Sue raised cashflow concern on Q3 call. Marcus aware. Director loan past Div 7A trigger.
Frameworks, Glossary and Key Concepts
A reference page for every named framework, methodology, and key concept introduced across the book.
Named Frameworks and IP
Capture → Structure → Intelligence → Scale
Introduced: Part 1, Chapter 1
The book's central framework. The four-stage progression that describes how an accounting firm builds institutional intelligence. Capture is the foundation. Without it nothing compounds. Structure makes the captured information queryable. Intelligence surfaces signals and patterns from the structured data. Scale is the outcome when all three are working.
→ See Diagram 04 — Capture, Structure, Intelligence, ScaleThe Four Levels of Structure
Introduced: Part 2, Chapter 5
A hierarchy for organising captured information. Raw capture at the bottom. Tagged and categorised above it. Linked and relational above that. Queryable and AI-accessible at the top. Each level makes the information more useful than the level below it.
The Multiplier Principle
Introduced: Part 2, Chapter 5
Structure multiplies the value of capture. The same information, structured at a higher level, produces exponentially more insight than the same information stored in a flat format. The investment in structure compounds over time.
The Personal Knowledge Trap
Introduced: Part 2, Chapter 5
The failure mode where institutional knowledge lives in one person's head rather than the firm's infrastructure. When that person leaves, the knowledge leaves with them. The intelligent firm moves knowledge from people's heads into the system deliberately and continuously.
The Two-Minute Close
Introduced: Part 2, Chapter 4
A meeting close technique named by Emma Baxter. At the end of every client meeting, the advisor states the actions clearly, for the client and for the transcript simultaneously. Produces clarity for the client about what happens next and a structured record for the system without any additional effort.
The Signal Stack
Introduced: Part 3, Chapter 8
A four-level hierarchy for categorising signals from the firm's data. Noise at the bottom — information that does not require attention. Soft signals above — patterns worth monitoring. Hard signals above that — movements that warrant a conversation. Critical signals at the top — situations requiring immediate human action. The system handles the lower levels. The human handles the upper ones.
→ See Diagram 06 — The Signal StackThe Deliverable Drift
Introduced: Part 3, Chapter 9
Named by Martin Kamenski. The gradual shift in how accounting firms perceive their value — from the relationship to the deliverable. When the deliverable becomes the product, the client feels processed. The relationship that produces the deliverable is where the real value lives.
The 25-Minute Meeting Model
Introduced: Part 3, referenced across Part 6
A structured advisory meeting format. Fifteen minutes of agenda driven by the pre-meeting brief and financial signals. Ten minutes of open conversation following wherever the client takes it. The structure creates space for the relationship.
The Middle Majority
Introduced: Part 7, Chapter 30
Named by Nick Ferguson. The largest group in any accounting firm's AI adoption curve. Cautious, competent, and persuadable. Neither early adopters nor active resisters. The group whose adoption determines whether a pilot becomes a firm-wide capability or a one-person experiment. Design the pilot for the Middle Majority.
The Deserve to Scale Test
Introduced: Part 6, Chapter 23 and Part 6 intro
Named by Amy Holdsworth. Four metrics that determine whether a firm is ready to grow. Core lodgement at ninety-five percent by the fifteenth of May. Capacity forecast at or below seventy percent of hours forward-booked. WIP lock-up under thirty days. Debtor days under fourteen, ideally zero. A firm that fails one of these is not ready to grow. It is ready to fix what it is failing on.
The Pricing Confidence Review
Introduced: Part 6, Chapter 25
A four-quadrant framework for preparing for any renewal or fee conversation. Quadrant 1: What was agreed. Quadrant 2: What was delivered. Quadrant 3: What changed. Quadrant 4: What is the client's situation now. Produces one of three outcomes: Maintain, Adjust, or Discuss. Turns the most uncomfortable conversation in accounting into one that starts from evidence rather than assertion.
→ See Diagram 19 — The Pricing Confidence ReviewThe Auto-Updating Loop
Introduced: Part 4, Chapter 16
The four-stage mechanism that keeps the Firm OS current without requiring a separate maintenance effort. Capture: work happens and the capture layer records it. Structure: the AI layer processes and tags the output. Flag: the system identifies what is potentially relevant to the Firm OS. Integrate: the AI Champion reviews and adds what is worth keeping. The loop runs as a byproduct of the work rather than as a separate activity.
→ See Diagram 10 — The Auto-Update LoopThe 90-Day Standard
Introduced: Part 7, Chapter 32
The training sequence for a new AI Champion. First thirty days: the approved tool stack and data policy, prompt engineering fundamentals, personal prompt library built on real tasks. Second thirty days: the second brain, the Firm OS, the signal infrastructure. Third thirty days: running the firm-wide usage audit, designing and running the first pilot, debriefing and communicating the outcome. By day ninety the Champion has completed one full adoption cycle.
The Human in the Loop
Introduced: Part 7, Chapter 30
A six-stage circular model showing how AI and human judgment work together continuously. Stage 1: Task Arrives. Stage 2: AI Processes. Stage 3: Human Reviews. Stage 4: Human Decides or Acts. Stage 5: AI Captures and Structures. Stage 6: AI Monitors and Surfaces. The loop is continuous. Stage 4 is the anchor — the moment that requires human accountability and cannot be delegated to the system.
→ See Diagram 23 — Human in the LoopThe 30-Day Pilot Structure
Introduced: Part 7, Chapter 31
Five stages for running an AI pilot that does not die in week three. Stage 0 (pre-week): Champion sets up, tests, and builds the prompt template. Stage 1 (week one): kickoff, participants try the tool once before leaving. Stage 2 (week two): Champion checks in, friction cleared. Stage 3 (weeks three and four): unassisted use. Stage 4 (end of week four): debrief and verdict. Verdict is one of three: expand, extend, or conclude.
→ See Diagram 25 — The 30-Day AI Pilot CalendarThe Numbers Mentor
Introduced: Part 6, Chapter 26
Named by Aynsley Damery. The advisory role that AI infrastructure makes scalable. An accountant who diagnoses before they prescribe, challenges as well as supports, and holds the business owner accountable to the plan. Distinct from a traditional accountant who reports and a business coach who advises without data. The intelligent firm delivers this role to every client who needs it, not just the ones whose partner is gifted at the conversation.
The CLEAR Methodology
Introduced: Part 6, Chapter 26
Developed by Aynsley Damery at Clarity HQ. A structured advisory conversation framework. Current: where the business is now across the seven key numbers. Leaderboard: benchmarks against the sector. Endgame: models the potential upside. Action: builds the specific plan. Review: monthly accountability that loops back to Current. The pre-meeting brief from the second brain is what makes CLEAR possible at scale.
The Systematic Engagement Model
Introduced: Part 6, Chapter 28
A four-layer client engagement architecture. Engagement Rhythm: a defined cadence for each client tier. Signal Monitoring: the intelligence layer that accelerates the rhythm when something warrants earlier contact. Advisor Brief: the pre-meeting summary the second brain produces before every planned or triggered contact. Record: every contact captured and fed back into the brief for the next contact. The loop runs continuously without depending on any individual's memory.
→ See Diagram 20 — The Systematic Engagement LoopThe Service Library
Introduced: Part 6, Chapter 24
A structured catalogue of every service the firm offers. Four components: the service definition (what is included and excluded), the pricing framework (how it is priced and under what conditions), the client eligibility criteria (which clients should receive it and when), and the delivery playbook (how it is delivered consistently). Connected to the capture layer, it converts client intelligence into structured service offers and surfaces scope exceptions in real time.
The Five AI Policies
Introduced: Part 4, Chapter 15
The five governance policies every accounting firm needs in the AI era. Data policy: what client data can go into which tools under what conditions. Tool approval policy: which tools are approved and how new ones are evaluated. Review policy: AI output does not go to clients without human review. Disclosure policy: whether and how the firm tells clients AI is used. Escalation policy: when AI output is flagged for human review and what happens next.
→ See Diagram 17 — The Five AI PoliciesThe Five-Layer Intelligent Firm Stack
Introduced: Part 5, Chapter 22
The technology architecture of the intelligent firm. Layer 1: Capture. Layer 2: Practice Management. Layer 3: Financial Intelligence. Layer 4: Intelligence (Claude, ChatGPT, Copilot). Layer 5: Workflow and Automation. Each layer depends on the ones below it. Skipping a layer does not remove the dependency.
→ See Diagram 18 — The Five-Layer Intelligent Firm StackKey Concepts and Terms
Second Brain
The firm's centralised, structured, queryable repository of institutional knowledge. Built from captured meetings, documents, SOPs, and client history. The foundation that every other layer of the intelligent firm depends on.
→ See Diagram 03 — The Second Brain ArchitectureFirm OS
The operating system the firm runs on. Documentation, knowledge base, policies, and the auto-updating loop that keeps all of it current. Distinct from the technology stack. The Firm OS is what the firm knows. The stack is the tools it uses to know it.
→ See Diagram 09 — The Firm OS LayersAI Champion
The person responsible for the firm's AI infrastructure and adoption. Not the most senior person and not the most technical. The person trusted by peers, capable without constant support, and curious enough to stay engaged. Four mandate dimensions: ownership of the approved stack, running pilots, answering questions, and maintaining the Firm OS connection.
→ See Diagram 26 — The AI Champion Role CardShadow AI
AI tool usage happening outside the firm's approved systems, without data policies applying, and without visibility into what client data is going into which tools. Present in almost every accounting firm that has not actively addressed it. The response is not surveillance. It is a better approved alternative.
MCP (Model Context Protocol)
A universal connection standard that allows AI-enabled tools to connect to each other without a custom integration for each pair. The standard that is making the ecosystem approach to technology viable for smaller firms.
→ See Diagram 05 — The MCP Connection MapThe Drift
Named by Aynsley Damery. The pattern by which small businesses fail slowly and invisibly rather than dramatically. A point here, a point and a half there in Gross Profit Percentage. Cash Days stretching. Revenue per Employee flatlined. Nobody having the conversation. The Numbers Mentor role exists to catch the drift before it becomes irreversible.
The Revenue Trap
Named by Aynsley Damery. Chasing revenue to fix a profitability problem is like turning on the taps to fix a leaking pipe. More revenue amplifies existing problems if the underlying model has not been fixed.
The Enriched Touchpoint
Any client communication that carries genuine, specific, contextual value. Distinguished from generic newsletters or bulk regulatory updates by specificity. Requires both signal intelligence (what is happening in this client's situation) and relationship context (what the firm knows about this person). Infrastructure makes enriched touchpoints systematic rather than exceptional.
The Human-Keep Categories
Five categories of work that stays with humans regardless of AI capability. Judgment under uncertainty. Difficult conversations. Relationship moments. Novel situations. Professional accountability. Not defensive positions. The definition of the profession's value.
→ See Diagram 24 — The AI Task Inventory & Human-Keep CardInstitutional Intelligence
The accumulated knowledge that the firm builds over time through capturing, structuring, and learning from its work. The thing that a firm starting today cannot buy, import, or replicate with better tools. Compounds like interest. More valuable in year three than year one.
The Compounding Adoption Curve
The pattern by which one successful pilot gives the firm permission to run the next. Each subsequent pilot runs faster because the team has seen the previous one work. Within six months a firm that started with one use case has a portfolio of embedded workflows.
Pre-Meeting Brief
The structured document the second brain produces before every client advisory meeting. Five components: relationship summary, outstanding commitments, financial signals, opportunity flags, and suggested agenda. Removes the preparation bottleneck and gives every advisor what the best partner carries in their head.
The Usage Audit
A people audit rather than a technology audit. Four questions asked of every team member: What AI tools are you currently using? What for? What client data goes into them? What would you need from an approved alternative to stop using the unapproved ones? The most useful intelligence-gathering exercise a firm can run before making any adoption decisions.
Diagrams across the book
Every figure that appears in the chapters, gathered here so they can be revisited without finding the page they sit on. Click any diagram to open it full size.
Diagram 01Part 1 · The Line in the SandChapter 1
The Three Eras of the Firm.
From the analog firm to the digital firm to the intelligent firm. Three waves of technology, each one not just faster than the last but changing what a practice could become. The shift now underway is the first that turns captured knowledge into compounding capability.
198019952006201520222025You are here
I
AnalogThe Analog Firm
1980 — 2005
Paper to disk.
Quicken, CCH, Thomson Reuters. Work that lived in paper ledgers moved onto hard drives. Email replaced the fax. The GST rollout forced an entire profession to rethink compliance overnight. The shape of the firm did not change — only the pace.
Wave · Desktop & Internet
II
DigitalThe Digital Firm
2006 — 2024
Disk to cloud.
Xero launched in 2006. QuickBooks Online followed. Sage rebuilt. The ledger moved off the desktop and into the browser. Real-time visibility became possible. Collaboration between advisor and client stopped requiring a physical handoff. Data became accessible — but not yet intelligent.
Wave · Cloud Accounting
III
IntelligentThe Intelligent Firm
2025 →
Data to intelligence.
A small firm operating with the institutional knowledge of a large one. A partner who knows what happened in every meeting without taking a single note. The intelligence layer turns captured conversation into compounding capability — and the foundation underneath it is the work of this book.
Wave · AI · Capture · MCP
Diagram 02Part 1 · The Line in the SandChapter 1
The Cost of Waiting.
Two firms start in the same place. One builds the foundation; the other waits for the dust to settle. The gap between them does not stand still. It compounds — quietly at first, then all at once.
Capability · Compounded Advantage
2025 — todayYear 1Year 2Year 3Year 4Year 5
Diagram 03Part 2 · The Second BrainChapters 3–6
The Second Brain Architecture.
Every firm has intelligence. It already exists — in meeting notes, partner heads, the EA's inbox, the call on the way out the door. The Second Brain is the structure that lets that intelligence survive the person who held it, so a firm can stop running on memory and start running on knowledge.
Intelligence layerClaudeOpenAIMCP
L 04IV
Intelligence
Ask any question, get a real answer.
Plain-language queries across every conversation, commitment and number the firm has ever held. The new advisor walks in with the institutional memory of a 20-year partner. The senior partner walks out without it leaving with them.
L 03III
Knowledge
Queryable, searchable, alive.
Captured records become a connected knowledge base. Client history, firm SOPs, regulatory context, the lessons learned across every engagement — all sitting where any team member can reach them.
L 02II
Structure
Raw capture, made meaningful.
Tags, relations, entities. A meeting becomes a record linked to a client, an engagement, a service line, and a commitment. The multiplier that turns data into something a machine — and a human — can reason about.
L 01I
Capture
Everything the firm says and hears.
Speech-to-text as foundational infrastructure. Every conversation, every commitment, every piece of context that currently lives in someone's head — written down before it evaporates. Nothing else works without this.
The argumentAI without a foundation produces confident, generic answers. A foundation without AI is a filing cabinet. The Second Brain is the layer where they meet.
The promiseA small firm operates with the institutional knowledge of a large one. A team's collective memory outlives any one person on it.
The orderCapture is not optional and not last. It is first. Structure follows. Knowledge follows that. Intelligence is what the previous three earn you.
The compoundingEvery engagement makes the next one cheaper, faster, sharper. The advantage is not the tool. It is the data the tool is allowed to see.
Diagram 04Part 2 · The Second BrainChapters 4–6
Capture, Structure, Intelligence, Scale.
The through-line of the entire book in four words. Each stage earns the next. Skip one and the stage that follows produces confident answers that are wrong, or generic answers that could apply to anyone. Built in order, the four stages turn a firm's daily noise into a compounding advantage.
Stage 01 · Foundation
CaptureEverything the firm says and hears.
Speech-to-text as infrastructure. Every meeting, every call, every voice memo, every commitment — written down before it evaporates. Nothing else in this book works without this.
YieldRaw, complete record.
Stage 02 · Multiplier
StructureRaw capture, made meaningful.
Tags, entities, relations. A meeting becomes a record linked to a client, an engagement, a service line, a commitment. The step that turns data into something a machine — and a human — can reason about.
YieldConnected, contextual data.
Stage 03 · Activation
IntelligenceAsk any question, get a real answer.
Plain-language queries across every conversation, commitment, and number. The new advisor walks in with the memory of a twenty-year partner. The senior partner walks out without taking it with them.
YieldQueryable, on-demand knowledge.
Stage 04 · Compounding
ScaleA firm that compounds without you.
Every engagement makes the next one cheaper, faster, sharper. New hires onboard against the firm's full memory. Senior people leave without taking the firm with them. The advantage widens with use.
YieldAn intelligent firm.
The order is the argument
Capture, then Structure, then Intelligence, then Scale.
You can't fix a structure problem with a bigger model. You can't fix a capture problem with better tagging. The order matters. Skip a stage and what comes next breaks.What each stage costs to skipNo capture → AI is generic. No structure → answers are confident but wrong. No intelligence → a tidy filing cabinet nobody asks. No scale → you built it for one person, not a firm.
What good looks likeThe bottom three stages are infrastructure decisions, made once, maintained quietly. The fourth — Scale — is the only one a partner ever needs to talk about. The rest just works.
Diagram 05Part 2 · The Second BrainChapters 5–6
The MCP Connection Map.
A firm's tools have always talked to each other badly, or not at all. MCP — the Model Context Protocol — is the standard that lets any AI talk to any tool in plain language. The same plug, on every side. Build once. Connect forever.
The connective tissue
One protocol, every tool.
MCP
Model Context Protocol
Open standard · 2024
Any tool that speaks MCP can be reached by any AI that speaks MCP — without a custom integration for either side.
The analogy
USB is to devices what MCP is to AI.
Before USB, every printer, scanner and camera needed its own port. After USB, one plug worked for all of them. MCP does the same thing for the conversation between AI models and the software your firm already owns.Why it mattersA practice management vendor that speaks MCP can be reached by every AI model that exists today and every model that ships tomorrow — without rebuilding anything. Lock-in goes down. Optionality goes up.
The question every firm should askWhen the rep from your practice-management, document or workflow vendor next walks into your office, the buying criterion is no longer “does it have AI features?” It's “is your platform MCP-compatible — and if not, what's the date you ship it?”
Diagram 06Part 3 · SignalsChapter 9
The Signal Stack.
Every client, team member, and number in the firm's GL is communicating something. Most of it is noise. A little of it is the future, arriving early. The Signal Stack is how a firm decides — deliberately — what gets filtered, what gets watched, and what gets acted on.
Level 04 · Critical
Act now.
Volume
Time-sensitive. Material risk. Already in motion.
Cashflow breach within 7 daysClient signalled exit in last callCompliance deadline missedSenior team member resignation risk
Action lane
Partner alerted within the hour.
A human decides. The system surfaces, escalates, and stops talking.
AI surfacesHuman decides
Level 03 · Hard signal
Look this week.
Volume
Concrete pattern. Repeated, measurable, named.
Three missed client responsesWIP creeping on a fixed-fee jobMargin compression two quarters runningEngagement score down 20%
Action lane
Owner reviews on Monday.
AI surfaces with context attached. Human assigns next step.
AI surfacesHuman reviews
Level 02 · Soft signal
Watch it.
Volume
Suggestive, not certain. Worth aggregating.
Tone change in last callSlight delay in paymentNew competitor mentionedQuiet client, no replies in three weeks
Action lane
AI logs & watches.
Aggregated weekly. Promoted to Hard if the pattern repeats.
AI logsAI patterns
Level 01 · Noise
Ignore.
Volume
Routine traffic. Doesn't change anything.
Standard meeting confirmationsRecurring lodgement remindersOut-of-office repliesBackground system notifications
Action lane
Filtered automatically.
Captured for the record. Never escalated. Nobody's inbox sees it.
AI filters
The argument
Most firms drown in noise and miss the signal.
The intelligent firm sets thresholds deliberately. The stack is not a feature of the AI — it is a policy decision the firm makes once and tunes over time.The volume ruleNoise is the loudest tier by far and Critical is the rarest. If your system is flagging Critical events all day, the threshold is set wrong. Tune it. The cost of a false alarm is fatigue. The cost of a missed signal is a lost client.
Who does whatAI sees everything. AI filters most of it. AI surfaces what humans should see. Humans decide on anything Critical. The promotion path — Soft → Hard → Critical — is the rule that lets the system get smarter without anyone watching it.
Diagram 07Part 3 · SignalsChapter 7
The Three Signal Types.
Every signal in the firm is one of three things. Something a client is saying. Something the team is feeling. Something the numbers are doing. The intelligent firm sees them all — and pays special attention to where they overlap, because that is where the future of the firm is usually written.
The three typesAnd the place where they meet
Type 01 · Client
What the client is telling you.
Engagement drop-off, tone shifts, the quiet client who used to reply on the day. The signals that arrive long before a retention conversation does.
Tone changeReply latencyScope driftQuiet client
Type 02 · Team
What the team is feeling.
Workload-to-output relationship. Wellbeing signals. Quiet high performers and the resignation that nobody saw coming six months out.
WIP loadBurnout cuesQuietly excellentRetention drift
Type 03 · Financial
What the numbers are doing.
The GL is always talking. Margin compression, cashflow stress, debtor behaviour. Engagement profitability that nobody's reading until quarter-end.
Margin compressionCashflow stressDebtor patternWIP creep
The overlap · All three
The whole-firm event.
A long-time client renegotiates terms. The senior who carried them is already stretched. Margin on the engagement was already thin. Three signals, one moment. The firms that win catch it as it arrives, not three months after.
Renegotiation riskCarrier burnoutMargin in retreat
The argument
Most firms see one type of signal at a time.
The partner watches the client. The practice manager watches the team. The CFO watches the numbers. None of them watches the place where the three meet — which is where most of the firm's risk and most of its opportunity actually lives.What the intelligent firm doesCaptures all three streams continuously. Lets the second brain join them by client, engagement, and time. Surfaces the joins to the right human at the right moment, instead of waiting for the quarterly review to discover them.
Where to startPick one client. Look at every conversation, every team note, every GL movement on their engagement over the last six months. The intelligent firm reads that file in seconds. The average firm has never read it at all.
Diagram 08Part 3 · SignalsChapter 11
The Financial Signal Workflow.
The general ledger is always talking. The intelligent firm doesn't just hear it — it captures it, joins it to context, classifies it, routes it, acts on it, and writes back what it learned. One signal, end to end, with the human in the seat where they belong.
Worked example“Client X · monthly revenue down 38% MoM.”Source · Xero · 06:14 AM Tuesday
Stage 01
Detect.
A movement in the GL.
The capture layer is watching the ledger continuously. A threshold is crossed. A new event lands — raw, untagged, alone.
Raw eventMonthly revenue threshold breached on Client X.
AI watches
Stage 02
Enrich.
Context, attached.
The event is joined to everything the firm already knows about this client — recent meetings, commitments, engagement owner, last six months of behaviour.
With contextWas $130K, now $80K. Last call 3 weeks ago discussed scaling. Owner: Sarah. WIP unchanged.
AI joins
Stage 03
Classify.
Noise, Soft, Hard, or Critical.
The signal stack assigns a level. Magnitude, recency, and pattern fit are weighted against thresholds the firm set deliberately.
Level assignedHard signal — material magnitude, concrete pattern, recent context.
AI grades
Stage 04
Route.
To the right human at the right time.
Hard signal → engagement owner's Monday review queue, with the full enriched record attached. No inbox. No noise. No partner pinged needlessly.
Where it landsSarah's queue, Monday morning. Pre-attached context. One click to open the client.
AI routesHuman receives
Stage 05
Act & learn.
A decision is made — and remembered.
Sarah calls the client. They paused a project; expanding next quarter. She logs the outcome. The system updates its pattern — next time, this shape becomes Soft, not Hard.
Outcome“Temporary pause, growth deferred to Q3.” Pattern updated. Watchlist promoted to expansion conversation.
Human decidesAI learns
The loopEvery decision the human makes is captured back into the second brain. The next signal of the same shape is classified more accurately. The system gets better the more the firm uses it — without anyone maintaining it.
What's different
The signal arrives with everything you need to act on it.
In the average firm, the same GL movement surfaces in the quarterly review — three months late, stripped of context, attached to no one. The intelligent firm catches it Tuesday morning, with the whole story.Where the humans sitAI watches, joins, grades, and routes. Humans decide. Nothing irreversible happens without a person in the loop. The system's job is to put the right human in front of the right decision at the right time.
The compoundingEvery action is a label. Every label refines the classifier. After six months of use, the firm's signal stack is tuned to its own clients, its own engagements, its own risk tolerance — and is impossible to replicate from outside.
Diagram 09Part 4 · The Firm OSChapters 12–16
The Firm OS Layers.
Every firm has an operating system. Most firms have not designed it — it accumulated, under pressure, from habits nobody decided on. The designed OS makes explicit what the inherited one leaves implicit. Four layers, kernel to surface, with one loop that keeps it current.
The four layersFrom the kernel outward
Layer 01 · Kernel
Identity
What the firm stands for, holds true, and will not compromise.
Values, standards of work, the line the firm won't cross with AI or with clients. Sets the constraints every layer above it must respect.
Layer 02 · System
Policies
What the firm permits, prohibits, and reviews — including AI usage.
Living policy frameworks tested against real situations. Data security, confidentiality, AI permissions, escalation rules. Decisions made once so they don't need to be re-made every day.
Layer 03 · Library
Knowledge
What the firm has learned, written down where any team member can reach it.
Client history, technical and regulatory knowledge, lessons learned across every engagement. The retiring partner's memory, preserved. The senior advisor's mental model, shared.
Layer 04 · Surface
Procedures
How the work actually gets done, day to day, on every engagement.
Living documentation — workflows, checklists, SOPs — that updates as the firm learns. The onboarding call becomes the SOP. The conversation becomes the documentation. Nobody writes it after the fact.
The inherited OS
Reasonable when normal. Unreliable when not.
The OS most firms run on accumulated by accident. It works as long as the right people are in the room. It fails in onboarding, handover, and consistency — the three places that decide whether a firm scales.The designed OSEvery process has an owner. Every handover has a structure. Every edge case resolved once is resolved forever. The OS is not a project that competes with the work — it is the infrastructure that makes the work sustainable.
The loop is the pointA static OS rots. The auto-updating loop (Ch 16) wraps every layer — the onboarding call becomes the SOP, the resolved edge case becomes the policy, the new lesson becomes part of the library. The system maintains itself.
Diagram 10
Part 4 · The Firm OS
Chapter 16
The Auto-Update Loop.
A static OS rots. The intelligent firm runs a loop — every engagement produces a small update to the system that ran it. The onboarding call becomes the SOP. The resolved edge case becomes the policy. The new lesson becomes part of the knowledge base. Nobody is documenting after the fact, because the work is the documentation.
One turn of the loopWork in. Smarter firm out.
01
Step 01 · Origin
Work.
An engagement happens.
A client meeting, an advisory call, an unusual situation. Real work — not a documentation exercise. The loop only ever starts here.
Firm02
Step 02 · Record
Capture.
Recorded, structured, joined to context.
The second brain ingests the engagement. Transcript, participants, decisions, commitments — attached to client, engagement, and service line automatically.
AI captures03
Step 03 · Detect
Flag.
Something the OS doesn't yet know.
A missing SOP. A novel edge case. A repeated question that has no canonical answer. The AI doesn't decide what matters — it surfaces candidates for review.
AI flags04
Step 04 · Decide
Review.
A human decides what becomes OS.
The OS owner sees a flagged item with full context attached. One click: keep as a new SOP, edit and accept, or discard. Five minutes a day, not a quarterly project.
Human decides05
Step 05 · Compound
Update.
The OS revises itself.
The relevant layer — Procedures, Knowledge, Policies — gets the change. Versioned, attributed, dated. The next engagement that needs it gets the new version automatically. Then the loop starts again.
AI writes back
The argument
The work is the documentation.
The onboarding call becomes the SOP. The resolved edge case becomes the policy. Nothing gets written after the fact, because the loop closed the moment the engagement ended.
Why most firms can't do this
Without capture, there is no record. Without structure, the record can't be flagged. Without policies, no human knows what to keep. Without owners, nothing gets written back. The loop only runs once Parts 2, 3 and 4 are in place — that's the order this book is in.
What changes after six months
The OS is no longer a folder the firm maintains. It's a living artefact the firm runs on. New hires onboard against the latest version. Senior people leave without taking the firm with them. The system gets better the more it's used — and is impossible to replicate from outside.
Diagram 11
Part 1 · The Line in the Sand
Chapter 2
AI Readiness is a spectrum.
Two firms can run the same model with the same prompt and get answers that look like they came from different products. The tools are the same. The data underneath them is what changes everything. Where your firm sits on this spectrum decides whether AI gives you confident-but-wrong or specific-and-actionable.
Input · One prompt, sent at both firmsIdentical
"What's happening with Acme Pty Ltd this quarter — and what should we do about it?"
Same model· Same prompt · Same query budget
Left end of the spectrum
The foundation-poor firm.
Same AI. Different answer. Same confidence.
Output · Generic— Source: pattern-matched
"Acme Pty Ltd appears to be a manufacturing client. Their account is typically managed at a quarterly cadence. Most clients in this stage of relationship benefit from a regular check-in to discuss any recent changes, review compliance obligations, and identify opportunities for cross-sell. We recommend scheduling a meeting."
- No reference to recent meetings, decisions or commitments.
- No financial signal — margin, cashflow, debtors invisible.
- Recommendation could apply to literally any client in the book.
Generic
Unsourced
Hallucinated
Plausible · Wrong
What the model can see · Underneath
Scattered fragments. No index.
Files live in folders, inboxes, heads — not records.
No tags. No links. No who-said-what.
Knowledge walks out with each resignation.
The model has nothing specific to ground on.
Right end of the spectrum
The foundation-rich firm.
Same AI. Different answer. Decision-grade.
Output · Specific— Source: capture layer · GL · second brain
Acme Pty Ltd · Q3 FY26. 3 meetings this quarter (04 Jun · 19 Jul · 22 Aug). Gross margin −4.2pp vs Q2 budget; debtor days drifted from 38 → 52. On 22 Aug, Director Sue flagged a supplier dispute affecting two largest customers. Open advisory commitment from 19 Jul: cashflow scenario by 30 Sep, owner Marcus. Recommendation: bring the cashflow forward to this week and pre-empt the call Sue will otherwise make on Friday.
- Every claim is tied to a specific event, date, or GL line.
- The commitment is in the second brain, not in someone's head.
- The recommendation is a decision — not a check-in.
Specific
Sourced
Dated
Actionable
What the model can see · Underneath
A connected, queryable record.
Every conversation joins to client, engagement, service.
Commitments and decisions are first-class records.
Financial signals live next to relationship signals.
The model retrieves; it doesn't invent.
01None
02Sparse
03Connected
04Structured
05Queryable
The argument
The tools are the same. The data is what changes.
AI doesn't degrade gracefully on missing data — it confabulates. Most firms experimenting with AI are sampling the left end of this spectrum and concluding the technology is unimpressive. They're sampling their own foundation.
Why this matters now
The gap between the two ends compounds. A foundation-rich firm trains its second brain a little more every engagement; a foundation-poor firm gets the same generic answer next year that it got last year. The same model — better data — wins.
Where to start
Don't buy more AI. Build the substrate first.
Capture the meeting. Structure the record. Join the GL. Tag the people. Then ask the model what's happening. The order is not a preference — it's how the loop closes at all.
Diagram 12
Part 3 · Signals
Chapter 11
The four financial signal categories.
The GL is always talking. Most firms only listen for what's about to break. The intelligent firm hears the same numbers and pulls out four very different conversations — two about protecting the client, two about helping them grow. Financial signals are not just about risk. They are about relationship.
Retrospective
Forward-looking
Risk · Protect
Opportunity · Grow
The four categories
Numbers + people
Signal · not summaryLiquidity
01 · Cashflow
Cashflow.
When the money is coming in, going out, and whether the gap is widening.
Signal · what the GL shows
Debtor days drift 38 → 52 over six weeks. Two large clients now 60+ days overdue. Working capital headroom down 22% vs. opening of quarter.
The conversation it opens
"Your top three debtors slipped past sixty days. Want us to draft the chase script and run a cashflow scenario before Thursday's board?"
Margin
02 · Profitability
Profitability.
Where margin is leaking, which jobs are losing money, and which clients are quietly subsidising others.
Signal · what the GL shows
Gross margin on Service Line B down −4.2pp vs Q2 budget. Realised hours on three engagements 1.6× the quote. WIP write-off ratio creeping 7% → 11%.
The conversation it opens
"This service is profitable on paper and unprofitable in practice. Let's renegotiate scope before renewal — or move it to a fixed-fee envelope."
Obligations
03 · Compliance
Compliance.
Filings, lodgements, statutory obligations — what's due, what's at risk, what's already been quietly missed.
Signal · what the GL shows
BAS for 3 entities drifting toward the 28-day window. PAYG instalment estimate 16% below actual trading. Director loan balance moved past Div 7A trigger.
The conversation it opens
"Three things will become problems in the next forty-five days. Here's the order to fix them — and the one that needs your decision before next week."
Expansion
04 · Growth
Growth.
Where the business is leaning forward — new revenue mix, new headcount, capital decisions, hidden capacity.
Signal · what the GL shows
Revenue from Segment C up +38% YoY and now 22% of book. Two new hires in payroll without an org-chart update. Inventory turnover lifting 5.1 → 6.4.
The conversation it opens
"Segment C is becoming the business. Want us to model what pricing, working capital and structure look like if it keeps growing at this rate?"
The argument
Not just risk. Relationship.
Two of these categories protect the client; two help the client grow. The same GL surfaces both. A firm that only listens for the risk signals is leaving the relationship half-built.
How it works in practice
The second brain joins the GL to the client record, the engagement, and the meeting capture. Each category has its own cadence — daily, weekly, monthly, quarterly — and its own escalation rule into the Signal Stack.
What it changes
The advisor arrives with the question already asked.
The client stops being surprised by their own numbers. Every check-in starts with something specific and recent — not a generic "how's business?". That is what makes the relationship worth paying for.
Diagram 13
Part 4 · The Firm OS
Chapter 13
The Documentation Lifecycle.
Most firms have documentation. Few have documentation that's true. The same SOP can be inherited, audited, or living — and the difference shows up the moment someone actually tries to use it. The point of the Firm OS isn't to write more documents. It's to make sure the documents reflect what the firm actually does.
01 · Inherited
Stale
Inherited.
Frozen at the moment someone last cared.
OriginCreated at a point in time, usually by someone who has since left.
LifespanAccurate the day it's written. Quietly decays from then on.
What it tells youWhat the firm used to do — at a date nobody can find on the document.
Accuracy over time−68% in 5 years
02 · Audited
False confidence
Audited.
A date stamp pretending to be a heartbeat.
OriginPeriodically reviewed by a committee. Re-stamped, rarely re-thought.
LifespanNever quite finished. The next quarter's audit always justifies it.
What it tells youThat someone signed off — not that the firm still does this.
Accuracy — Claimed vs Actualgap widens each cycle
03 · Living
Live
Living.
Connected to the work. Updated by the work.
OriginBorn from the capture layer — meetings, comms, decisions feed it in real time.
LifespanAlways current. Versioned, attributed, dated — every change has a source.
What it tells youWhat the firm actually does — right now, with the receipts to prove it.
Accuracy over timestays at the line
The argument
Documentation that doesn't update isn't documentation. It's archaeology.
The first two columns are what most firms have today. The third is what the Firm OS is for. Without the capture layer underneath, every SOP eventually becomes column one — no matter how often it's audited.
How living works in practice
A meeting captures a decision. The decision is structured into an event. The event is owned by a procedure. The procedure rewrites itself, with the meeting and the decision linked as evidence. A human reviews — five minutes — and the change is in the next version. No-one wrote a doc. The doc wrote itself.
What it changes for the team
Onboarding works. Audits stop lying. New hires can trust the SOP.
The procedure folder is no longer the place where last year's truth goes to die. It's the artefact the firm actually runs on — and the AI assistant can finally retrieve from it without making things up.
Diagram 14
Part 4 · The Firm OS
Chapter 13
Ask the firm a question.
A natural-language interface on top of the second brain. Plain question on the left, sourced answer on the right — every claim tied back to an SOP, a client record, or a technical note. The point isn't the chat. The point is that the firm finally has something worth asking.
Context: Acme Pty Ltd
Engagement: Onboarding
You: Sue Henderson · Partner
S.H
You· 09:42 today
How do we onboard Acme Pty Ltd from their previous accountant — and is there anything specific to their situation I should handle first?
AI
Firm Brain· retrieved from 12 sources · 1.8s
Three steps from the SOP, then one thing specific to Acme.
Standard onboarding from another firm
- Issue an ethical-clearance request within 48 hours of engagement-letter signing. SOP-04 §2.1
- Notify the previous accountant in writing and request handover by Day 7. SOP-04 §2.4
- Reconcile opening balances against the prior return before the first lodgement. Tech Note T-12
Specific to Acme
On the handover call on 19 Aug, the prior accountant flagged a Div 7A loan balance moving past the trigger threshold. Marcus is aware; Sue is not yet. Address this before Q3 lodgement, not after. Client · Acme · Meeting 19 Aug
Suggested next step
Schedule a 30-min Acme onboarding meeting with Marcus before Friday. I can draft the agenda and the ethical-clearance letter — say the word.
SOP-04 · Client Onboarding
"Within 48 hours of engagement-letter signing, the engagement partner must issue an ethical-clearance request to the prior accountant…"
Tech Note T-12
"Opening balances are reconciled against the prior return using the three-statement walk-back method described in §3…"
Meeting · Acme handover
"Prior accountant flagged director loan balance moved past Div 7A trigger. Marcus aware. Action: address before Q3 lodgement."
Ask the firm anything · scoped to Acme Pty Ltd…
Ask →
The argument
A query interface on a firm that's worth querying.
The chat is the easy part. The hard part is everything you can't see — the captured meetings, the structured records, the SOPs that actually reflect the work. Without that, this same UI returns generic nonsense.
Who this is for
Every partner, every advisor, every new hire.
The partner uses it to prepare. The new hire uses it instead of finding "the person who knows." The firm stops being a network of memories and starts being a system.
What it changes
Senior knowledge stops walking out the door.
The senior partner can retire and the firm still answers questions the way the senior partner would have answered them — because the answer was always in the SOP, the client record, and the meeting note. The interface just made it reachable.
Diagram 15
Part 4 · The Firm OS
Chapter 13
The AI Email Agent Flow.
Every inbound email runs the same loop. The agent reads it. It checks the knowledge base and the second brain. A decision rule fires. Most emails come back as a sourced draft — sent or queued. The ones that shouldn't be auto-handled get a human, with all the context the human needs to answer in two minutes instead of twenty. The Signal Stack, applied to the inbox.
01
Email arrives
Inbox.
A real email from a real client.
MW
Marcus Williams
team@theintelligentfirm.com · to team@theintelligentfirm.com
Quick question about quarterly BAS — overdue?
"Hi team, I just got a reminder from Karbon about our Q3 BAS — is it overdue? We thought it was lodged last week. Can someone confirm what's happening? Thanks, Marcus"
02
Agent retrieves
Read · join · scan.
The agent goes looking for what it needs.
Retrieval · 1.6 sec
12 sources checked · 4 hits
Client
Acme Pty Ltd · Tier 1 · Advisory
✓ hit
Procedure
SOP-09 · Client comms · BAS
✓ hit
Lodgement
Q3 FY26 BAS · filed 14 Aug · ref XYZ
✓ hit
Meeting
Acme handover · 19 Aug
✓ hit
Policy
Auto-reply scope · Tier 1 OK
— check
Tone
Sender language · neutral
— check
The answer exists. The Q3 BAS was lodged on 14 August. The Karbon reminder is for Q4 prep, not Q3.
03
Decision rule
Confidence × risk.
Can it answer? Should it?
Confidence
96%
Relationship risk
Low
Auto-reply checks · all must pass
✓
Answer fully grounded in retrieved sources
pass
✓
Tone neutral · no sensitive topic flagged
pass
✓
Client tier permits auto-reply
pass
✓
No financial commitment / scope change
pass
!
Partner audits sample · 1 in 10
sampled
Auto-respond — draft & send.
→
04
Output · two paths
The reply, or the brief.
The agent picks one. The other shows what would have happened.
Auto-reply · drafted & sent
12 sec ago
Re: Quick question about quarterly BAS
Hi Marcus — quick answer: your Q3 BAS was lodged on 14 August (ATO ref XYZ-1184). The Karbon reminder is for Q4 prep, which we'll start the week of 2 Sep. I'll flag it in your next status note. — Hello, the firm
◆ Lodgement · Q3 BAS
◆ SOP-09 §3
◆ Engagement · Acme
Logged · Acme · BAS
Sampled by Sue
04 ⤬
If a check failed → human escalation
Escalated · advisor brief
would queue · 0 sec
Routed to SUE HENDERSON · ACME PARTNER
"Marcus is asking about Q3 BAS status — neutral tone, but our records show a recent Div 7A loan flag on the handover call. Worth a 2-min phone call, not an email."
Context bundle attached
◆Client record · Acme Pty Ltd
◆Recent meeting · Acme handover · 19 Aug
◆Draft reply · ready to send / edit
◆Suggested talking points · 3
Status · awaiting human
Slack DM · Sue
The argument
A draft is not a reply. A check is not a delay.
The agent never sends what it isn't sure of, and never escalates what doesn't need a human. The volume of "where are we up to?" emails — which is most of inbound — drops to a one-line audit on the partner's phone.
How it mirrors the Signal Stack
Auto / Auto-with-audit / Human / Critical.
Same four-tier logic Chapter 8 applied to financial signals. Level 1–2 the agent handles. Level 3 the agent prepares. Level 4 the partner gets a phone-call alert before the email is even opened.
What it changes
The inbox stops being the firm's bottleneck.
Partners stop being the routing layer. New hires stop ghosting on emails they don't know how to answer. The client gets a response in minutes — sourced and correct — and nobody worked late to send it.
Diagram 16
Part 4 · The Firm OS
Chapter 14
The knowledge base is three things, connected.
Most firms call any single body of writing "the knowledge base." The intelligent firm separates client knowledge, technical knowledge, and operational knowledge — and then makes a queryable layer sit across all three. The answer to most real questions lives in the join, not in any single silo.
The queryable layer
One question, three sources.
A natural-language layer sitting across the three silos. The model retrieves; it doesn't invent.
Sue at Acme wants to recapitalise via a Div 7A loan rollover — what's our position and how should we frame the conversation?
Position is yes, with conditions.
On Sue's 19 Aug call she signalled cashflow concern as the reason CLIENT · meeting 19 Aug. Under Div 7A a 7-yr unsecured rollover qualifies if benchmark interest applies TECH · TR 2023/4. We handled the same pattern for Calvert & Co in March — drafted board minute, ATO acknowledged in 9 days, the template's in the OS OPS · Calvert · lessons. Lead with the cashflow risk; the rollover is the answer, not the topic.
Retrieved from
Client knowledge3 hits
Technical knowledge2 hits
Operational knowledge2 hits
Synthesis time
1.4 s12 sources scanned
01 · Client
312 entries · active
Client knowledge.
Who they are. What they want. What we've already told them.
Relationship history
Personal context
Decisions made
Advice acted on
Advised against equipment hire-purchase. They went ahead. Margin -2.4pp confirms the call.
Annual planning: keep family trust distribution flat at $180k. Reviewed Aug-26.
Sue (decision-maker), Marcus (founder, sentiment), Karen (CFO, exec). Family also clients.
Owner · Sue Henderson
Live
02 · Technical
488 entries · active
Technical knowledge.
What the firm knows about the work. Tax positions, standards, the field.
Tax positions
Accounting std
Approaches
Regulatory research
7-yr unsecured rollover OK if benchmark interest applied; secured 25-yr if real prop. Cite TR §3.1
EV exemption thresholds for non-luxury cars. Firm position: apply blanket to all motor.
Bendel update reshapes UPE treatment from 1 Jul. Firm position memo + client letter template.
Revenue / assets / employee tests for size class. Decision tree v3 · live in tool.
Owner · Tax tech committee
Live
03 · Operational
196 entries · active
Operational knowledge.
What we've learned doing the work. Edge cases. Mistakes. Process variations.
Edge cases
Lessons learned
Process variations
Mistakes
Same Div 7A rollover pattern. Drafted board minute + ATO ack in 9 days. Template T-44
Missed UPE flag on Tetra. Cost the client $14k. New checklist step added · §3.6.
SMSF in-house asset rule with related-party lease. Resolved with private ruling — see PR-2025-08.
Onboarding from competitor: 48-hr ethical clearance is the speed limit, not the formality.
Owner · Each partner contributes
Live
The argument
A folder of SOPs is not a knowledge base.
Most firms mash three different kinds of intelligence into one drive and wonder why nobody can find anything. Separating the silos — and then connecting them — is what turns a folder into an asset.
Why the connection matters more than the silos
The answer is almost always in the join.
Knowing Div 7A doesn't help unless you also know Sue's mood. Knowing Sue's mood doesn't help unless you also remember how Calvert went. The queryable layer is the bit that does the join — and it's where AI earns its keep.
What it changes
Senior knowledge stops being a person. It becomes a place.
The senior partner can leave. The new graduate gets the same answer the senior would have given — sourced, with the precedent, with the conversation cue. The firm's intelligence is finally separable from the firm's people.
Diagram 17
Part 4 · The Firm OS
Chapter 15
Five AI policies — and what makes them specific.
The policy framework is not complicated. Five policies cover almost every situation a firm will run into. What most firms are missing isn't the topic — it's the specificity. A policy that says "use judgment" is not a policy. It's a wish.
The Policy
Vague
what most firms have
Specific
what good looks like
01 · Data
Data policy.
What information may go into which AI tool, and in what form.
Owner · AI Committee
Vague — non-policy
"Don't put confidential client data into AI tools."
No definitions of "confidential," "AI tools," or "into." Every employee resolves it differently. In practice — they don't.
Specific — operable
Client documents may be uploaded to Claude (Enterprise), Microsoft Copilot (M365), and ChatGPT Team. Never to free or consumer tiers. TFN · DL · Passport redacted before any prompt.
02 · Tool approval
Tool approval.
Which AI tools the firm permits — and what gets a person before it gets used.
Owner · AI Champion
Vague — non-policy
"Use AI tools sensibly. Talk to a partner if unsure."
Sensible to whom? Which partner? Shadow AI grows in this gap — staff use what's free and don't say.
Specific — operable
Approved list: Claude Enterprise · Microsoft Copilot · Vinyl AI Notetaker · Karbon AI · FYI Copilot. Anything else needs AI Champion review + partner sign-off before client work. List refreshed quarterly.
03 · Review
Review.
What level of human review applies before AI output leaves the firm.
Owner · Engagement partner
Vague — non-policy
"Always review AI output before sending."
"Review" by whom, against what, with what evidence? The reviewer becomes the bottleneck — or stops reviewing.
Specific — operable
Any AI output that leaves the firm or affects a relationship gets a named reviewer before send. Internal drafts may be unreviewed. 1-in-10 sampled audit on auto-sent items.
04 · Disclosure
Disclosure.
When and how AI use is disclosed to the client.
Owner · Engagement partner
Vague — non-policy
"Be transparent with clients about AI."
Transparent when, where, at what granularity? Either you disclose nothing or you disclose everything — both annoy clients.
Specific — operable
Engagement letter clause v3 states AI is used for transcription, drafting, and analysis. Individual outputs aren't flagged. Opt-out available on request. No disclosure required mid-engagement.
05 · Escalation
Escalation.
Where AI stops and human judgment is non-negotiable.
Owner · Managing partner
Vague — non-policy
"Use judgment about when to bring AI in."
A policy that defers to judgment isn't a policy. Every team member draws the line in a different place — and the firm carries the liability.
Specific — operable
AI does not draft initial scope, sign off on financial statements, propose new pricing, or send communications on a client's tax position. These are partner decisions; material ones need a second-partner review.
The argument
A policy that uses the word "judgment" is not a policy.
The five topics are not the hard bit — most firms have all five on a page somewhere. The hard bit is committing to specifics: which tools, which fields, which reviewer, which clause, which line never gets crossed.
How to use this page
Print it, mark which column you're in for each row.
If you're in the left column on three or more, you don't have an AI policy — you have a wish. Move one to the right column this quarter, the next two next quarter. The five-row frame stays the same.
What it changes
Staff stop guessing. Partners stop carrying liability they don't know about.
Shadow AI surfaces — because the approved list exists. Reviewers stop being the bottleneck — because the trigger is named. The firm stops apologising to clients about AI — because the engagement letter already covered it.
Diagram 18
Part 5 · Build vs Buy
Chapter 22
The intelligent firm stack.
Five layers. Data captured at the bottom, intelligence and action at the top. Each layer has many vendors and no single right answer — the right answer is the one that talks to the others. MCP is the standard that makes the talking possible.
Layer 01 · Foundation
Capture.
Meetings, calls, comms, documents — heard, structured, joined to records.
VVinylMeeting AI · structured
TMS TeamsMeetings · transcripts
ZZoomMeetings · transcripts
SSharePointDocs · source files
◆Email · DMsInbound comms
MCP ↑ data up
Layer 02 · Record of work
Practice management.
The system of record for clients, engagements, jobs, tasks, and billables.
FFYIPM · docs · workflow
XXPMXero Practice Manager
IIgnitionProposals · engagement
AAccountKitWorking papers
LLevvyPM · next-gen
MCP ↑ data up
Layer 03 · The numbers
Financial intelligence.
General ledger, reporting, signals — the financial truth, queryable.
XXeroCloud GL
MMYOBCloud GL · enterprise
FFathomReporting · analysis
SSpotlightReporting · cashflow
SSyftMulti-entity analytics
MCP ↑ data up
Layer 04 · Reasoning
Intelligence layer.
Where retrieval, reasoning, drafting, and signal detection happen — over everything underneath.
CClaudeAnthropic · agents
MCopilotMicrosoft 365 AI
GChatGPTOpenAI · Team / Enterprise
GGeminiGoogle Workspace AI
PPerplexitySourced research
MCP ↑ action up
Layer 05 · Action
Workflow & automation.
What the firm does with the intelligence — agents, triggers, drafts that send themselves.
ZZapierNo-code automation
MMakeVisual workflows
nn8nSelf-hosted automation
PPower AutomateMicrosoft workflow
◆Anthropic SkillsReusable agents
The argument
No single vendor is the stack. The connections are the stack.
This diagram is on purpose vendor-light. Tools change every twelve months; the layers don't. The job of the firm is to pick one tool per layer that integrates well — not to find the one tool that does all five.
Why MCP matters here
The Model Context Protocol is the USB-C of this stack.
Without a shared protocol between layers, every integration is bespoke. With MCP, the intelligence layer can ask the practice manager what's open, ask the GL what's changed, and write back into both — without anyone wiring a webhook in the middle.
What it changes for buying decisions
"Best in its layer" beats "best at everything".
Stop shopping for a single product that does capture-plus-PM-plus-GL-plus-AI. There isn't one and there won't be. Buy the best capture, the best PM, the best ledger, and a model that can read all three. Vendor lock-in disappears when MCP is the contract.
Diagram 19
Part 6 · Revenue & Relationships
Chapter 25
The Pricing Confidence Review.
Four questions, four sources, one decision. Run this on any engagement before the renewal conversation. The data is already in the firm — agreement, capture, gap analysis, signals. The review just makes it visible. By the time you're in the meeting, you already know whether to maintain, adjust, or discuss.
01
What was agreed.
Source · Service LibraryServiceAdvisory · Tier 1
CadenceMonthly · QBR
Scope12 mtgs · 1 plan · BAS
Annual fee$48,000
Engagement letter v2 · signed 14 Jul FY26. The clause that defines the boundary.
02
What was delivered.
Source · Capture layerMeetings (YTD)14 of 9 planned
Ad-hoc calls+ 22 hrs
Extra advisoryCashflow scenarios x3
Realised cost$61,400
From transcripts, time, and the meeting record. The work the firm actually did.
03
What changed.
Source · Gap analysisScope drift+ 28%
Realised vs quoted1.6×
Effective rate$182/hr
Versus target− 26%
The delta between what we said and what we did. Scope crept; price didn't move.
04
What is the situation now.
Source · Financial signalsAcme margin− 4.2pp
Debtor days38 → 52
RelationshipStrong · Sue trusts us
Capacity for fee talkOpen this quarter
The client's financial reality plus the relationship signal. Both matter to the price.
Outcome
One of three.
✓
Maintain.
Quadrants line up · scope, delivery, change, situation all green. Renew at price. No meeting needed.
▸
Adjust.
Scope or delivery has shifted. Update engagement letter at next renewal · no surprise conversation.
!
Discuss.
Material gap and the relationship can carry it. Acme is here. Book the call. Lead with what changed, not what we want.
The argument
Price conversations should be small because the data did the work.
Most firms walk into a renewal with a feeling and a number. The intelligent firm walks in with four quadrants of evidence and a recommendation. The fee adjusts because the work changed — and the client can see it.
How to use this page
Print it. Fill the four sample boxes. Pick one outcome.
The four sources are already in the second brain. The four boxes are the agenda for a 15-minute internal review before any client-facing pricing conversation. Then the outcome decides the calendar.
What it changes
Firms stop undercharging quietly and stop apologising loudly.
Scope creep becomes visible the moment it happens. Discounts become deliberate, not accidental. The annual price-rise email goes from a sentence to a one-page brief — and the client signs, because the maths is right there.
Diagram 20
Part 6 · Revenue & Relationships
Chapter 28
The Systematic Engagement Loop.
Client engagement isn't a sequence; it's a rhythm that doesn't stop. The intelligent firm runs five stages on a continuous loop — clockwise, with no start and no end. The data flows around the circle. Every turn makes the next turn better.
The five stages of one turnSet the cadence. Listen. Brief. Talk. Record. Back to listening.
01
Stage 01 · Cadence
Engagement Rhythm.
When each client should be touched — and how.
Per-tier defaults set by the firm. Monthly for Tier 1; quarterly for Tier 2; annual for Tier 3. Signals can override the minimum at any time.
Firm policy02
Stage 02 · Listen
Signal Monitoring.
Watch the GL, the meetings, the comms, the calendar.
The second brain runs continuously. Margin drift, debtor slip, tone change, milestone approaching — anything that should change the next touch.
AI listens03
Stage 03 · Prepare
Advisor Brief.
Signals + history compressed into a structured pre-meeting brief.
One page. What's changed since last touch. What to ask. What to recommend. What the client almost certainly doesn't know yet.
AI drafts04
Stage 04 · Engage
Client Contact.
The conversation the brief makes possible.
Call, meeting, email — whatever the rhythm and the signal require. Advisor leads with something specific, recent, and useful. Not "how's business?".
Human leads05
Stage 05 · Close the loop
Record.
Decisions, commitments, sentiment — back into the brain.
The capture layer joins the contact to the client record, the engagement, and the next-touch trigger. The next turn of the loop starts the moment this stage closes.
AI captures
The argument
A loop, not a campaign.
Most firms run engagement as a quarterly project — a push of newsletters, a campaign of check-ins, a flurry of statements. The intelligent firm runs it as a permanent rhythm where five things are always happening, just at different positions on the wheel.
Why circular, not linear
There is no first stage. There is no last stage.
Record doesn't finish the engagement; it starts the next signal scan. A linear funnel implies completion. The reality is that the moment a client has been contacted, something has changed — and the brain is already onto the next thing.
What it changes
Retention stops being a quarterly worry.
Every client gets a properly prepared advisor every time they're contacted. Drift gets caught early. Expansion opportunities surface in stages 02 and 03 rather than at renewal. The advisor stops winging it — because the firm stopped letting them.
Diagram 21
Part 6 · Revenue & Relationships
Chapter 28
The Client Segmentation Matrix.
Three honest scores per client — fee level, complexity, relationship stage — and the tier writes itself. The point isn't the names of the tiers. It's that every client gets a deliberate engagement cadence instead of whatever the calendar happens to allow this month.
Client profileA representative type.
Fee level$ relative to book.
ComplexityWork behind the engagement.
Relationship stageTrust + history.
→ TierEngagement cadence.
High-trust advisor.
LMH
$ 60k+ pa · top decile of book.
LMH
Complex · group structures, FBT, advisory.
LMH
Established · 5+ yrs, strong trust.
Tier 1.
Monthly + signal-triggered
Growth business.
LMH
$ 24k pa · expanding scope.
LMH
Complex · funding, hiring, expansion.
LMH
Growing · 2 yrs · still proving.
Tier 1.
Monthly + signal-triggered
Established SME.
LMH
$ 18k pa · long-tail.
LMH
Moderate · single entity, payroll.
LMH
Established · 7+ yrs, low churn.
Tier 2.
Quarterly + signal-triggered
New medium client.
LMH
$ 14k pa · could grow.
LMH
Moderate · two entities + super.
LMH
New · < 12 mo · building.
Tier 2.
Quarterly + signal-triggered
Compliance-only SME.
LMH
$ 4-8k pa · bottom quartile.
LMH
Simple · BAS + tax only.
LMH
Established · 6+ yrs · stable.
Tier 3.
Annual + signal-triggered
Individual return.
LMH
$ 600 pa · transactional.
LMH
Simple · 2-3 line items.
LMH
New / casual · < 12 mo.
Tier 3.
Annual + signal-triggered
At-risk veteran.
LMH
$ 42k pa · top decile.
LMH
Moderate · steady scope.
LMH
At-risk · 3 mo no contact · drift.
Tier 1.
Override — partner contact this week
Tier definitions
What each tier means in practice.
Tier 1.Monthly + signals
Proactive advisory cadence. Partner-led. The brief, the call, the record — every month, plus anything the signal layer surfaces in between.
Tier 2.Quarterly + signals
Reliable review rhythm. Advisor-led. Quarterly check-in plus signal-triggered touches — drift, milestones, deadlines that move.
Tier 3.Annual + signals
Compliance cadence. Team-led. Annual return + statutory touches; the signal layer handles the rest. Tier upgrade triggers automatically.
The argument
Three honest scores. One deliberate cadence.
Segmentation isn't a marketing exercise. It's the firm telling itself who deserves what kind of attention — so the partner stops spending Tier 1 hours on Tier 3 work, and Tier 3 clients stop being ignored until something breaks.
How to use this page
Print it. Score every client in 30 minutes.
One row per client, three pills filled in. The tier on the right is the output — it should be obvious. Edge cases (last row) are where the override exists: relationship stage can pull a Tier 2 client up to Tier 1 for a quarter.
What it changes
The firm stops treating every client the same.
Engagement rhythm becomes deliberate. Reviews are scheduled, not begged. Expansion conversations happen with the clients ready for them. Retention stops being a defensive worry — and the partner gets their evenings back.
Diagram 22
Part 6 · Revenue & Relationships
Chapter 28
Engagement rhythm by tier.
Twelve months. Three rhythms. Tier 1 sees a partner every month. Tier 2 every quarter. Tier 3 every year. On top of all three, the signal layer adds contact whenever something deserves it — debtor drift, a missed BAS, an industry change. The minimum is scheduled. The maximum is what the firm notices.
Legend
Scheduled touchpoint · minimum cadence
Signal-triggered · added when something changes
FY26 · 12 months
JUL'25
AUG'25
SEPQ1
OCT'25
NOV'25
DECQ2
JAN'26
FEB'26
MARQ3
APR'26
MAY'26
JUNQ4
Tier 01
Monthly.
Partner-led advisory. Brief, call, record.
12 scheduled · plus signal triggersA
JulA
AugQ
Sep QBRA
OctA
NovQ
Dec QBRA
JanA
FebQ
Mar QBRA
AprA
MayQ
Jun annual!
debtor drift!
margin alert!
Div 7A flag
Tier 02
Quarterly.
Advisor-led review at the close of each quarter.
4 scheduled · plus signal triggersQ1
SepQ2
DecQ3
MarQ4
Jun annual!
BAS missed!
milestone
Tier 03
Annual.
Team-led return and statutory work.
1 scheduled · signals only otherwise●
annual return!
PAYG variance!
tier-up trigger
The argument
Cadence is a promise, not a calendar invite.
The scheduled touchpoints define what every client at that tier can rely on. The signal markers define what the firm catches without being asked. Together they replace the "we'll call you when we need to" relationship with one the client can actually feel.
How signals layer over rhythm
Signals only add — they never remove.
A Tier 3 client still gets their annual. A signal-triggered touch is an addition, not a substitution. Over time, repeated signal touches on a Tier 3 client become the upgrade conversation: the rhythm should be quarterly.
What it changes
No client gets ghosted. No partner gets surprised.
The chart is the firm's commitment to every client written down once, not relitigated every month. Capacity becomes visible. Tier upgrades and downgrades become evidence-driven. And the partner stops apologising for being too busy to talk.
Diagram 23
Part 7 · Your Team
Chapter 30
Human in the loop.
Six stages, one continuous loop. AI does what AI is good at — reading, structuring, drafting, monitoring. Humans do what humans must — reviewing, deciding, owning. The whole loop exists to protect one stage: the moment of decision. Everything else is the work that gets the human to that moment with everything they need.
Six stages of the loopThree things AI does. Two things only humans do. One anchor.
01
Stage 01 · Input
Task Arrives.
A piece of work enters the firm's workflow.
A transcript, a query, a dataset, a document. The trigger that starts a turn of the loop.
System02
Stage 02 · First pass
AI Processes.
Transcribe, structure, draft, extract, flag, summarise.
The mechanical work. Tokens not judgement. Speed at scale.
AI03
Stage 03 · Review
Human Reviews.
Qualified person checks the AI output.
Correct, refine, or approve. Quality stays high because a human stayed in the loop.
Human04
Stage 04 · The anchor
Human Decides · Acts.
Accountability. Relationship. Judgement under uncertainty.
The moment the whole loop exists to protect. No AI sign-off here.
Human only05
Stage 05 · Memory
AI Captures.
Decision + outcome structured back into the second brain.
What the human decided, what they said, what they meant. Joined to client, engagement, signal.
AI06
Stage 06 · Watch
AI Monitors.
Patterns surface; the next moment is flagged.
The signal layer is always on. It opens the next loop — back to Stage 1.
AI
Five human-keep
Things AI never decides.
Relationships.
Anything that changes how a client feels about the firm.
Accountability.
Sign-off that carries professional liability.
Pricing.
Scope, fee, scope-change conversation.
Judgement.
Decisions under uncertainty where the answer isn't in the records.
Final word.
Anything that leaves the firm with the firm's name on it.
The argument
The loop is the answer to "where does AI stop?"
It doesn't stop. It hands off — twice. To a human reviewer at Stage 3, to a human decision-maker at Stage 4. Then it picks the work back up at Stage 5. The human is in the loop, not outside it.
Why Stage 4 is the anchor
Five other stages exist to make Stage 4 cheaper.
If the human arrives at the decision with everything they need — sources, draft, precedent, signal — the decision takes two minutes and the answer is right. If they arrive cold, the decision takes an hour and the firm rolls a dice.
What it changes for the team
AI doesn't replace the role. It changes the work.
Less typing. Less re-reading. Less remembering. More deciding. The most valuable hour in the day stays where it was — the partner's hour. It just gets ten times as much done.
Diagram 24
Part 7 · Your Team
Chapter 30
The AI Task Inventory & Human-Keep card.
A two-sided reference card. Side A maps every common firm task onto a role and a tier — what AI does, what AI drafts, what AI assists with. Side B names the five categories that never leave a human's hands, with examples. Pull it out for any AI adoption workshop. Print it. Pin it.
Side A
5 roles · 3 tiers · 50+ tasks
AI Task Inventory — by role × tier.
Tier ↓ · Role →
Read across to find what AI does for each role.
01Partner
02Manager
03Senior
04Bookkeeper
05Admin
Tier 1
Fully automate.
AI does it. Audit only.
Meeting transcription & summary
Calendar prep & brief auto-pull
Inbox sorting by client + topic
Post-meeting action capture
Status report rollups
WIP & budget variance flags
Workflow nudges & reminders
Capacity heatmap refresh
First-pass review checklists
Reconciliation difference scans
Standing query bundles
Document tagging & filing
Bank rule-based coding
Receipt + invoice OCR
Bill payment scheduling
BAS pre-fill from GL
Client onboarding intake
Engagement letter drafts
Auto-replies for FAQs
Document e-sign chase-ups
Tier 2
AI drafts · human reviews.
Output goes nowhere until signed.
Client comms & status emails
Engagement scoping notes
Pre-meeting advisory briefs
Board paper drafts
Team status / 1:1 prep notes
Engagement margin reviews
Client account summaries
Workflow exception triage
Working paper drafts
Tax position notes
Client query responses
Variance commentary
Edge-case classifications
Reconciliation narrative
Director loan reviews
Payroll exception drafts
Client outreach drafts
Meeting agendas & minutes
Renewal & expansion letters
Internal status memos
Tier 3
AI assists.
Human leads. AI on demand.
Pricing & scope decisions
Client relationship calls
Strategic plan & risk sign-off
Hiring & team decisions
Escalations to partner
Difficult conversations · staff
Client expansion proposals
Engagement decisions
Complex client queries
Judgement on technical positions
Mentoring & supervision
Client review meetings
Material classification calls
Client-facing explanations
Process changes & sign-off
Account close decisions
First contact with new clients
Sensitive comms triage
Process improvement input
Cultural & team moments
Side B
Things AI never decides
Five Human-Keep categories.
Relationships.
Anything that changes how the client feels about the firm — or how the team feels about each other.
For example
First client call after a price increase
A difficult performance conversation
A "thank you" to a long-term client
Never AI: tone & trust live with people.
Accountability.
Sign-offs that carry professional liability or regulatory weight.
For example
Financial statement sign-off
Tax return final lodgement
Audit opinion & engagement letter signing
Never AI: the name on the file is a human's.
Pricing.
Scope, fee, and scope-change conversations — the price the firm charges to do the work.
For example
New engagement quotes
Annual price-rise conversations
Scope creep "discuss" outcome
Never AI: price is a relationship signal.
Judgement.
Decisions under uncertainty where the answer isn't in the records — the technical, ethical, or strategic call.
For example
Technical tax position with unclear case law
Material accounting estimate
Client risk acceptance / decline
Never AI: judgement is what you're paid for.
Final word.
Anything that leaves the firm with the firm's name on it — a regulator, a client, a counter-party.
For example
Letters to the ATO
Final advice memos to a client
Engagement-defining outputs
Never AI: last review is human, always.
The argument
A reference card, not a philosophy.
Side A is what people in your firm can do tomorrow with AI; Side B is what they should never let AI do — even when the prompt looks tempting. Two sides. One conversation. Pull it out at the start of any AI adoption discussion.
How to use it
Workshop agenda in a single page.
Walk Side A column by column. Each role finds three cells they want to start with this quarter. Then turn the card over and confirm Side B — these don't move. Adoption stops being a debate and becomes a decision.
What it changes
Shadow AI stops being a shadow.
Staff stop pretending they aren't using AI; partners stop pretending they don't know. The card is permission slip and stop sign in one — and the firm gets to the conversation everyone has been avoiding for two years.
Diagram 25
Part 7 · Your Team
Chapter 31
The 30-day AI pilot calendar.
Five weeks. Three tracks. One decision at the end. The pre-week is for the Champion alone; weeks one through four bring the pilot group in, then taper the Champion back out, with the rest of the firm watching the whole time. The week-four debrief is the moment the firm answers a single question: keep, modify, or kill.
Track ↓ · Week →
Three swim lanes across thirty days plus a setup week.
Pre-week
Setup.
Day −7 · −1
Week 01
Kickoff.
Day 1 · 7
Week 02
Clear friction.
Day 8 · 14
Week 03
Hands off.
Day 15 · 21
Week 04 · Anchor
Decide.
Day 22 · 30
Track 01
The pilot group.
3–5 staff doing real client work with the tool.
Not yet involved
Day 1▸
Kickoff session.
90 min intro · install · first prompt. Champion runs.
Day 2 – 7▸
One real task each.
Every pilot user applies the tool to a real piece of client work this week.
Day 8 – 14▸
Use + log.
Daily pilot log entry — wins, friction, time saved or lost.
Day 10▸
Office hours.
30 min drop-in with Champion. Friction list cleared.
Day 15 – 21▸
Unassisted use.
No Champion intervention. Pilot users figure it out. Log continues.
Day 19▸
Peer share.
Pilot group shares prompts and wins with each other — no Champion.
Day 22 – 29▸
Final unassisted use.
Last week of real client work with the tool. Log everything.
Day 30▸
Debrief survey.
Each pilot user submits answers to the five debrief questions before the meeting.
Track 02
The AI champion.
The single named owner running the pilot — taper out by week three.
Day −7 · −2◆
Setup & edge cases.
Tool access provisioned. Pilot group chosen. Edge cases tested in private.
Day −1◆
Prompt template.
First prompt template built and saved for the pilot group to use day one.
Day 1◆
Run kickoff.
Lead the session. Set the log. Set the rules.
Day 4 · 7◆
Async check-ins.
Read the log daily; respond to friction within a day.
Day 10◆
Midweek check-in.
Office hours. Clear friction. Update the prompt template if needed.
Day 14◆
Log reminder.
Light nudge to keep logging through the unsupported week.
Day 15 · 21◆
Stay out.
No intervention. Read the log. Note the gaps. Resist the urge to help.
Day 30◆
Debrief + decision.
Walk the log. Score the wins. Make the call: keep · modify · kill. Sign-off with managing partner.
Pilot anchor · decision point
Track 03
The wider firm.
Everyone who isn't in the pilot — what they see, when they hear.
Day −1◇
Announce.
All-firm note: pilot starting Monday, who's in, what we're trying, why this not that.
Day 7◇
Wins & learnings · 1.
End-of-week note: 3 wins, 3 frictions. Honest. No spin.
Day 14◇
Wins & learnings · 2.
Second weekly. The friction list is shorter. The wins are sharper.
Day 21◇
Wins & learnings · 3.
Third weekly. Without Champion support, what's still working.
Day 30 + 1◇
Rollout note.
All-firm: what we decided · who's next · when the rollout begins.
Legend
Pilot event · blue
Champion activity · amber
Firm-wide comm · green
Anchor · decision point
The argument
A pilot dies in week three. Design for it.
Most AI pilots stall in week three — the Champion has moved on, the pilot group has lost momentum, and the firm has stopped looking. The week-three "hands off" block is on the page on purpose: it's where the question "does this stick without me?" gets answered.
How to use this page
Print it. Pin it. Fill the dates.
This is a planning template, not a Gantt chart. Set the start date, replace the placeholder content with your tool, your group, your prompts. The structure stays; the specifics are yours. Downloadable at theintelligentfirm.com.
Week-four is the anchor
Keep · modify · kill.
The whole calendar exists to make Day 30 a decision, not a debate. Pilot group submits the survey. Champion brings the log. Managing partner makes the call. Then the firm-wide note goes out the next day. No drift. No quiet death.
Diagram 26
Part 7 · Your Team
Chapter 32
The AI Champion role card.
Hand this card to a new Champion on day one. The four mandate dimensions define what the job actually is. The 90-Day Standard says how to start. The Committee diagram is who they report to and who they call when something is on fire. Three things, one page.
Mandate
What the job actually is
Four dimensions of the role.
01
Capability.
Build and maintain the firm's working AI toolkit. Prompts, templates, agents, integrations.
Owns
Approved-tool list & access provisioning
Firm prompt library & agent templates
MCP & integration map
02
Adoption.
Move tools from "interesting" to "embedded." Pilots, training, peer-share, momentum.
Owns
Pilot calendar & debrief decisions
Onboarding for new hires (AI track)
Internal show-and-tell rhythm
03
Governance.
Hold the line on policy, data, and risk. Say no when it matters. Say yes loudly when it doesn't.
Owns
The five AI policies & the version log
Vendor review · approval · sunset
Incident response & the kill-switch list
04
Communication.
Translate. Up to partners, across to teams, out to clients. Make the work visible.
Owns
Weekly wins-and-learnings note
Quarterly partner brief
Client-facing disclosure & FAQs
90-Day Standard
How to start
From day one to day ninety.
30
Days 0–30
Foundation.
See what's already happening. Don't add anything yet.
By day 30
Shadow-AI audit done · what's already in use
Five policies in draft · two signed off
Approved-tool list v1 published
Champion office hours · weekly
60
Days 31–60
Pilot.
Run one real pilot end-to-end with the 30-day calendar.
By day 60
One pilot complete · log filed
Keep · modify · kill decision posted
Prompt library v1 · 8–12 prompts
First quarterly partner brief
90
Days 61–90 · Anchor
Scale.
Make adoption durable without you being the bottleneck.
By day 90
Two more pilots running · owned by others
AI Committee meeting cadence live
Engagement letter clause v3 in use
Successor candidate identified
Committee
Three nodes · clear lines
Who the Champion answers to and works with.
MP
Sponsor
Managing partner.
Final authority. Sets the mandate. Signs off on policy and spend.
Provides
Strategic intent & budget envelope
Air cover for the unpopular calls
Monthly · 30 min
CH
Operational owner
AI Champion.
Runs the four mandates. Owns the 90-day. First point of contact for the firm.
Held by
One named person · 1.0 FTE or 0.5 + backup
Bias to action; technical credibility
All day · every day
CM
Working group
AI Committee.
Cross-discipline panel. Reviews pilots, votes on tool list, advises on policy.
Members
Champion (chair) · 1× partner
1× tax · 1× audit · 1× ops · 1× IT/risk
Quarterly · 90 min
The argument
A role becomes a role when it has a card.
Without a card, every AI conversation in the firm flows to whoever happened to read the most LinkedIn posts last week. With a card, there is one named human, four named mandates, and one ninety-day plan. The work stops being political.
How to use it
Hand it over on day one.
The partner gives this card to the new Champion on day one. Within a week the Champion ticks off the day-30 column dates with their own. Within a quarter, they're running the committee. Within two, they're training their successor.
What it changes
AI stops being everyone's side hobby.
Capability gets built deliberately. Adoption survives the pilot. Governance moves at the speed of business. Communication stops being silence. And one person is accountable when the firm asks "who owns this?"
Diagram 00 · variation C
The whole book · as an orbit
Everything orbits capture.
Four concentric rings, one centre. Capture sits at the core because nothing further out is real without it. Structure rings capture; intelligence rings structure; scale is the outermost — the visible orbit the market actually sees.
Reading from the centre outEach ring rests on the one inside it.
01
Core · Capture
The work becomes the record.
Without the core, every ring further out is built on someone's memory.
1 · Line in the sand
2 · Second brain
02
Ring · Structure
Captures are joined to context.
Client, engagement, person, decision. The folder becomes an asset.
2 · Second brain
4 · Living docs
03
Ring · Intelligence
The firm becomes queryable.
Signals surface. The OS holds what's decided. The stack does the talking.
3 · Signals
4 · Firm OS
5 · Build vs Buy
04
Orbit · Scale
What the market sees.
Revenue, relationships, team, the future. The outermost orbit — the only one most firms try to fix.
6 · Revenue
7 · Team
8 · Future
The metaphor
A solar system, not a chain.
The chain reading suggests linear motion. The orbit makes dependency obvious — the inside rings are why the outside ring exists. Pull capture out of the centre and the whole picture collapses.
Where firms misread it
They optimise the orbit they live on.
Most firms only feel the outer ring — revenue, team, growth — so that's where they spend. The book is a slow walk inward to where the real work is.
What it reframes
Capture isn't a project. It's the centre of gravity.
Treat capture as the orbital anchor and every other decision — what to buy, how to scale, who to hire — gets its physics right.