Signals

Your firm is swimming in information it is not reading.

Not because the information is hidden. Because it has never had anywhere useful to go.

Every client interaction produces signals. The response time that has been getting slower over the last three months. The tone shift in the last two meetings that never made it into the notes because nobody wrote them up. The question the client stopped asking that they always used to ask. The number that moved in a direction that should have prompted a conversation and did not because nobody was watching it closely enough at the time.

Every team interaction produces signals. The advisor whose output quality has been quietly dropping. The three people who went quiet in the same meeting. The pattern across multiple client relationships that suggests a process is failing, not an individual.

The numbers produce signals constantly. GL movements that should not have happened. Margin compression that started three months before anyone named it. Cashflow patterns that predicted a problem the client is about to call you about.

The signals are there. They have always been there. The question has never been whether the information exists. It is whether the firm has the infrastructure to hear it.

Most firms do not.

Something changed recently that makes this more urgent than it has ever been.

In 2025, both Xero and Intuit announced partnerships with Anthropic, the company behind Claude. What this means in practice is that the financial data sitting inside the two most widely used accounting platforms in the world is becoming natively queryable by one of the most capable AI reasoning systems available. The GL is no longer just a ledger. It is a data layer that an AI can read, interpret, and draw conclusions from in real time.

Karbon, Ignition, and a growing number of practice management and workflow platforms are moving toward MCP support. MCP, Model Context Protocol, is the universal connector that allows AI systems to access data across different software without custom integrations for each one. Think of it the way USB changed computing. Before USB, every device needed its own connector. After USB, one standard worked for everything. MCP does the same for AI. When your practice management tool is MCP-enabled, the client record, the workflow history, the billing data, and the engagement timeline all become part of what the AI can see when it is helping you.

This is the moment where signals stop being something a firm reads manually and start being something the firm's AI infrastructure surfaces automatically. The relationship signals in the meeting history. The financial signals in the GL. The team signals in the workflow data. All of it connected, all of it queryable, all of it visible to an AI that can surface the right thing to the right person at the right time.

The firms that have built the second brain in Part 2 are now positioned to take advantage of this. The firms that have not are about to have very powerful AI tools sitting on top of very incomplete data. And as established in Chapter 4, incomplete data does not produce worse AI output. It produces confident AI output that is wrong.

Before going further, it is worth being honest about the main AI tools a firm is likely to encounter and what each one is actually for.

Nick Ferguson is Head of Professional Services and AI Solutions at Strategic Group in Brisbane, Australia. He leads a team of consultants delivering advice across AI, digital transformation, security, and change management, helping organisations apply AI in ways that actually work for people rather than just impressive in a demo. He has deployed these tools inside accounting and professional services firms at scale, which means he has watched what happens when firms pick tools without understanding the tradeoffs.

Nick rates the three main tools across two dimensions that actually matter for firm deployment.

Capability: Claude leads at ten out of ten. ChatGPT at nine. Copilot at eight. For complex reasoning, long document analysis, and tasks requiring deep context across large amounts of information, Claude consistently produces the most accurate and useful output.

Security and integration: Copilot leads at ten out of ten. Claude at six. ChatGPT at five. Copilot inherits Microsoft's existing role-based access controls, meaning it can immediately access the full Microsoft environment relevant to each user. Outlook, Teams, Excel, Word, SharePoint. If the user can find the file, Copilot can too.

That is not a complicated framework. Copilot for the firm. Claude for the people doing the most complex and high-stakes work. The two tools complement rather than compete.

Nick's broader observation is worth sitting with. Partners who fixate on Claude's superior capability and reject Copilot are optimising for the wrong dimension. The rest of the team does not need to be innovating. They need to be billing. And for billing work inside a Microsoft-first environment, Copilot's integration advantage outweighs the capability gap.

The security question is not something to figure out after deployment. It is the first question every firm should ask before any AI tool touches client data. Where does the data go? Who can see it? Is it used for model training? What is the retention policy? The tool that scores highest on a demo does not automatically have the right answers to those questions.

Do your due diligence. Build your AI policy before you give anyone access. Every firm is different. The right answer for your firm depends on your size, your data sensitivity, your existing infrastructure, and your risk appetite. Chapter 16 in Part 4 covers this in full.

The practical output of a signal infrastructure is not a dashboard nobody checks. It is a rhythm.

A daily briefing, delivered to the practice manager's inbox or the firm's Slack or Teams channel, built overnight by an AI that has read every meeting summary from the day before, checked every outstanding action item, and cross-referenced client deadlines against team capacity. The three things that need attention before noon. The two client relationships that have shown a pattern worth watching. The one financial movement in a client's accounts that is worth a conversation before it becomes a problem.

A weekly summary, broader in scope, showing the patterns across the whole firm. Which clients are highly engaged and which have gone quiet. Which team members are at capacity and which have headroom. Which financial signals have been compounding across multiple clients in the same industry.

A threshold alert, immediate and specific, when something crosses from soft signal to hard signal. A client whose response times have dropped below a certain threshold. A GL movement outside the expected range. A team member whose output quality has shifted in a way that warrants a conversation.

This is what the intelligent firm's signal infrastructure actually produces. Not information overload. The right information, to the right person, at the right time, with enough context to act on it.

The Signal Stack is the framework that makes this manageable.

Not every piece of information the system surfaces requires action. Not every movement in the data is a signal worth responding to. The Signal Stack has four levels.

Level 1 is noise. Information the system processes and handles without surfacing to any human. Routine patterns within expected ranges. Data that confirms what is already known.

Level 2 is soft signals. Patterns worth watching but not acting on yet. The system monitors them, notes when they are compounding, and keeps a flag on them. No human attention required unless they escalate.

Level 3 is hard signals. A threshold has been crossed. Something has moved outside the expected range in a way that warrants a human decision. The system surfaces it to the right person with the context they need to act.

Level 4 is critical signals. Something requires immediate human judgment. The system gets out of the way entirely. The flag goes up, the human takes over, and the system supports rather than leads.

Most firms operate without this framework. Every piece of information either gets noticed or does not, depending on who happened to be paying attention that day and how busy they were. The intelligent firm is different. It knows what it is looking for. It knows what each signal means. And it knows exactly which level of response each one deserves.

Part 3 builds this capability across five chapters.

Chapter 7 is about the foundational idea. Everything is a signal. Every interaction, every number, every pattern in client behaviour is information the firm can learn from. The intelligent firm treats data as a continuous conversation rather than a periodic report.

Chapter 8 is about the Signal Stack in practice. How to separate noise from signals worth watching, and signals worth watching from signals worth acting on.

Chapter 9 is about client signals. The ones that predict churn before it happens, identify expansion opportunities before the client articulates them, and tell you more about the health of a relationship than any financial report.

Chapter 10 is about team signals. The early warning system most firms do not have for their most important asset.

Chapter 11 is about financial signals. The numbers that are telling you something, now connected to AI systems that can read them in real time across every client simultaneously.

The second brain is listening. Part 3 is about learning to hear what it is saying.