Your business doesn’t need another AI workshop. Read the exhaust first.

The workshop is not the problem. The starting point is.

Ask a leadership team where AI could help and you will usually get a familiar list: draft emails, summarise documents, automate reporting, improve customer service. None of those ideas is necessarily wrong. They are simply untethered from the operating reality of the business.

Workshops privilege what is visible, recent and easy to describe. The loudest frustration gets airtime. The process everyone has learned to tolerate disappears. A senior person's view of the workflow can be entirely rational and still differ from what happens at 4:40 on a Thursday when the system is slow, the customer is chasing and somebody needs to get the job out.

That is why we prefer to begin with evidence the business is already producing. Not because data is objective and people are not, but because the gap between the record and the explanation is often where the useful question lives.

Every operating business leaves an exhaust

The exhaust is the trail left by ordinary work. A quote is created, revised and accepted. A job moves through stages. An invoice is raised and paid. A customer writes, calls, waits and follows up. A manager's calendar fills with recurring interventions. A spreadsheet appears because the main system cannot answer the question somebody actually needs answered.

Individually, these records are mundane. Connected around a decision or workflow, they can reveal where work slows, where margin leaks, where context is lost and where experienced people are quietly holding the operation together.

The useful evidence is rarely contained in one perfect system. It may sit across accounting software, a CRM, email, project tools, calendars, call records and the local spreadsheet that everybody says is temporary. The job is not to centralise everything. It is to connect enough of the trail to understand one consequential piece of work.

• Timestamps reveal waiting and handoff delays.
• Edits and overrides reveal where the standard process does not fit reality.
• Repeated customer contact reveals where confidence or context breaks down.
• Margin by job, customer, product or lane reveals where activity and value diverge.
• Calendars and recurring meetings reveal where management attention is being consumed.
• Side spreadsheets reveal questions the official systems cannot answer.

Stated strategy and revealed strategy are not always the same

A business may say that customer retention is its priority while its best people spend every week rescuing new work. It may say that a service line is strategic while leadership time, hiring and investment point elsewhere. It may believe a workflow problem belongs to operations when the evidence shows that pricing, sales qualification or product variation created it upstream.

This is not an accusation of hypocrisy. Businesses adapt. Teams create workarounds to protect customers. Managers fill gaps because the formal process has not caught up. The revealed strategy is simply the strategy expressed through attention, money and repeated decisions.

AI introduced without understanding that reality tends to automate the visible task. It may make the workaround faster while leaving the underlying constraint intact. In the worst case, it hardens an accidental process into infrastructure.

People explain what the systems cannot

Reading the exhaust is not a case for replacing interviews with dashboards. The record can show that jobs with one characteristic sit forty per cent longer at a particular stage. It cannot reliably tell you whether the cause is supplier uncertainty, a missing approval, an experienced employee protecting quality or a field that nobody trusts.

The evidence makes the human conversation better. Instead of asking, ‘Is this process efficient?’, we can ask, ‘These jobs wait here longer than the others. What is different about them?’ The person closest to the work can challenge the data, explain the exception and point to the actual decision that needs to change.

That conversation should never become employee surveillance dressed up as transformation. Use the minimum evidence required, agree access and purpose, avoid individual performance inference where the question is about the system, and let people see and correct the interpretation of their work.

The order matters

A disciplined discovery does not need months of analysis. It needs a specific business question and a sensible sequence. Define the constraint. Find the systems and people that hold evidence about it. Connect a small sample. Look for patterns and contradictions. Ask the people doing the work to explain them. Estimate the commercial consequence. Only then choose the smallest intervention that could prove value.

The answer may be a clearer rule, a changed handoff, a report people can finally trust, configuration in software already owned, an integration, or an AI-assisted workflow. The point is not to avoid AI. It is to make AI earn its place.

• Name one decision, handoff or operating constraint.
• Read the existing trail before requesting new reporting from the team.
• Treat every pattern as a hypothesis until people and a second source support it.
• Separate the root cause from the workaround that made it visible.
• Put a value and an owner against the change before building.

A better workshop comes after the evidence

Once the operating picture is visible, a workshop becomes genuinely useful. Leaders can make trade-offs with a shared view of the business. Teams can distinguish an irritating task from an expensive constraint. Technology choices can be evaluated against cost, flexibility, control and the way the business actually works.

The first deliverable is not an AI roadmap. It is a better question, backed by enough evidence to decide what happens next. Sometimes that leads quickly to deployment. Sometimes it prevents the business spending six months automating the wrong thing. Both are valuable outcomes.