Start here
The answer is already in your task list
Most firms asking “where do we start with AI?” already have the answer sitting in their weekly task list. They just have not looked at it the right way.
The problem is not a shortage of tools. It is that most leaders approach AI adoption by browsing vendor demos, watching webinars, and asking “could this help us?” That is the wrong question. It produces a long list of possibilities and zero clarity on what to do first — generic advice that sounds useful but cannot survive contact with a real brief.
This audit takes five minutes. It replaces the noise with a simple scoring method that tells you which tasks are worth automating and which to ignore entirely.
The filter
The three-box test
Before you score anything, write three words at the top of a page: repetitive, rules-based, time-consuming. These are your three boxes. Every task you consider for automation passes through them. Tick all three and AI can probably help. Miss one and you move on.
- Repetitive. Your team does this at least weekly. Often daily. Sometimes several times a day.
- Rules-based. There is a clear right and wrong way to do it — you could write a one-page set of instructions a new starter could follow from day one. If it needs significant judgement or changes substantially each time, it is not rules-based enough.
- Time-consuming. It takes at least 30 minutes each time, or happens often enough that the total adds up to something meaningful across the team.
Score your tasks against three words, focus on the threes, and the noise disappears.
The method
How to run the audit
Step one: list your tasks. On the left of a page, write down every task your team does regularly. Think about what happens every Monday morning, what makes people say “not this again,” what a new hire spends their first months learning. Common ones in professional-services firms: responding to enquiries, updating CRM records after calls, writing social posts, summarising meeting notes, chasing invoices, producing weekly reports, and answering the same client questions repeatedly. Give yourself two minutes and aim for at least ten tasks.
Step two: score each task 0 to 3. One point for each box it ticks — one if repetitive, one if rules-based, one if time-consuming (30 minutes or more per occurrence, or high volume).
Step three: focus on the threes. Any task scoring three is a strong candidate for automation. Anything below two is probably not worth prioritising yet.
Calibration
Two worked examples
Responding to customer enquiries. Repetitive? Yes — daily, sometimes hourly. One point. Rules-based? Partially; some enquiries follow a predictable pattern, others need real judgement, so call it half a point. Time-consuming? Yes — across the team it easily adds up to an hour or more a day. Another point. Total: 2.5. Worth investigating, but not the top priority.
Updating the CRM after a sales call. Repetitive? Yes — every single call. One point. Rules-based? Yes — you are pulling specific information into specific fields. One point. Time-consuming? Yes — 15 minutes per call, several calls a day. One point. Total: 3. Exactly the kind of task AI handles well: structured, predictable, and currently draining time your fee earners could spend on client work.
The pattern
What the audit usually reveals
The tasks that score highest are almost never the interesting ones. They are the back-office operations — the data entry, the formatting, the status updates, the chasing. Not the creative work. Not the strategic delivery. The admin.
That is not a surprise. It is the point. Repetitive, structured, rules-based work is where automation earns its place — not writing your firm’s methodology, not replacing the expertise clients pay for. The businesses getting real results from AI are not automating their core value; they are automating the tasks that drain time away from it. Remove that drag and you start to see where your real capacity gains sit, and that is where the return lives.
In the KWA framework — Knowledge, Workflow, Agents — this audit sits at the Workflow layer: before you can automate anything, you need to know which workflows are actually worth standardising. The three-box test is how you find them.
The next move
What to do with your high scorers
For each task scoring three, answer three questions before you touch a single tool:
- What triggers the task? An email arriving, a meeting ending, a date in the calendar, a form submission. Be specific.
- What is the output? A message sent, a record updated, a document created, a report shared.
- Where does the information come from? Your inbox, a spreadsheet, your CRM, a call transcript.
With those three answers you have enough to research what solutions exist — a workflow tool like n8n or Zapier, a purpose-built integration, or something worth building with support. What you should not do is buy anything yet. Do not sign up to subscriptions. Do not brief a vendor. The audit tells you where to look; validate that a solution genuinely addresses the problem before you spend money. Most AI tool purchases that fail do so because the buyer skipped this step.
The principle
Start with the tasks, not the tools
Repetitive, rules-based, time-consuming. Those three words are your filter. Score your tasks against them, focus on the threes, and work out what triggers each one and what the output looks like before you research solutions.
That is what making AI work so your team can deliver actually means in practice: you do not start with tools. You start with the tasks that are draining the time your people should be spending on the work that moves the firm forward.