guide · illustrative
Which workflow should a professional-services business automate first?
A practical way to select repeated work that is useful to improve, safe to review and honest to measure.
The first AI workflow should not be the most glamorous idea on the list. It should be a piece of work the business understands well enough to improve responsibly.
That usually means a job that is repeated, has recognisable inputs, produces a useful output and still leaves room for a person to review the result.
Use five questions to shortlist the work
| Question | Why it matters |
|---|---|
| Does it happen regularly? | A repeated job makes learning and measurement possible. |
| Can the trigger and expected output be described? | A workflow cannot be designed around a vague aspiration. |
| Is the context available and owned? | A system cannot use information that is missing, stale or inaccessible. |
| Can a person review the output safely? | The first workflow should make review clearer, not hide risk. |
| Would an improvement be visible? | Look for turnaround, rework, quality, capacity or client-experience signals. |
No single “score” should decide the answer. The purpose of the questions is to expose what still needs to be understood.
Good first-workflow characteristics
A good candidate is often mechanical in its preparation and meaningful in its outcome. It may involve routine intake, document collection, a report structure, recurring follow-up, source organisation or an internal hand-off.
These tasks are not trivial. They often absorb senior attention because the current process is scattered, every input arrives differently or nobody is certain which material is current. Making that preparation more consistent can free experts to focus on the decisions that genuinely need them.
Poor first-workflow characteristics
Avoid beginning with a rare, high-stakes exception or a task nobody owns. Also avoid work where the organisation cannot explain what “done” means, has no usable source material or expects a model to make an accountable client decision without review.
A poor first workflow often produces a polished demo and little operational learning.
Separate preparation from decision
One way to make a workflow safer is to split it into parts:
- Prepare: collect inputs, organise sources, draft a starting point or identify what is missing.
- Check: compare the result against agreed rules and flag uncertain or unusual cases.
- Decide: let the accountable person make the professional judgement or client commitment.
This does not make a workflow less ambitious. It makes the responsibilities visible enough that the team can trust what changes.
Choose a measure before you build
Do not wait until the end of a pilot to decide whether it helped. Agree a baseline first.
For example, a reporting workflow might track the time required to prepare a first draft, the number of clarification cycles, reviewer effort and on-time completion. An intake workflow might track incomplete submissions, follow-up time and the time before work can start.
The metric needs to describe the actual work. Every released hour is not automatically a financial result; it may become capacity, quality or a better client experience instead.
A practical next step
Write down one job that happens at least weekly. Note what starts it, what information it needs, who checks it and what “done” looks like. That is enough to begin a useful workflow conversation.
The assessment helps turn that outline into a clearer starting point.