Every growing consultancy, agency or professional service firm has a person who “just knows how things work”.
They know the client onboarding steps. They know which invoice rules apply to awkward cases. They know how the monthly report is compiled, which spreadsheet needs checking, and which email template should be used when a project changes scope.
That knowledge is useful, but it is also a risk. If it lives mainly in one person’s head, the business has a single point of failure. The risk usually becomes obvious when that person is off sick, on holiday, overloaded, or leaving.
Most firms know they should document their processes. Fewer know where to start. The task feels too big, so it gets postponed.
An example AI workflow: finding documentation gaps
This is an example workflow, not a real client case study. The aim is not to let AI write official process documents on its own. The aim is to find the gaps so the team can prioritise what to document first.
The assistant scans a defined set of internal sources. That might include a shared drive of existing SOPs, a list of recurring internal tasks from a project management tool, and, if appropriate, recent internal emails or Slack threads where people have described “how to” steps.
It then produces a gap report showing:
- recurring tasks that appear to have no written SOP at all;
- existing SOPs that have not been updated in over a year;
- processes where several versions of the truth appear to exist;
- a suggested priority order based on frequency, business risk and dependency on one person;
- a draft outline for the top one or two missing SOPs.
The output is a ranked report, not a finished operations manual. A manager or operations lead reviews the findings before anything becomes official guidance.
Why this is useful
A manual process audit can take hours before anyone writes a single useful document. Someone has to search folders, ask colleagues, compare task lists and work out whether old instructions are still current.
AI can speed up that discovery work. It can compare a list of recurring tasks with the documents that already exist. It can spot when “client onboarding” appears in the project management system but no current onboarding SOP exists in the shared drive. It can also draft a skeleton structure, such as purpose, owner, inputs, steps, exceptions and approval points.
That gives the team a starting point. The person who understands the process can then correct and fill in the outline instead of starting from a blank page.
Good boundaries keep it safe
This is a low-risk workflow when it is kept to audit and drafting. The assistant should not publish anything as official process. It should not update the source of truth. It should not tell staff that a draft SOP is approved.
Access also matters. If internal communications are included, the tool should only read sources already visible to the person running the audit. In many cases, you can start with the shared SOP folder and project management task list only.
The final sign-off belongs with a manager or process owner. AI can suggest structure; it cannot know all the exceptions, client promises or commercial judgement behind a process.
What you need for a first version
A simple version needs three inputs: a folder of current SOP documents with last-modified dates, a list of recurring internal task types, and a short set of rules for prioritising risk.
Optional internal messages can be useful where processes are described informally, but they are not essential for a first pass.
The assistant then produces a markdown or PDF report, plus draft outlines for the highest-priority missing SOPs. No write access to your systems of record is needed.
A practical first step
Ask your team this question: if our most experienced person was unavailable next week, which recurring tasks would become difficult within 48 hours?
Those tasks are your first documentation candidates.
If you want help building a safe, practical process gap-finder for your own firm, book a short call with Stuart Cole Consulting. We can identify the right sources, guardrails and approval steps before you give any AI tool access to internal information.
