To document your firm's SOPs with AI, record a staff member talking through a task, feed the transcript to an AI tool, and ask it to draft a step-by-step standard operating procedure you then correct. AI turns a 20-minute conversation into a usable first draft in minutes — the part that used to stall because nobody had time to write it down. For an Australian accounting, legal, real estate or conveyancing firm, this is the step that makes everything else possible: you cannot automate a process you have never written down.
This article is for practice managers and partners who know their best processes live in one or two people's heads, and who want them captured properly before anyone talks about automation.
Why document SOPs before you automate anything
Automation amplifies whatever process you point it at. If the process is messy, undocumented or exists only in a senior staffer's memory, you will automate the mess — or worse, you will build an automation that nobody can maintain once that person leaves.
A standard operating procedure (SOP) is simply the agreed, repeatable way your firm does a task: new client onboarding, a trust account receipt, a BAS lodgement, a settlement booking. Writing it down does three things at once. It surfaces the steps that are actually inconsistent between staff. It gives you a baseline to measure improvement against. And it becomes the specification a developer or a tool like Zapier or n8n builds from.
We cover sequencing in detail in our guide to deciding which tasks to automate first, and the build process in the AI automation implementation guide. SOPs are the connective tissue between the two. Skip them and you will keep re-explaining the same process to every new hire and every new tool.
How to turn team knowledge into documented processes
The hardest part of SOPs has never been the writing — it's extracting what's in people's heads. Experienced staff do tasks on autopilot and skip the steps they consider obvious. AI changes the economics here because it is a patient, tireless interviewer and transcriber.
Here is the sequence that works:
- Pick one task with a clear start and end. "Onboard a new conveyancing client" is good. "Manage client relationships" is not — it's too broad to capture in one pass.
- Record the person doing or describing it. Use a meeting notes tool to capture a screen-share walkthrough or a verbal explanation. We've written separately about using AI meeting notes in professional services — the same transcription turns a casual walkthrough into raw material.
- Let the AI interview the gaps. Paste the transcript into your AI tool and ask it to list every point where a step is ambiguous or a decision was made without explanation. This is where the hidden knowledge lives.
- Go back and answer those questions. Ten minutes with the staff member filling gaps produces a far better SOP than an hour of them trying to write from scratch.
The shift is subtle but important: the expert talks, the AI writes. People who would never volunteer to "document the process" will happily spend 20 minutes explaining how they do their job.
How to write standard operating procedures with AI
Once you have a transcript and the gaps filled, drafting is fast. The skill is in the prompt and the correction, not the typing.
Start with a prompt like this, pasted above your transcript:
"You are helping a Brisbane accounting firm document an internal procedure. Below is a transcript of a staff member explaining how they process a client's monthly bookkeeping. Turn it into a standard operating procedure with: a one-line purpose, the trigger that starts it, numbered steps in plain English, the software used at each step, who is responsible, and a checklist of what 'done' looks like. Flag anything that sounds inconsistent or risky."
The first draft will be roughly 80% right and 20% invented or vague. That 20% is the job. Read it against reality and correct it:
- It assumed a step. AI fills gaps with plausible-sounding defaults. If it wrote "reconcile in Xero" but you use a specific review step first, say so and regenerate.
- It merged two steps that should be separate. Tell it to split them — separate steps are what automation hooks onto later.
- It used generic language. Replace "relevant documents" with the actual document names your firm uses. SOPs fail when they're too abstract to follow.
Keep the format consistent across every SOP. A good template has: purpose, trigger, inputs, numbered steps, responsible person, tools, and a definition of done. When every procedure follows the same shape, staff can read any of them without relearning the layout.
A note on tooling: use your firm's existing licensed AI where you can. If you're on Microsoft 365, Copilot works inside your tenant; on Google Workspace, Gemini does the same. For SOP drafting specifically, you can often work faster in a dedicated workspace — our guide to building a custom GPT or Claude Project for your firm shows how to give the AI your template once so every SOP comes out in-house style.
AI process mapping for professional services
A written SOP tells you the steps. A process map shows you how the work actually flows — the handoffs, the decision points, the places work waits. AI can draft these too, and for professional services firms the handoffs are usually where time leaks.
Take the SOP you just wrote and ask the AI: "Convert this procedure into a process map. For each step, show who does it, what triggers the next step, and any decision point where the path can branch. Mark every handoff between people or systems."
What you get back is a structured description you can read in plain English or paste into a diagramming tool. The value isn't the pretty diagram — it's that process mapping forces you to name the decision points. "If the contract has special conditions, escalate to the principal" is a branch. Branches are exactly what an automation has to handle, and the ones you don't map are the ones that break later.
For firms weighing whether a process even warrants automation versus another hire, mapping it first makes the decision concrete. Our article on what admin work to automate before hiring is a useful companion here — a mapped process tells you which steps are genuinely repeatable and which still need a person's judgement.
How to keep SOPs up to date automatically
Most SOP projects die the same way: a burst of documentation, then the procedures drift out of date and everyone goes back to asking the person who knows. A "living SOP" is one that gets updated as a matter of routine, not as a special project.
AI makes maintenance light enough to actually sustain:
- Review on a trigger, not a calendar. Tie SOP review to real events — a software change, a staff departure, a near-miss. When your practice management system updates, that's the prompt to re-check any SOP that references it.
- Let AI compare old and new. When a process changes, record a fresh walkthrough and ask the AI to compare it against the existing SOP and list only what changed. You update the diff, not the whole document.
- Keep SOPs where people work. A procedure buried in a shared drive nobody opens is dead. Keep them in the tool staff already use daily, and link them from the task they describe.
- Assign an owner per SOP. One named person is responsible for each procedure being current. AI drafts; a human owns.
"Automatically" has limits worth being honest about. AI can flag when a SOP is likely stale and draft the update, but it cannot know your process changed unless something tells it. The reliable pattern is a human trigger plus AI doing the tedious rewriting — not a system that silently keeps itself current.
What to keep human, and the privacy line
Some things should never be delegated to the draft. Anything involving professional judgement, regulatory compliance, trust accounting rules or client-specific legal advice needs a qualified person to own the content. AI is excellent at structure and terrible at knowing which step your professional obligations make non-negotiable.
On data: documenting SOPs should not require real client information at all. Write procedures with placeholders — "the client's TFN", not an actual number; "the matter reference", not a real file. If a walkthrough transcript happens to contain client details, redact them before pasting anything into an AI tool.
The OAIC's general guidance is clear that the Privacy Act applies the moment personal information goes into an AI system, and that publicly available chatbots should not be fed personal information. Use enterprise or API tiers where data isn't used for training, prefer the AI already licensed inside your Microsoft or Google tenant, and keep a short record of what data each workflow touches. Our overview of AI and client data under the Australian Privacy Act covers this properly. For SOP documentation specifically, the easy win is simply not putting client data in at all — procedures describe the steps, not the cases.
When documenting SOPs with AI is the wrong move
Don't document a process you're about to change. If a task is obviously broken, fix the process with your team first, then capture the improved version. Writing an SOP for a bad process just entrenches it.
And don't try to document everything at once. Pick the three or four processes that are most painful when the key person is away, or most error-prone, and start there. A complete library of SOPs nobody uses is worse than four good ones your team actually follows.
Documenting your SOPs is the unglamorous step that makes every automation afterwards cheaper, safer and easier to hand over. Get the procedures written and owned first, and the rest of your automation work has something solid to build on. When you're ready for that next step, our free guide, 10 AI Workflows to Save 10+ Hours a Week, shows where documented processes most often turn into real time savings for Australian firms.
Common questions
How do I write standard operating procedures with AI?
Record a staff member explaining or demonstrating the task, then paste the transcript into your AI tool with a prompt asking for a purpose, trigger, numbered steps, responsible person, tools used and a definition of done. The first draft will be roughly 80% correct — your job is correcting the invented or vague 20% against how the task actually works.
How can AI turn team knowledge into documented processes?
Experienced staff rarely want to sit and write procedures, but they will happily talk through a task. AI removes the writing bottleneck: transcribe a short walkthrough, let the AI identify where steps are ambiguous, then spend ten minutes filling those gaps. The expert talks, the AI drafts, and knowledge that lived in one person's head becomes a shared document.
Can AI keep our SOPs up to date automatically?
Partly. AI cannot detect that your process changed on its own, but it makes maintenance light enough to sustain. Tie reviews to real triggers — a software update, a staff departure, a near-miss — then record a fresh walkthrough and ask the AI to list only what changed. You update the difference rather than rewriting the whole document, with one named owner per SOP.
Is it safe to put our procedures into an AI tool?
Documenting SOPs should not need real client data — write with placeholders like 'the client's TFN' rather than actual details, and redact any identifiers in walkthrough transcripts. The OAIC's general guidance is that the Privacy Act applies once personal information enters an AI system. Use enterprise or API tiers, or the AI already licensed inside your Microsoft or Google tenant, and keep a record of what each workflow touches.
Why document SOPs before automating?
Automation amplifies whatever process you point it at, so automating an undocumented or messy process just locks in the mess. A written SOP surfaces inconsistencies between staff, gives you a baseline to improve, and becomes the specification a developer or tool builds from. It also means the automation survives when the person who 'just knows how it works' leaves the firm.
Want these ideas working in your firm? We build controlled AI workflows for Australian professional services firms — starting with a free automation audit.