Training your team on AI in 30 days: a week-by-week plan
A 30-day, week-by-week plan for training an accounting team on AI, starting with safe use and the firm’s policy before any prompting. It covers champions, three shared firm tasks, role tracks, training reviewers and measuring use, with the whole plan as a table to copy.
Trent McLaren · Publisher, The Firm
9 November 2026 · 9 min read
Part of our AI in accounting coverage. See the full AI for accountants guide →
Learning check · before you start
What do you already have in place?
A few quick questions, so you can skip the steps you have already done. Each one has a line explaining what it means. Not sure, or never heard of it? Choose “Not yet / not sure”: that is where most firms start, and the playbook walks you through it.
This playbook is for the owner of a firm of five to twenty people who has bought AI logins, or is about to, and wants the team using them well a month from now. You finish with a 30-day rollout plan as a table, one or two named champions, a short list of firm tasks everyone can do with AI, and a way to tell whether it worked. It costs about six hours of team time over four weeks, plus an hour a week from each champion.
The order matters more than the content. Most firms start with prompts because prompts are fun. This plan starts with safe use and the firm’s policy, because a team that is unsure what it is allowed to paste either pastes everything or uses nothing.
Teaching a team to use AI productively and securely is among the most-registered themes at The Firm’s AI in Practice Summit this month: a panel with Beau Gaudron (Growthwise), Gus McLennan (AI Dojo), James Rose (Content Snare) and Travis Bacon (TaxDigital) is on the agenda for 11 November. Notes will be at the summit notes page afterwards.
Step 1: Get ready before day one
Your task before training starts, get a one-page AI policy signed, set up firm-managed AI accounts for the first group, and name one or two champions with an hour a week set aside.
You said you already have this covered. Skim this step, or skip to the next.
Three things need to exist before anyone sits a session, and none of them is a prompt.
- A written policyOne page naming the approved tools, what client data may go in, and who reviews output. If you have none, our one-page AI policy guide has wording you can adapt in an afternoon.
- Firm-managed accountsBusiness-plan seats for whoever is in the first cohort, set up by the firm, not personal subscriptions expensed later.
- A named owner and one or two championsThe owner answers for AI in the firm. Champions are the people others ask at their desk. Pick the curious person, not the most senior one, and give them an hour a week on the timesheet for the month.
Same tools, same Claude. The difference is whether they wrote down what they were doing and gave the guardrails for their teams.
Twyla was describing Karbon’s State of AI in Accounting report, which she said found that firms with an AI strategy or policy saved more time with AI than firms without one.
Done when: the policy is signed, the seats exist, and the champions know it is part of their job this month.
Learning check
What would you do? Pick one; every answer is explained.
Step 2: Week one is safe use and the policy
Your task run a 45-minute safe-use session using three real examples from your own files, then give everyone one small, safe task to finish by Friday.
You said you already have this covered. Skim this step, or skip to the next.
The first session is 45 minutes and has no clever prompting in it. Cover four things.
- The account rule. Client work happens in the firm’s account, never a personal one. Show the team the difference on screen.
- What goes in. Walk through three real examples from your own files: one fine to paste, one that needs the client name taken out, one that never goes in (a TFN, a bank account number, a login).
- Review. Every output that reaches a client is checked by a qualified person. In Australia, the Tax Practitioners Board’s July 2026 guidance on AI, TPB(GS) 55/2026, says practitioners should verify and review AI-generated content and document those steps.
- Mistakes. Who to tell if something goes in the wrong place, and the promise that reporting it is never punished.
Then give everyone one small, safe task to finish by Friday: rewrite an internal email, summarise a public ATO or IRS page, or draft a checklist from an SOP. The aim is a first success, not a breakthrough.
Done when: everyone has read the policy, said so in writing, and completed one task in the firm account.
Learning check
What would you do? Pick one; every answer is explained.
Step 3: Week two is three firm tasks, not prompting tips
Your task pick three tasks the firm does every week, have a champion write and save one tested prompt (the instructions you give the AI) for each in a shared place, and run a 30-minute session on live jobs.
You said you already have this covered. Skim this step, or skip to the next.
Generic prompt training fades because nobody remembers a technique without a job to attach it to. Instead, the owner and champions pick three tasks the firm does every week and build the team’s AI habit around them. Good candidates: drafting the client request email at the start of a job, turning meeting notes into action items, and producing a first-pass variance comment on a set of accounts for a reviewer to correct.
For each task, the champion writes one tested prompt and saves it where everyone can reach it: a shared project, a saved instruction or a skill, depending on your tool. Then a 30-minute session where each person runs all three on a real job and compares notes.
So if you’ve ever prompted something more than once, stop and ask: should this be a skill?
Link the tasks to the system you already run on. If the work lives in Xero Practice Manager, FYI or Karbon, the prompt should start from the job or the email thread there, so AI becomes part of the workflow rather than a tab people forget to open. If you are still choosing what to automate first, our guide to the first problem worth handing to AI helps narrow it.
Done when: three tested prompts are saved in a shared place, and each team member has used all three on live work.
Learning check
What would you do? Pick one; every answer is explained.
Step 4: Week three is role tracks and the review habit
Your task run one short session per role, such as bookkeepers, accountants and admin, that adds one task for that role, and have every reviewer check at least one AI-assisted piece of work against its source.
You said you already have this covered. Skim this step, or skip to the next.
By now the team splits. Bookkeepers, accountants and admin staff want different things, and the champion’s job shifts to coaching. Run one short session per role, each adding one task specific to that role.
This is also the week to train reviewers, which most plans forget. A senior who reviews AI-assisted work needs to know what tends to go wrong: invented references, figures that do not tie back to the ledger, and confident wording on a point the source never made. Our piece on how juniors learn when they stop asking the partner covers the other side of that habit.
Done when: each role has one added task, and every reviewer has checked at least one AI-assisted piece of work against its source.
Learning check
What would you do? Pick one; every answer is explained.
Step 5: Week four is measure, fix and decide
Your task collect who is using the AI tool, how many of the three shared firm tasks each person ran last week and how many near misses were reported, then hold a 30-minute review.
You said you already have this covered. Skim this step, or skip to the next.
Measure use plainly. Most business plans give admins some view of who is active; check what yours shows rather than assuming. Pair it with two numbers you collect yourself: how many of the three firm tasks each person ran in the past week, and how many near misses were reported. A near miss reported is a good sign, because it means people trust the rule.
Then hold a 30-minute review. Which task saved time? Which one nobody used, and why? What did people try to paste that the policy did not cover? Update the policy, retire the dead task, add one new one, and decide whether to roll seats out to the next group.
Logins go unused for reasons that show up in that review: nobody owns the tool, the first task was not part of real work, or people were never sure what was allowed. Our piece on why teams stop using their AI logins sets those against the National AI Centre’s adoption guidance, which treats training as ongoing work rather than a one-off.
Done when: you have the usage numbers, a revised policy, and a decision on the next cohort.
The 30-day AI training plan. Owner: [name]. Champions: [names]. Start date: [date].
| Days | Focus | Session | Homework | Check |
|---|---|---|---|---|
| Before day 1 | Policy, seats, champions | Owner and champions, 60 min | Champions try the tool on internal work | Policy signed, seats live |
| Days 1 to 7 | Safe use and the policy | Whole team, 45 min | One safe internal task in the firm account | Everyone confirmed the policy in writing |
| Days 8 to 14 | Three firm tasks | Whole team, 30 min | Run all three tasks on live jobs | Three tested prompts saved and shared |
| Days 15 to 21 | Role tracks and review | Per role, 30 min each | One role task; reviewers check AI work against source | Each role has an added task |
| Days 22 to 30 | Measure, fix, decide | Whole team, 30 min | Tell the champion what you tried that failed | Usage numbers, near misses, policy revised |
| Every week | Champion desk time | Champion, 60 min open | Bring one stuck task | Questions logged and added to the FAQ |
Rules for the month: client work in firm accounts only. Nothing restricted (TFNs, bank details, logins) ever goes in. Every client-facing output is reviewed. Report mistakes the same day; reporting is never punished.
Learning check
What would you do? Pick one; every answer is explained.
Free to join
Join The Firm free to read the rest
Members get every playbook in full, plus free templates, downloads and tools across the site. Join more than 35,000 accounting professionals. All it takes is your email.
Where we could be wrong
The assumption most likely to be wrong is that three shared tasks are enough to form a habit in a month. In a firm deep in tax season, three may be two too many, and the plan should stretch to six weeks rather than skip the safe-use week. We also have not measured this plan across a set of firms: it is built from the TPB and National AI Centre guidance and from what practitioners described at the June summit, not from outcome data. And we could not confirm what usage reporting each vendor’s business plan provides as at November 2026, which is why Step 5 tells you to check yours.
This is general guidance on training, not legal advice. Your obligations on client data come from your professional body, the TPB (for Australian tax and BAS agents) and privacy law. For the wider picture of tools, see our AI for accounting firms hub.
Frequently asked questions
Why not start with prompt engineering?
Because a team that is unsure what it may paste will either avoid the tool or paste everything. Safe use takes one session and removes the hesitation that stops people trying. Prompt skill grows faster once people are using the tool on real jobs every day.
Who should be the AI champion?
The person others already ask about software, whatever their title. What matters is protected time: an hour a week on the timesheet, and the owner backing their answers when the policy is unclear.
Should partners do the training too?
Yes, and in the same sessions. Partners are usually the reviewers, and a reviewer who has never used the tool cannot tell a sound AI-assisted draft from a fluent wrong one.
What if usage drops after the first month?
Expect some drop as novelty wears off. Ask which of the three tasks people stopped running and why. Usually the task was not tied to work they do every week, or the saved prompt broke when a template changed. Replace it rather than re-running the training.
Do we need a paid training course?
Not to start. This plan uses your own policy, your own jobs and your own champions. External training earns its place later, for specific skills such as building shared projects or connecting tools to your practice data, once the team knows what it wants to do.
Your one-page summary
Your answers, choices and decisions, on one page
Sponsored
AI TrainingTeamAI AdoptionAI PolicyPlaybook
Get the editable 30-day AI training tracker Download the 30-day AI training tracker
The Excel workbook (.xlsx), ready to fill in. Your profile unlocks every template on The Firm.
Or build it with AI
Open Claude or ChatGPT with the prompt filled in. It builds the workbook for your firm straight away, then asks a few questions to tailor it. No client data needed.
Works without client data. Don't paste client names, tax file numbers or ledgers into an AI tool your firm hasn't approved for them. AI data safety playbook.
