Firm-wide AI rollout for an accounting firm: the calendar

A firm-wide AI rollout at an accounting firm succeeds or fails on two decisions: when in the firm's year the work happens, and whether each department ends up with a few workflows it depends on. Build and train between the end of spring busy season and late fall, freeze changes before January, and review what broke in May.

The rest of this article lays that plan out for a managing partner, COO, or firm administrator who has already bought licenses or run a pilot and now has to make it hold across departments.

Why do licenses and a training session not produce adoption?

A license gives a staff accountant a blank prompt box, and a lunch session shows them someone else's examples. Neither one changes the organizer, the workpaper, or the proposal that is due on Thursday, so on Thursday most people do the task the way they did it last year.

What changes the work is a small number of workflows that a department already relies on, each one built with the person who owns that task, each with its written instructions and templates stored in the firm's shared workspace. The person who helped build it is using it by the time it is finished, and the next person learns it from them on a real file.

Start with what staff are already doing

The rollout does not begin from zero. In a 2026 Thomson Reuters survey of 538 professionals at tax and audit firms in 40 countries, about a third of them in the United States, 81% said they use AI tools at least several times a week, and 35% said they use AI tools their firm has not authorized. A separate Thomson Reuters Institute survey of more than 1,500 professional services respondents in 27 countries found that only 18% knew their organization was tracking the return on its AI tools in some manner.

Those two findings set the first two jobs. The first is to bring the use that already exists inside the firm's own workspace and rules, which starts with asking each department what people are using and for which tasks, without penalty for an honest answer. The second is to measure something, which we come to below. If the workspace and the client data rule are not in place yet, our guide to setting up a CPA firm to use AI covers that step.

Lay the rollout against the firm's calendar

Our view, from building inside firms like these, is that the calendar decides more than the technology does. A tax-heavy firm has a limited stretch of the year when people can absorb a new way of working, and a workflow introduced in January competes with the deadline and loses. Adjust the months below to your own mix of tax, audit, and advisory work and to your extension deadlines.

When	What happens	Who is involved
May	Collect what staff used during the season, what they wanted, and where the existing process hurt most. Pick the first department and its first three workflows.	Rollout lead, partner sponsor, department heads
June through August	Build each workflow against last season's closed files, with the person who owns the task in the working sessions.	Rollout lead, the workflow's owner, the builder
September through November	Use the workflows on live work around the extension deadlines. Take the "after" measurements. Train the rest of the department on their own files.	The department, its champion
December	Freeze. Write down the version of each workflow that will run through busy season and update the policy and the quality management documentation to match.	Rollout lead, quality management lead
January through April	Run what was frozen. Change nothing except to fix an error. Keep a running list of problems and requests.	Everyone
May	Review the list, fix what broke, and start the next department.	Rollout lead, partner sponsor

Closed files are the right training ground for a reason beyond scheduling. The firm already knows the correct answer for last year's return or last year's audit file, so a reviewer can compare the AI's draft with what the firm delivered and correct the instructions until the draft is an acceptable starting point.

Pick the first department and the first three workflows

Choose the department by its people before its workload. The best first department has a partner who will use the result where staff can see it and a senior or manager who is curious enough to sit in the working sessions.

Then choose workflows that recur, start from a clear source document, and already have a reviewer. Three is a sensible number for a first round because it is enough for the department to feel a difference and few enough to finish before fall. Typical candidates include a client-specific organizer built from the prior-year return, a workpaper tie-out to the trial balance, and a proposal drafted from past proposals. Our article on where AI fits in tax, audit, and client advisory work goes through the options by service line.

Build with the person who owns the work, and train inside the build

Each workflow gets a short working session every week with the person who does that task today. They bring last season's files, the builder brings a working draft, and the owner corrects it: this schedule is missing, that note is worded the way a first-year would word it, the partner always asks for this comparison. Those corrections become the written instructions.

This is also the training. The owner learns how to direct the tool on documents they know cold, which in our experience sticks better than a general course, and they finish the build already using it. We describe the method in more detail in how to run AI training on your team's own work. The instructions, templates, and a one-page description of each workflow stay in the firm's shared workspace so the firm owns them and a new hire can find them.

Who owns it?

A rollout needs three roles, and in our view the first is the one that is easiest to leave underfunded.

• A named lead with time allocated. One senior manager, director, or partner whose hours for this are set aside in the budget and in their chargeable-hour target. A lead who is expected to do this on top of a full client load has no hours left for it once busy season starts.
• A partner sponsor who uses it. A partner who opens the workflow in a team meeting and reviews its output in the normal course does more for adoption than a steering committee does.
• A champion in each department. Usually the person who owned the first workflow. They answer the day-to-day questions and keep the list of what needs fixing.

A July 2026 Journal of Accountancy article by Nicole L. Graham of Aon suggests firms consider designating an AI lead or committee to review proposed use cases, approve tools, revise and implement the policies, monitor employee use, and address incident reports. The named lead is the natural holder of those duties, with a committee added only if the firm's size calls for one.

Where does it sit in the quality management system and the usage policy?

Firms that perform engagements under the auditing, attestation, or accounting and review services standards were required to establish a system of quality management by December 15, 2025, the effective date of the AICPA's new quality management standards. The same Journal of Accountancy report says a firm must evaluate that system within one year of implementing it or by December 15, 2026, and annually after that. The resources component of the standards now includes requirements for technological and intellectual resources in addition to human resources, and that article describes technological resources as the IT applications a firm uses to support the quality management system and engagement performance. An AI workflow used on attest work plausibly falls under that heading, so ask the person who runs your quality management system how each workflow should be recorded in the risk assessment and what review responds to it. The December freeze is a convenient point to bring that documentation up to date, and the annual evaluation is a convenient point to review it.

The usage policy carries the day-to-day rules. The Journal of Accountancy article on AI policies suggests an approved tool list, human review and approval under the firm's existing supervision policies, documentation in the client file of the prompts used, how outputs were verified, and who reviewed them, and integrating AI review expectations into the quality control system instead of running them as a separate, informal process. CNA, writing for the AICPA Member Insurance Programs, tells firms to draft a firmwide policy, identify the specific tasks for which staff may use generative AI, prohibit sharing confidential client and firm information with these tools, train personnel on the policy, and monitor usage.

Where your firm sets its client data rule is a decision for the partners with counsel and the firm's carrier, and nothing here is legal advice. Client instructions belong in the policy too: in the Thomson Reuters Institute survey, 40% of firm respondents said they had received orders from various clients both to use AI on matters and not to use it. Our guide to writing an AI acceptable use policy for a professional firm covers the drafting.

What should we measure before and after?

Measure one thing per workflow that a partner already tracks or would recognize, and take the first reading before the build starts. Login counts and prompt volumes do not tell a partner whether returns are moving faster or review is getting lighter. Examples of measures that do:

• Days from client documents received to return in review
• Hours to a first draft of a proposal
• Review notes per return or per workpaper section at first review
• Hours from trial balance received to tie-out complete

Pull the "before" figure from last season's time and workflow records where they exist, and take the "after" figure on live work in the fall. Report both to the partners in the same units, including the cases where the number did not move, because that is how the firm learns which workflows to keep.

What should we freeze before busy season and collect after it?

In December, fix the version of each workflow that will run through April. Record its instructions, its templates, who reviews its output, and who to call when it produces something wrong. New ideas go on a list and wait.

During the season the champion in each department keeps that list. Each entry should say which workflow, what the input was, what went wrong, and what the reviewer did about it. In May the lead and the sponsor read the list with the department and decide what to fix, what to retire, and what to build next.

Extending to the next department without starting over

The second department starts with more than the first one had. The workspace, the policy, the quality management documentation, and the measurement habit all exist, and so does a set of written instructions that show what a finished workflow looks like at this firm. What does not carry over is the content: audit's documents and review points differ from tax's, and the workflow has to be built with the person in that department who owns the task.

Keep the same shape each time: a named owner for each workflow, closed files first, a before-and-after measure, and a freeze before the department's own deadline. If you would like a senior person to build the first round alongside your team, you can get in touch.

Common questions

How long does a firm-wide AI rollout take?

Plan on more than one annual cycle. A named set of working workflows for one department is roughly eight to twelve weeks of build time, which is how our own engagements are shaped, and that work has to fit between busy seasons. A reasonable plan is one department in the first year and the others in the off-season windows that follow.

Should every employee get an AI license on day one?

Give everyone access to the firm's approved workspace early, because in a 2026 Thomson Reuters survey 35% of tax and audit firm professionals said they use AI tools their firm has not authorized, and the likely alternative to a firm account is a personal one. Access is the starting condition, though, and the rollout itself proceeds one department and one workflow at a time.

What is the best time of year for an accounting firm to roll out AI?

For a tax-heavy firm, the months between the end of spring busy season and late fall are when people have room to learn a new way of working. Use that window to build and train on last season's closed files, and stop making changes before January.

How do we handle partners who will not use it?

Do not start with them. Start with a department whose partner will use the workflow in view of the staff, measure the result, and show the other partners that number in their own terms. The review and documentation rules in the firm's policy still apply to every partner whose staff use the tools, whether or not the partner does.

Do we need an AI committee?

You need a named lead with hours set aside and a partner sponsor. A July 2026 Journal of Accountancy article by an Aon risk consultant suggests designating an AI lead or committee to approve tools, maintain the policy, and monitor use, so either form fits that suggestion, and in our view a lead with allocated time moves faster than a committee without it.