The AI readiness checklist for accounting firms: 14 questions to ask

By Kenny MacAulay | November 16, 2026
The AI readiness checklist for accounting firms: 14 questions to ask

An AI readiness checklist for an accounting firm is a firm-level audit, run once, that shows whether you can account for the AI already in the building. The fourteen questions below come in three groups: what is running, whether you could defend it, and what your clients and insurer can see. Each has a fail signal underneath. A question you cannot answer is information. One you answer badly is a decision waiting to be made.

What is an AI readiness checklist for accounting firms?

An AI readiness checklist is a short set of questions a firm answers about itself before, or while, it deploys AI. This AI readiness checklist comes from Before you deploy AI, a guide built with contributors from ACCA, HLB International, Apteus Group and a private equity backed accountancy group. The questions are the guide's own view.

Each question carries a fail signal: one line saying what a bad answer means. That matters more than the question. Most firms can answer "do we have an AI policy" with a yes. Far fewer can say who owns it, or when it was last updated. The fail signal is what turns a box-ticking exercise into something a partner can act on, and it is the part most generic checklists leave out.

The checklist works as an AI readiness assessment if you run it properly. John Toon, Head of Technology at HLB International, sets the order: start with an audit of what you have, including the subscriptions nobody declared. Then triage what is old, what could be replaced tomorrow and what has to wait for the market to catch up. Then keep the plan live, because something will be released that changes it.

Group one: What AI is running in your firm?

The first five questions establish what AI is in the firm, where client data goes and who else touches it. Most firms find this group the hardest, because the honest answer starts with an inventory nobody has finished.

  1. Can you list every tool in the firm with an AI feature, including those switched on inside software you already owned? Fail signal: if it is shorter than your software inventory, it is not finished.
  2. Which of those tools move client data outside the UK or EU, and where does it land? Fail signal: "The vendor handles it" is not an answer you can give a client.
  3. How long are you keeping client data, and does your AI reach all of it? Fail signal: everything you have retained is in scope the day you point a tool at it.
  4. Has anyone asked your providers whether your data trains their model? Fail signal: if nobody has asked, you do not know. Neither does your client.
  5. Who becomes the subprocessor for your client data? Fail signal: your own clients will eventually put this to you.

The second and fifth questions are easier for your own core systems if you know their data residency and audit logging. The third is where preparing accounting data for AI usually begins.

Group two: could you defend the work?

The middle six questions test whether AI-assisted work would survive a reviewer or an insurer asking how it was produced. This is where firms with good intentions and no records fall down, because a review that leaves no trace cannot be shown to have happened.

  1. For a piece of AI-assisted work, can you reconstruct the inputs, the prompt, the tool and its version, the output, and who accepted it and why? Fail signal: if not, you cannot explain it to a reviewer, an insurer or a client.
  2. Have you defined what human review means for each use case, or does one policy cover them all? Fail signal: a review nobody has specified is a rubber stamp with a job title.
  3. How quickly could you stop an AI tool that was doing the wrong thing? Fail signal: nearly 60% of European organisations cannot say.
  4. Are there escalation procedures for AI-related incidents? Fail signal: an incident with no route out of the team is absorbed quietly.
  5. Does every AI initiative have a named, documented owner? Fail signal: "IT" is not an owner.
  6. Which named department is accountable for reviewing and updating the AI policy? Fail signal: a policy nobody owns is a policy nobody updates.

The figure behind question eight comes from ISACA's March 2026 research across 681 European organisations. Question seven is easier for firms that already run maker-checker approvals on client work. Question six is the one most firms fail outright. Acting Office keeps a full audit trail of every user action on the client record, which gives reconstruction a record to start from.

Group three: what can your clients and insurer see?

The final questions cover what sits outside the firm, from the engagement letter to the training your people have had. These are the questions an insurer or a buyer is most likely to ask first, and the ones where the evidence either exists already or does not.

  1. Do your engagement letters disclose AI use and the extent of reliance on AI outputs? Fail signal: your insurer is forming a view now, not at renewal.
  2. Do you record explicit consent before AI-assisted recording of client meetings? Fail signal: consent you did not record is consent you cannot evidence.
  3. Have your people been trained, or are they teaching themselves? Fail signal: 41% of finance professionals globally are pursuing AI training in their own time.

The figure behind question fourteen comes from ACCA and Chartered Accountants Australia and New Zealand, in a 2026 survey of 1,600 finance professionals worldwide. Question twelve is the cheapest to fix on the AI readiness checklist: a paragraph in the engagement letter, applied consistently when the engagement sits on the same client record as the work.

How should you use the checklist?

Run the checklist once, at firm level, with someone who has the authority to ask every partner and provider. Record the answers, including the ones you cannot give. Then put four sentences in front of your people while the gaps are fixed, because they will do more than a policy nobody reads.

Tell them the firm is not banning AI, because a ban moves the problem onto personal phones. Tell them which tools are approved, and that the list will grow. Tell them client data does not go into anything off that list, because since January 2026 that has been a conduct matter under PCRT. And tell them who to ask, by name.

Then ask what they are already using. Most firms learn more in that conversation than in any audit.

George Llewellyn, Head of Technology at Apteus Group, would go further. If investment or a sale were three years out, he would run due diligence on the firm himself, starting by writing everything down. This checklist is a reasonable first page of that file, and Acting Office published the full guide so the rest of it has a structure too. Download the full guide for the reasoning and evidence behind every fail signal. It also covers the nine questions partners are asking about AI that sit behind these fourteen.

Download the AI readiness checklist

All fourteen questions and fail signals on one page, with space to record your answers.

FAQ

How many questions are in the AI readiness checklist?

The checklist from Before you deploy AI has fourteen questions in three groups. Five cover what AI is running in the firm, six cover whether the firm could defend AI-assisted work, and three cover what clients and insurers can see. Each question carries its own fail signal.

What is a fail signal?

A fail signal is a one-line statement under each checklist question saying what a bad answer means. For example, if your list of AI tools is shorter than your software inventory, it is not finished. It turns a yes or no question into a test a firm can actually fail.

How often should a firm run an AI readiness checklist?

The checklist is designed as a firm-level audit, run once, with the resulting plan kept live. Toon's advice is to keep the plan under review, because new releases will change it. Revisit the inventory whenever a product renews or adds an AI.

Note: Please note that the content of the above blog and the aforementioned information are solely for the purpose of awareness and are informative in nature. The content is designed with intent to ease the understanding while preserving the essence and importance of the compliance rules and shall not be considered as an ultimate replication of the rules. ActingOffice does not own any responsibility whatsoever for any unpleasant event that may arise due to the misinterpretation of a specific part or whole of the information.