Start with the people who will review an agent's work before it reaches a client: managers, seniors and the bookkeepers who approve categorisations. Test them on core ideas set in accountancy and read the results by concept. The report decides which agent goes first, who trains on what, and whether the firm is ready before the next filing deadline.
What is the assessment for, and who owns it?
A readiness assessment measures whether a team understands what a governed agent will put in front of it: depth, held actions, verdicts, provenance and the record. It is a diagnostic, not a certificate. In an accountancy firm, its job is to stop the firm putting an agent in front of reviewers who will wave its drafts through because they look like the work of a capable junior.
The practice operations lead usually owns it, with a sponsoring partner who will act on the result. The partner matters: the report often recommends a smaller, slower start than the firm hoped for. For the general idea, see what a readiness assessment is.
Who in the firm should take it?
Pick one service line first, not the whole firm. Bookkeeping and accounts production is the usual candidate, because agents there categorise bank transactions, reconcile accounts and draft year-end adjustments, all of which a person reviews.
The cohort should be everyone who will see the agent's output in that line: bookkeepers, seniors, the managers who sign off files, and one or two client service staff who answer client queries. Twenty to 25 people is a good size. Partners take it separately, later, so their presence does not change how juniors answer. The question of who in an accountancy firm needs training first is answered partly by these results.
Tax and audit teams can follow. Audit is a special case: independence rules shape what any tool may do on an audit file, so assess audit staff only once the firm has decided what agents will and will not touch there.
How to run it, step by step
- Person (operations lead): picks the service line and the cohort, and books the sitting away from a filing deadline.
- Person (operations lead): asks for the test set in accountancy, so questions are about held categorisations, client documents and adjustments, not generic examples.
- Person (each team member): sits the test alone, in one sitting, with no preparation. A prepared team hides its gaps.
- Person (operations lead): reads the report by concept, not by average.
- Person (sponsoring partner and operations lead): choose the first agent that matches the team's strengths, and the depth it starts at.
- Person (operations lead): sets training on the weakest concept first, for the people who will approve.
- Agent, depth 1 then 2: runs in shadow on past client files, drafting categorisations the team compares with what was actually posted.
- Person (team): reassesses on the weak concept before the agent drafts on live files.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Choose cohort | Nothing yet | None | Service line and timing |
| Sit the test | Nothing yet | None | Each person's own answers |
| Read the report | Nothing yet | None | Interpreting gaps |
| Pick first agent | Is named as the candidate | None | The choice and its depth |
| Targeted training | Runs in practice labs | Practice | Learning the weak concept |
| Shadow | Drafts on past files | 1 then 2 | Comparing with real postings |
| Reassess | Nothing new | 2 | Deciding on live drafts |
What does the test reveal in an accountancy firm?
Provenance is the usual gap. Accountants are trained to trust bank data and question client explanations, which maps neatly onto sealed and declared facts. Where teams stumble is the external class: an emailed receipt, a supplier statement forwarded by the client, a PDF invoice. Many treat these as evidence on a par with the bank feed. An agent will label them correctly; the question is whether reviewers read the label.
The decision chain is the second gap. Reviewers used to checking a junior's work look for errors. With an agent, they also need to read why something was held: which of the five bounds decided. A team that cannot tell a hold on the firm's own materiality limit from a hold on a weak source will clear both the same way.
Depth usually scores well, because "drafts but does not file" matches how practices already work.
What does the report change?
It changes the first agent. A team strong on verdicts but weak on provenance should start with an agent that drafts replies to routine client queries, not one that categorises transactions from mixed sources.
It changes who trains first. If managers score well and seniors do not, seniors are the bottleneck, because they review most drafts.
It changes the date. A weak result two months before a deadline is a reason to move go-live to after the deadline, not to compress training.
Which controls does the assessment leave evidence for?
Who took it. Evidence: the cohort list and individual results, held privately.
What was found. Evidence: the report by concept, with the gaps named.
What was decided. Evidence: the partner's recorded choice of first agent and starting depth, with reasons.
That the gap closed. Evidence: the reassessment result on the weak concept.
What should the firm measure?
- Scores by concept before and after targeted training.
- Agreement between shadow drafts and real postings during the shadow period.
- How many reviewers in the line hold a current reviewer certificate before live drafting.
Where does it go wrong in accountancy firms?
Testing partners with juniors. Juniors answer cautiously and the result flatters the team.
Running it in January. Or any month before a deadline. People rush it and the scores mean little.
Ignoring a weak result. The firm reads the report and goes live anyway. Then the first wrong categorisation is approved by someone the report said would approve it.
Law firms face a similar shape of problem with different documents; compare the law firm readiness assessment and the property management firm version. Partners need their own preparation, covered in what AI agent training an accountancy firm board needs. The topic guide is training teams for AI agents.
Where the Academy fits
Pryme Intelligence Academy runs a free readiness assessment for teams of up to 25. The team takes the Foundations test set in its own industry, and the report shows scores by concept, the gaps, and the first agent to start with. Reviewers who then approve drafts can go on to the Operator track.
Questions people ask
Should results be shared with the team?
Share the team's results by concept. Keep individual results for one-to-one coaching, or people will answer to protect themselves next time.
Do bookkeepers need to take it if they do not sign files?
Yes, if they will approve any agent draft, including categorisations. Approving is the moment the test is about.
How often should the firm reassess?
Before each new agent, before raising an agent's depth, and after training on a weak concept.