Clearing AI mapping proposals before signing
Why a reviewer should clear the AI mapping proposals in a draft before the partner signs, even though nothing in the Studio forces it, and a rule to adopt.
By Achekam · · 4 min read · For Accountants & audit firms in South Africa, Nigeria, Kenya
Some controls are built into software and some are left to people. Clearing the AI mapping proposals in a draft set of statements is one of the second kind. Achekam lists the proposals, shows where each account has been placed and lets a person accept, move or set each one aside. It does not stop a partner signing while some are still open.
That design puts the responsibility where it belongs, with the firm, but only if the firm actually picks it up. This article explains how proposals get into a draft, what each one shows, how to work the list, and a rule worth adding to the firm's own policies.
How AI proposals enter the draft
When a trial balance is loaded, Achekam suggests where each line belongs, drawing on the file's own mapping column, earlier mappings, AI or keyword rules. Without an AI key, keyword rules are used and labelled as such. A suggestion is not an approved mapping until a person approves it.
The Studio then treats AI suggestions in a particular way. When it builds a draft, those with a confidence of 0.5 or more are placed straight into the draft, so the draft is complete enough to read, and they are listed as proposals a person can accept, move or set aside.
The consequence is easy to miss. A draft can look finished, with every account on a line and every total agreeing, while some of those placements are still nothing more than proposals.
What a proposal shows, including its confidence
Each proposal shows the account, the line it has been placed on and the confidence attached to the suggestion. Confidence describes how sure the suggestion is; it is not a check anyone has made. A proposal near the bottom of the accepted range deserves more scrutiny than one near the top, but neither has been looked at by a person until someone acts on it.
Use confidence to order the work, not to skip it. Start with the lowest-confidence proposals and with those on lines that matter most to readers: revenue, loans, related-party balances and provisions. A high-confidence proposal on an account with a vague name, such as "Sundry" or "Clearing", still deserves a look at the ledger before it is accepted.
Accept, move or set aside: working the list
For each proposal, the reviewer has three choices:
- Accept it when the account clearly belongs on the proposed line. Check the account description and, if unsure, its contents in the client's ledger.
- Move it when it belongs elsewhere. The workbench under Canvas > "Mapping" shows each account's line and note and lets staff move accounts by hand; the menu entry "Trial Balance & Source Mapping" opens the Source Room instead, so do not look for it there.
- Set it aside when the account needs more information before it can be placed, and follow up with the client. A set-aside proposal is an open question, not an answer.
A worked example in words: an account called "Advances" is proposed into trade receivables with modest confidence. The reviewer looks at the ledger and finds the balance is salary advances to staff. It is moved to other receivables, and because one of the staff members is a director's spouse, it is flagged for the related-party note.
How suggestions are approved more generally is covered in TB mapping suggestions and approvals, and moves by hand in moving an account to another statement line.
The safeguards are people, not a block
Nothing in the Studio stops signing while AI proposals are open. The safeguards are the review stages and the partner's signature. Each set names a preparer, a reviewer and an engagement partner; only a user with a partner role can sign, and the person recorded as preparer or reviewer cannot issue the set.
The same pattern runs through the AI features in the statements. Where an AI key is set up, AI writes narrative text that is checked in code and labelled "AI draft - review before issue", with no separate accept step per paragraph. Staff can run an automated review of a draft that lists findings for a person to accept, resolve or dismiss; it is advice only and does not block signing.
That can be the right design: a hard block would push teams into accepting proposals simply to clear it. But it means the firm needs a rule of its own.
A review rule to write into the firm's own policy
A firm's system of quality management is where a rule like this belongs, beside its other review requirements. The IFAC's Quality Management Toolkit for Small- and Medium-Sized Firms offers illustrative policies, checklists, sample letters and forms, plus a risk matrix, which can help frame it.
A rule that works in practice:
- No set of statements goes to the partner for signing while any AI mapping proposal is open.
- The reviewer, not the preparer, clears proposals on revenue, loans, related-party balances and provisions.
- Every proposal set aside is followed by a client request, and the set waits for the answer.
- Before signing, the partner asks whether the proposal list is clear, and the reviewer confirms it.
- Any exception is recorded with a reason on the engagement file.
The rule costs a few minutes per set. It turns a list on a screen into a control the firm can show it operates, which is what an inspector, or a client who later disputes a figure, will want to see.
General guidance only, not advice on your own circumstances. Tax and payroll rules change: check the current position with the authority or your adviser.
