Onboarding a client onto automated books

Onboarding decides how a client is served for years. Get it right and the client is profitable and low-friction. Get it wrong and you carry the consequences until the relationship ends, because habits established in month one are very hard to change in month twenty.

With automated books the stakes are higher, because the setup determines what the system learns.

Before you process anything

Get the history, and look at it. A year or two of prior transactions is what makes automated coding useful from the start. It also tells you what you have taken on — if the previous coding is inconsistent, you will inherit that inconsistency as a learned pattern, and it is far cheaper to decide the correct treatment now than to unpick it later.

Prune the chart of accounts. Whatever arrives will contain dead accounts and duplicates. Fixing it before processing begins is a couple of hours; fixing it after six months of learned patterns is a project. See data quality is the real constraint.

Deduplicate suppliers and customers. The same supplier under three spellings caps matching accuracy permanently and enables duplicate payment.

Reconcile the opening position. Every balance you inherit should decompose into items you can name. If the opening debtors will not break down, that is a problem you are adopting, and it is much easier to raise before you sign off the first period than afterwards.

The two conversations practices skip

How documents will arrive

Not a preference — an agreement. One channel, documents sent as they occur rather than in a batch after period end, and what happens when they do not arrive.

This is the conversation that determines whether the client is profitable. Skipping it means accepting whatever they currently do, which is exactly the arrangement that made their previous accountant slow.

Have it during onboarding, when the client is expecting change and motivated to cooperate. The same conversation in month eight is a complaint.

What the client is responsible for

Automated books create a shared process. The client's part — submitting documents, answering queries, approving what needs approval — has to be stated, because when accounts are late the question of whose fault it is will arise.

Worth putting in the engagement letter rather than leaving to goodwill.

The first two months

Month one: expect it to be poor. The system has history but no corrections. Coding proposals will be roughly right and frequently need adjusting, and the exception queue will be long — partly genuine, partly historical mess.

This is the month to correct properly inside the workflow rather than fixing things downstream with journals, because corrections made in the workflow teach and journals do not.

Month two: it should be visibly better. Recurring suppliers handled, exceptions dropping. If month two looks like month one, something is wrong — usually inconsistent history, or corrections being made downstream.

Month three: assess honestly. Is the exception queue short? Are the client's documents arriving as agreed? Is the time per client where you expected? If not, fix it now — at month three it is a conversation, at month twelve it is the way things are.

What to tell the client

Clients ask whether AI will make mistakes with their accounts. The answer that works is neither reassurance nor technical detail:

"Routine transactions are processed automatically and anything unusual comes to us. We check a sample of the automatic ones every quarter. You will see fewer questions from us and get your accounts faster. If something is wrong it is our responsibility, exactly as it was before."

That covers what they actually want to know: is it accurate, does someone check, and who is accountable. See explaining AI bookkeeping to a sceptical client.

The onboarding checklist

• History imported and reviewed for consistency
• Chart of accounts pruned, each remaining account defined in a sentence
• Suppliers and customers deduplicated
• Opening balances reconciled and decomposed
• Document channel agreed and set up
• Client responsibilities documented in the engagement
• Exception routing decided — what comes to you, what goes to the client
• Review checklist established
• A date set for the month-three assessment

The last one is the one that gets dropped, and it is the one that catches the problems while they are still fixable.

How long this onboarding takes says as much about how well the client's own processes were ever explained as it does about the automation itself — see what onboarding actually measures.

Common questions

What should be done before processing a new client's transactions?

Import and review the transaction history for consistency, prune the chart of accounts and define what belongs in each remaining account, deduplicate suppliers and customers, and reconcile the opening balances so every inherited balance decomposes into nameable items. Each of these takes hours at onboarding and becomes a project once six months of learned patterns sit on top of them.

What conversations should happen during onboarding?

Two that are frequently skipped: how documents will arrive — one channel, sent as they occur rather than batched after period end, with an agreed consequence when they do not — and what the client is responsible for, documented in the engagement letter. Both are straightforward during onboarding when the client expects change, and become complaints if raised months later.

How long before automated books settle down for a new client?

Month one is usually poor, with proposals needing frequent correction and a long exception queue containing both genuine items and inherited mess. Month two should be visibly better as recurring patterns are learned. If month two resembles month one, the usual causes are inconsistent historical coding or corrections being made downstream through journals rather than inside the workflow, where they would teach the system.

What should you tell a client about AI handling their books?

That routine transactions are processed automatically, anything unusual comes to a person, a sample of the automatic ones is checked periodically, they will receive fewer queries and faster accounts, and responsibility for correctness remains with the practice exactly as before. That answers what clients actually want to know — accuracy, oversight and accountability — without technical detail they did not ask for.

Related: standardising processes across a client base · explaining AI bookkeeping to a sceptical client · preparing your data before automating