Choosing the first accounting process to automate

The general framework for picking a first automation is in which process should you automate first. This is the narrower question: given the specific processes a finance function runs, which one should go first?

The criteria that matter for a first choice are different from the ones that matter generally. A first automation has an additional job: it has to succeed visibly enough that the second one gets approved.

What makes a good first choice

High volume. Enough transactions that the benefit is obvious within weeks.

Low judgement. So disagreements between the system and your team are resolvable rather than debatable.

A checkable right answer. Ideally an independent source to reconcile against.

Reversible. If it goes wrong, you reclassify rather than recover money.

Contained. Failure does not cascade into other processes.

Currently painful. So the team wants it to work, which matters more than any technical factor.

The candidates, scored

Supplier invoice capture — strong first choice

Volume: high. Judgement: low for most invoices. Checkable: yes, against the document and the purchase order. Reversible: yes. Contained: mostly. Painful: almost universally.

The standard recommendation, and it deserves to be. The work removed is visible immediately, the right answer is on the document, and errors are reclassifications rather than lost money.

Watch: capital-versus-revenue coding needs a rule requiring human classification above your capitalisation threshold, or it will be the source of most of your errors.

Bank reconciliation — strong first choice

Volume: high. Judgement: low. Checkable: yes, against an independent source. Reversible: yes. Contained: yes. Painful: yes.

Arguably the best technical candidate because the bank statement is genuinely independent — you are not checking the system against itself. Feeds arrive structured, which removes document-reading risk entirely.

Watch: clear the historical unmatched items first or the exception queue arrives full of archaeology.

One sales channel's reconciliation — strong, if it applies

Volume: very high. Judgement: low. Checkable: yes. Reversible: yes. Contained: yes. Painful: severely, for anyone doing it manually.

For businesses selling through marketplaces or taking card payments, this is often the highest-value first choice, because the manual version is close to impossible at volume and money is genuinely being lost in the gap.

Watch: pick one channel, not all of them.

Expense claims — reasonable, lower impact

Volume: high in count, low in value. Judgement: low. Checkable: yes. Reversible: yes. Contained: yes. Painful: yes, and disproportionately to the amounts.

Good for goodwill because everyone in the business benefits, not just finance. Lower financial impact, which makes it a weaker business case and a strong culture case.

Accounts receivable chasing — good, but not first

Everything about it is suitable except one thing: it depends on cash application being current. Chase a customer who has already paid and you damage a relationship and your credibility.

So it is a good second or third choice, after reconciliation is running. Doing it first is how implementations generate a burst of embarrassing emails.

The month-end close — poor first choice

Volume: low, it happens twelve times a year. Judgement: high. Checkable: partly. Reversible: partly. Contained: no, it touches everything. Painful: intensely.

The pain makes it tempting and the other five criteria make it wrong. It has the least learning opportunity, the most judgement, and the most visible failure mode.

The close does get faster — as a consequence of automating the processes that feed it. Attacking it directly is attacking the symptom.

The honest recommendation

For most businesses: supplier invoice capture or bank reconciliation. For businesses selling through marketplaces or gateways: one channel's reconciliation, which usually has the largest recoverable money in it.

Whichever you pick, pick one. The failure pattern is starting three simultaneously, at which point nothing can be run in parallel, nothing can be attributed, and if anything goes wrong the whole programme is in question rather than one process.

The question that settles it

If you are torn between two candidates, ask: which one would the team be most relieved to stop doing?

That is not a soft criterion. The people who have to work the exception queue, correct the coding and report the problems determine whether an automation succeeds. A technically superior choice that nobody wanted loses to a slightly worse one the team is invested in.

Common questions

Which accounting process should be automated first?

For most businesses, supplier invoice capture or bank reconciliation — both are high volume, low judgement, checkable against an independent source, reversible and contained. For businesses selling through marketplaces or payment gateways, reconciling one sales channel is often better still, because the manual version is impractical at volume and money is genuinely being lost in the gap.

Why not start with the month-end close?

Because it fails every criterion that matters for a first automation: it happens only twelve times a year so there is little to learn from, it is dense with judgement, it touches every other process so failure cascades, and it fails visibly in front of management. The close does get faster, but as a consequence of automating the processes that feed it rather than by being attacked directly.

Why is accounts receivable chasing not a good first choice?

Because it depends entirely on cash application being current. If receipts are applied weekly, the ledger is wrong most of the time and automated reminders will reach customers who have already paid — one such message costs more credibility than ten correct ones build. It is a strong second or third choice, once reconciliation is running.

Can I automate several processes at once?

It is the common failure pattern. Running several simultaneously means none can be run in parallel with the old method, problems cannot be attributed to a specific process, and a failure anywhere puts the whole programme in question rather than one part of it.

Related: starting with AI in accounting · when not to automate an accounting process · running an AI accounting pilot