The Cost of Automating Accounting: Real Ranges by Volume

The cost of automating accounting in a small or mid-sized company is, in most cases, €4,000-15,000 to build, plus annual maintenance of 10-20% of that figure. The range narrows as soon as three facts are on the table: how many documents arrive each month, how many systems have to be touched, and how many exceptions your real process carries — which is almost always more than the manual claims.

We give ranges rather than a fixed price because accounting automation quotes vary for a legitimate reason: you are not paying for software, you are paying to adapt a process. Two companies with identical invoice counts can differ threefold in price if one has a clean chart of accounts and a single ERP while the other runs three legal entities, two cost allocation criteria and a supplier master full of duplicates. This article breaks down what moves the price, what is included, and how long it takes to pay back.

What actually drives the price?

Four variables, in this order of weight. The first three can be measured in an afternoon; the fourth is what the project discovers.

• Monthly document volume. Supplier invoices, sales invoices, receipts, bank movements and payroll inputs. Below a certain threshold, deep automation does not pay — and we say so even when it costs us the work.
• Number of systems involved. An ERP with an API is one project; an ERP without an API plus a bank plus a shared folder plus email is four integrations, each with its own cost and its own fragility.
• Exception density. Suppliers with odd formats, invoices without a purchase order, multi-cost-centre expenses, partial VAT recovery, intra-EU transactions. Exceptions do not make the build more expensive — they make the rules and the testing more expensive.
• How much autonomy you want. Having the agent propose entries for a human to approve is markedly cheaper than reaching automatic posting with thresholds, controls and full audit. Starting with the former and evolving costs less than attempting the latter on day one.

What does it cost at your volume?

This table reflects what we see in real SME projects. These are build ranges, with human review at the start, scoped to the purchase, sales and banking cycles.

One important caveat about the first row. Below 300 documents a month, the hours saved rarely justify a custom project. The honest advice in that bracket is to start with standard capture tools, tidy the process, and revisit the conversation when volume grows — or when the problem is not cost at all, but that a key person is trapped in a task that blocks more valuable work.

What is included in that price, and what is not?

A serious accounting automation quote includes six things. If yours does not list them, ask, because whatever is missing from the quote reappears later as a change request.

1. Analysis of the real process and a measured baseline: documents, minutes per document, error rate, days to close.
2. Building the agent and its rules for classification, accounts, tax treatment and duplicate detection.
3. Integrations with the accounting ERP, the bank and the document sources.
4. Testing against your historical data, not demo samples. It is the phase that builds the most trust and the first one cut when budgets tighten.
5. Training and a support period for the team that will review and approve.
6. Monitoring and alerting so you find out when something stops working, which it will.

It usually does not include — and should be budgeted separately — cleaning up the supplier master data, migration if you decide to change ERP, and licences for the ERP itself or for a capture tool if one is needed. Nor does it include model consumption, which in these flows is typically a small share of the total but should still be estimated and monitored from month one.

How long does it take to pay back?

Simple arithmetic you can run today. Posting a supplier invoice manually takes 2-4 minutes when everything goes well, and considerably longer when a missing detail has to be chased. At fully loaded internal cost, each document consumes roughly €1-2.50 in time alone, before errors and rework.

At 1,200 invoices a month and 3 minutes each, that is 60 hours monthly. A deployment that removes 70% of that time frees around 42 hours a month; over twelve months, an investment in the €8,000-15,000 bracket typically pays back inside the first year — faster if it also shortens the close or stops duplicate payments. We apply the same calculation framework, in more detail, in invoice automation with AI.

Two returns almost nobody budgets tend to outweigh the hours saved: the duplicate payments that stop leaving the building, and the days shaved off the close. Closing four days earlier changes the quality of management decisions, and that never appears on a timesheet. We develop it in AI accounting close automation.

Which parts of accounting should not be automated?

The parts that require professional judgement and carry personal responsibility. The line here is clear and we do not move it for budget reasons.

• Tax treatment of doubtful transactions. Debatable deductibility, related-party dealings, timing of recognition. The agent assembles the complete file; a person decides.
• Close adjustments, provisions and accruals. Supported with data and proposals, but decided by people.
• Filings with the tax authority. Prepare and cross-check, yes; submit without sign-off, no.
• Changes to the chart of accounts or allocation policy. That is design, not execution.

The rule we always apply: automate the process, never the judgement. If a decision changes the tax position or the reported result, a person signs it.

How to request a quote you can actually compare

1. Bring your numbers. Documents per month by type, number of legal entities, ERP and version, and whether the bank feed is automatic.
2. Ask for phased scope, with phase one priced firm and later phases estimated. A single price for everything usually hides assumptions.
3. Insist the baseline is in the quote. If the vendor does not propose measuring the "before", they will not be able to prove the "after".
4. Ask what share of documents will still need review at month 3 and month 6. The honest answer is never "zero".
5. Confirm what maintenance covers exactly: ERP version changes, rule tuning, incidents and monitoring.

We run our own businesses this way and we quote client work the same way: by process, with a baseline, with human review at the start. If you want the full map of the finance function before deciding where to start, it is in our AI finance automation guide; and if you would rather work through your specific case, that is what we do in our AI consulting practice.

Frequently asked questions

Can we automate accounting without changing ERP?

Yes, and that is the normal case. The agent works on top of the ERP you already have, through its API where one exists and through import files where it does not. Changing ERP at the same time as automating doubles the project risk and multiplies the timeline: if the ERP has to change, do it before or after, never during.

What if an external accounting firm keeps our books?

Then the calculation moves, but it does not disappear. What you automate is your side of the cycle: preparing, classifying and delivering documentation in an orderly, complete way. Many firms charge by volume and by exceptions, so reducing the mess reduces the fee and, above all, removes the email ping-pong that consumes your admin team.

How long does the project take?

A scoped deployment — the purchase cycle with human review — goes live in 4-8 weeks. The full cycle with sales, banking and controls typically runs a quarter. Timelines slip for two reasons almost every time: access permissions that take weeks to be granted, and dirty master data.

Is it safe to let AI touch the accounts?

With the right design, yes, and it usually reduces errors: an agent does not miss a duplicate because it is seven in the evening. The three guarantees to demand are minimum permissions, thresholds by amount, and a complete log of every action so any entry can be reconstructed months later. Without traceability there is no defensible accounting project.

Is it worth it at low volume?

Below 300 documents a month, usually not as a custom project. It is worth it when the problem is not cost but that a specific person is trapped in the task and that blocks higher-value work, or when document chaos is stretching your close. In that case, start by tidying and capturing rather than by building.