Accounting automation: which step to tackle first, and what it saves

Accounting automation does not start with the supplier invoice. It starts with the steps that repeat identically every month: pulling source data, reconciling intercompany balances, consolidating and assembling the reporting pack. That is where most of the manual work sits, where the least judgement is required, and where close lead time drops fastest.

For a controller running several entities this is a sequencing question rather than a tooling question. The input side of the ledger has usually been partly automated by the accounting package already, while the back of the chain – reconciling, consolidating, reporting – still runs on spreadsheets. Start there and you save days. Revisit invoice processing that already works and you save hours.

What accounting automation actually takes over

Break the record-to-report chain into its parts and seven steps remain. How much of each can be automated differs sharply.

Step	Can be automated	Stays manual
Pulling source data from the ledgers	Fully	Monitoring that the connection still runs
Reading and coding supplier invoices	Largely	New suppliers, capitalise or expense
Matching bank transactions	Largely	Partial payments and payment differences
Recurring monthly entries	Fully, rule based	New contracts, provisions, estimates
Intercompany reconciliation and elimination	Largely	Explaining a remaining difference
Consolidation	Fully	Ownership percentages and structural changes
Assembling and distributing reports	Fully	Commentary and interpretation

Only two of those seven steps are really about bookkeeping. The other five are moving, matching and adding up data: work without judgement, which is exactly why it lends itself to automation. What happens at the input side with models is a separate topic, covered in AI bookkeeping.

How to approach an accounting automation project

1. Measure lead time per step first. For one full month, record when every step starts and when it finishes. Without that baseline you automate on instinct, and instinct almost always points at the most irritating step rather than the longest one.
2. Replace exports with connections. A monthly export from the accounting package is a manual step that quietly produces stale figures. Smartbooks refreshes data several times per hour from Exact Online, Twinfield, AFAS, Odoo, QuickBooks and Exact Globe, and shows per source when the last successful refresh happened. Each entity can sit on its own source system.
3. Standardise the chart of accounts before you consolidate. Unique ledger codes across all entities, identical subfolder codes for identical cost types, and at every level either subfolders or accounts, never both. Dull work you do once, and it makes every later step cheaper.
4. Capture intercompany relations in a matrix. On that basis, internal transactions are matched automatically, differences in amount or timing are flagged, and elimination entries are proposed. A transaction with a counterpart in an adjacent period is marked as partially matched – a timing difference, in other words, and no error.
5. Let consolidation follow period publication. Once the underlying entities publish their period, the consolidated set appears within minutes. You move from adding figures up to verifying them.
6. Fix the reporting pack in templates. One set of report definitions for every entity and for the consolidation. Rebuilding a spreadsheet each month is repetition, not automation.
7. Move the freed-up hours into analysis. Otherwise automation buys you an earlier deadline and no extra insight. See how the month-end close relates to the rest of the chain.

Example: closing a group of four entities

A neutral example of a group with four operating companies and a holding, with steps 2 through 6 above implemented. The figures illustrate proportions; they are not a customer result.

Step	Before	After
Collecting figures from five ledgers	1.5 days	0 – continuous
Intercompany reconciliation	1 day	2 hours
Consolidation and elimination	1 day	minutes
Assembling the reporting pack	1 day	0.5 hours
Analysis and commentary	0.5 day	1.5 days
Total	5 days	2 days

Look at the line above the total: analysis gets longer. That is the point. A close that drops from five days to two while analysis triples delivers more than a close that drops to a single day and leaves nobody looking at the numbers.

Limits and risks

• Data quality sets the ceiling. An automated chain processes a misposted entry just as diligently as a correct one. Automation accelerates the error too.
• A broken connection breaks silently. Check the refresh status per source before every close. Authorisations expire, and a report built on three-week-old data looks exactly like a report built on current data.
• Auditability. Your auditor will ask which entries were generated automatically and under which rule. Proposed eliminations should trace back to the underlying transactions, not to a balance.
• AI commentary is a draft, never a conclusion. A model explaining a variance uses the same confident tone whether it is right or wrong. Always make the answer resolve back to the journal entries beneath it.

Pitfalls in accounting automation

• Starting with invoice processing because it is the most visible step, while consolidation costs three times as many days.
• Doing everything at once, so that at the first discrepancy nobody knows which step caused it.
• Consolidating on a chart of accounts with duplicate ledger codes, then repairing the aggregation by hand every month.
• Filling in the intercompany matrix only halfway. Relations that are not mapped cannot be detected and return as an unexplained difference.
• Measuring success in hours saved instead of corrections booked after the close.

Frequently asked questions about accounting automation

What should you automate first in accounting?

The step that costs the most days and requires the least judgement. In a group with several entities that is almost always collecting and consolidating figures, not invoice processing. Measure lead time per step for a month before choosing.

What is the difference between accounting automation and AI in bookkeeping?

Chain automation runs on fixed rules and connections: the same action, every month, without intervention. AI in bookkeeping proposes treatments in situations no rule covers, such as an invoice from an unknown supplier. In a mature finance function the two sit side by side.

Does automation remove the controller's role?

The role shifts. Adding up, rekeying and reconciling disappear; judgement on valuations, explanation of variances and accountability towards the board and the auditor remain. In practice the analytical part of the role grows, because figures arrive earlier.

How many entities do you need before automation pays off?

From two entities with transactions between them the arithmetic starts to work, because intercompany reconciliation and consolidation then recur every month. With a single entity the gain sits mainly in reporting templates and the link to the accounting package.

Automation is a sequence rather than a project: first the steps without judgement, then the steps with judgement, and the freed-up hours into analysis. Read on about designing bookkeeping as a record-to-report process, or about what AI does to month-end closing and financial reporting.