Ask any group controller what they dread most about the monthly close, and intercompany eliminations come up quickly. Not because the concept is complicated — it isn't. But because doing it accurately, across multiple entities, different ERPs, and a moving set of transactions, is where the manual process reliably falls apart.

01 — Where the manual process breaks

When one entity sells to another, both sides record the transaction. At group level, neither entry should exist — it was money moving from one pocket to another. Eliminations remove those entries before you report. In principle, straightforward. In practice, three things make it hard.

Timing. Entity A books the sale on the 28th. Entity B books the purchase on the 3rd of the following month. Nothing in either system resolves that automatically. Someone has to find it and decide how to treat it.

Naming. Entity A calls the account "IC Revenue — Group." Entity B calls the same flow "Intragroup Sales — Domestic." They're the same thing, but nothing in either ERP knows that. Someone has to maintain the mapping, keep it current as accounts change, and make sure it holds every month.

Volume. These problems don't stay manageable as groups grow. A group with fifteen subsidiaries trading internally can have hundreds of intercompany flows to reconcile each close. Most finance teams handle it with Excel, email, and institutional knowledge. It works until someone leaves, until an entity is added, or until the auditors ask for documentation.

02 — What AI actually changes

The honest answer: AI doesn't remove the complexity — it handles the mechanical parts so a controller can focus on the judgement calls.

Pattern recognition across accounts. AI identifies which accounts map to each other across different ERPs, even when naming is inconsistent. It suggests the mapping; a controller approves it. The system doesn't act until a human confirms.

Continuous matching. AI matches intercompany transactions as they post — not in a batch at month-end. Timing differences surface immediately rather than during close week. By the time the controller sits down to close, the backlog is a short list of items that genuinely need a decision, not 300 rows to work through.

Consistent FX handling. For the majority of intercompany trading flows, FX differences follow defined rules. AI applies them every time, without the calculation errors that creep in when someone is doing it manually under close-week pressure. Edge cases that require controller judgement get flagged — not auto-resolved.

What AI shouldn't do is make judgement calls without oversight. When a transaction is ambiguous, the system should flag it, not resolve it. A good AI-driven elimination process leaves every real decision with the controller.

‍

03 — Three questions to ask any vendor

Does the AI suggest, or does it act?

Any system that eliminates transactions automatically without human review introduces more risk than it removes. Look for a process where AI proposes and a controller approves — every time.

Is matching continuous or month-end?

Batch matching means you still discover problems during close week. Continuous matching means problems surface as they happen. The operational difference is the close getting shorter and the conditions that create reconciliation errors being systematically removed.

What's the audit trail?

Regulators and auditors need to see how each elimination was derived — not just the net result, but the individual matches, approval timestamps, and rationale for any manual overrides. If a vendor can't show you that clearly, keep looking.

‍

Hours — Time to first consolidated view, not months

95% — Reported time savings on consolidation cycles