← Back to guides
Finance7 minUpdated 2026-04-30

How to automate month-end close

The mid-market close runs eight to ten business days, the top quartile is at four point eight, and the gap is mostly manual reconciliation. This is a playbook for closing the gap with agents on top of the existing ERP, walked through the way a controller would actually run it.

Pick the close where the agent's work is provable

Run the agent in parallel against last month's close, not next month's. The controller's actual numbers from the prior period are the ground truth. The agent drafts the same close pack against the same data, and the controller compares. Differences surface fast, and the controller calibrates the agent's voice and the variance commentary against their own. Two cycles of parallel running is enough to trust the agent's output for the live close.

Decide on the read-write boundary on day one

Some controllers want the agent to read only and post nothing. Some want the agent to draft and post on approval. Both are first-class workflows. The recommended starting point is read-and-draft for the first month, draft-and-post-on-approval for the second month onward. The controller's audit posture stays the same in both modes.

Wire the inputs the agent needs

Bank feed (Plaid or the bank's direct feed). The ERP (NetSuite, Sage Intacct, QuickBooks, Xero, SAP, Oracle, Dynamics 365, Ramp). The payroll provider (direct or through Finch). The expense management system (Ramp, Expensify, Concur). The AR system if it lives outside the ERP. The agent reads from all of them and assembles the close pack against the chart the customer already uses.

Calibrate the variance commentary

The agent writes the variance explanation in the controller's voice. The controller seeds the voice by editing the first month's draft, and the agent calibrates against the edits. By the third month, the variance commentary reads exactly like the controller wrote it, because the agent has been reading the controller's edits for two cycles.

What to measure

Close days, controller hours per cycle, AP exceptions resolved before the close, and the variance commentary turnaround time. The first three drop on cycle one. The variance turnaround drops on cycle two. The compounding outcome is FP&A getting the numbers six business days earlier, which moves the board pack forward by the same six days.

Frequently asked

What if the close depends on a system the agent does not adapt to?

The agent can read CSVs and sub-ledger exports for systems without an API. That covers most of the long tail. The dashboard surfaces the missing items so the controller knows where to upload.

Does this work for multi-entity close?

Yes. Each entity reconciles independently, and the consolidation pack joins across entities with intercompany elimination drafted automatically.

Keep reading

Try it on your stack.

Start free