AI transformation that actually changes the books

The first reason is scope. A programme that touches every function at once has no single owner and no clear before and after.

The second reason is the pilot trap. A pilot proves a model can do something. It does not prove the work changed. Plenty of companies finish a year of pilots with the same headcount doing the same steps.

The third reason is the system underneath. If the ledger is batch, if data lands weekly, and if the group view is rebuilt by hand every month, an assistant on top cannot fix that. It just drafts faster into the same queue.

The fourth reason is measurement. If nobody wrote down how long the close took before, nobody can say whether it got shorter.

Finance is the best place to start because the work is repeatable and the outcome is countable. Days to close. Invoices touched by a human. Receipts still missing at month end.

Step one is where documents arrive. Bills and receipts come in by mail, upload, or card spend. AI reads them, including line items, tax, and terms, so nobody types them again.

Step two is approval. Approvals move to Slack or Teams, where people already are, rather than a portal they log into once a week.

Step three is the ledger. This is the part that decides everything else. On a batch ledger, entries land in cycles and the group picture is assembled afterwards. On a live ledger, the picture is current, so an agent can act on it instead of waiting for a period rollover.

Step four is the agents. In Light, agents run AP, AR, reconciliation, and the close on that live group ledger. Finance reviews the result rather than assembling it. That is different from a copilot, which drafts while a person still posts.

Step five is the group. Multi-entity and multi-book are the core, and revenue recognition is natively included. For a group with two or more legal entities, that removes the monthly consolidation scramble instead of speeding it up.

Step six is control. Reviews, audit trail, and payment approvals stay with finance. Native bill pay runs from the company's own bank across more than eighty countries.

Step seven is the honest boundary. Light is a finance system, not a full operational ERP. It will not run inventory or a shop floor. Named customers include Tillo, KeyShot, and Alva Labs.

Change that crosses many functions and needs political cover. A large firm can hold a programme together across operations, supply chain, and finance in a way a tools purchase cannot.

Regulated or heavily audited environments where documentation and governance are the deliverable.

Companies with no internal owner available. If nobody can be accountable full time, a tools-led route will drift.

The trade is time and cost. Expect quarters before anything changes in the books, and expect the knowledge to walk out at the end unless you staff alongside.

Write down three numbers before you start: days to close, share of invoices touched by a human, and hours spent on consolidation.

Change one workstream fully rather than five partially. Finance is a good first choice because those three numbers move or they do not.

Give each workstream an internal owner with authority to change the process, not just the tool.

Re-measure after one full quarter, including a real month end. Then decide whether to widen.