What changed
For most of the last decade, a growing company's finance function was mostly assembly: exporting from billing, pasting into a spreadsheet, reconciling against the ledger, building the same charts. In 2026 a model like Claude, connected to QuickBooks, Stripe and a CRM, can do most of that assembly reliably and overnight.
That doesn't make finance less important. It moves the value. When the numbers are cheap to produce, the scarce things are knowing which numbers matter, knowing whether they are right, and deciding what to do about them.
Automate now
- Data pulls and syncs. Every system, every night, into one place.
- Reconciliations. Billing to ledger, CRM to billing, cash to bank, with breaks listed instead of buried.
- First drafts of variance explanations. “Hosting is up 14% because of three new enterprise customers on dedicated instances” is a draft a model can write and a person can confirm.
- Monthly package assembly. Statements, metrics and charts built the same way every month.
- Collections follow-ups. Drafted, specific and polite, sent only after approval.
- Monitoring. A Monday brief that flags anything unusual before your first meeting.
Keep a person on these
- Definitions and policy. What counts as churn, when revenue is recognized, how a customer is counted. These are choices, and they need an owner.
- Estimates and judgment. Accruals, reserves, capitalization, anything where reasonable people could disagree.
- Anything that leaves the building. Board decks, investor updates, lender reports and diligence answers need a person who signs them and stands behind them.
- Decisions. Whether to hire, raise prices, cut spend or take the term sheet. AI can model the options. Someone has to choose.
- Saying no. A CFO's most valuable sentence is sometimes “we can't afford that yet.”
Guardrails that make it safe
- Read-only by default. Connections read data. Anything that writes or sends starts as a draft.
- Tie-outs before trust. Every automated number reconciles to its source, and the output says so when it doesn't.
- One source of truth. Metrics are calculated once, from written definitions, not re-derived in every chat.
- An audit trail. You can see where each number came from and who approved it.
- A named human. Someone is accountable for every number that reaches a decision-maker.
The model we run
Our shorthand: Perch has the numbers, and your analyst has your back. Syncing, checking and first drafts are automated. The judgment calls, the follow-ups and the board prep are a person who knows your business and works in your Slack. When you ask a question in Perch, you get an answer in minutes with its sources attached, and when it needs judgment, your analyst picks it up within one business day.
How to start
Start with three workflows that save time every week and are easy to check: a Monday brief, an invoice chaser and a month-end close packet. Run them on your real data, check their output against your books for a month, and expand from there. That is exactly what our Finance Automation Sprint sets up in two weeks.
Common questions
Can AI replace a CFO?
Not in 2026. AI can do much of the mechanical work a finance team does, but definitions, estimates, decisions and anything that goes to a board, lender or investor still need a person accountable for it.
What finance tasks should a small company automate first?
A weekly cash and revenue brief, collections follow-up drafts, and month-end close packet assembly. They save time every week and are easy to check.
