Five back-office jobs AI already does well (and three it doesn't)

Most AI advice for small businesses is written by people selling AI to small businesses. So here is the unsold version: a field guide to where today's tools are genuinely dependable in a five-to-fifty-person company, and where they will quietly cost you more than they save.

Five jobs it does well

Inbox triage. Sorting, labeling, and routing inbound email — support versus sales versus supplier — is the closest thing to a solved problem. Errors are cheap (a mislabeled email) and volume is high.

First drafts of routine replies. Quote follow-ups, booking confirmations, FAQ answers. The trick is the word "draft": a person hits send, the AI just kills the blank page.

Meeting and call notes. Transcription plus summary plus action items is reliable enough that not using it is now the expensive option, especially for sales calls that otherwise live in one person's memory.

Document extraction. Pulling fields out of invoices, receipts, and forms into a spreadsheet or accounting tool. Accuracy is high when the documents are consistent, and the failure mode — a flagged field — is visible.

Internal search and Q&A. "What's our refund policy for X" answered from your own documents. The knowledge was always there; AI makes it retrievable by the newest hire.

Three jobs it doesn't

Anything that moves money unsupervised. Payments, refunds, payroll changes. The error rate is low, but the cost of one error is not, and automation without review concentrates exactly the errors a person would have caught.

Final-say customer communication in a crisis. A templated apology written by a machine reads like one. When the stakes are a relationship, the human costs twenty minutes and is worth it.

Judgment calls dressed as admin. Approving a discount, prioritizing a late order, deciding which supplier to burn. These look like workflow steps; they're actually the business. Automate the paperwork around the decision, never the decision.

The pattern

Reliable automation in a small company follows one shape: high volume, low stakes per item, and a person positioned where errors surface. Hold every pitch — including ours — against that shape and you'll make good calls.