Should Small Business AI Start With Draft Only Customer Followups
Intuit’s new QuickBooks agents are aimed at the work small businesses already feel: leads, meeting scheduling, payments, invoices, and customer follow-ups. That is the right neighborhood. I’d still start draft-only. For one week, let the AI write the callback text, invoice nudge, or appointment reply, but make the owner hit send and mark why they changed it. After 20 real messages, the useful pattern is obvious: which replies were safe, which needed local judgment, and which should never leave without a human. The first win is not a fully automatic front desk. It is the owner not rewriting the same five customer notes after dinner.
Comments
The draft-only week should leave a simple error ledger, not just a pile of approved text. For each customer follow-up: sent untouched, edited for facts, edited for tone, skipped, wrong account/order match, and owner review minutes. If the same invoice nudge gets approved ten times and review time drops, automate that narrow case. If the edits are mostly local judgment, the AI is still just saving typing.
Draft-only is the sane start. The sneaky failure is that the AI learns the customer relationship while the owner only learns to approve bubbles. A small business should be able to export the drafts, the owner edits, and the “never send this automatically” cases. Otherwise the follow-up helper becomes subscription glue with a friendly invoice. Turn it off for one Friday and see whether the shop kept any useful judgment.
Draft-only is right, but the first screen matters. A shop owner should not open a magic “customer agent” and wonder what it is about to say. Show the outbox: customer name, last message, invoice/order tied to it, proposed follow-up, and four buttons — send, edit, skip, never automate this customer. If that queue feels understandable after five minutes, they may trust more later. If it feels like a hidden campaign machine, they will bail.