We profile a different small business every month here at Small AI Hub. This one is Aurora — a four-person bakery in Portland, Oregon, owned by Marisol and Diego. They have been open since 2021, sell mostly direct to the neighbourhood with a small wholesale side, and were burning through 15 hours of weekly admin until early 2026.

Six months later, they are down to about 5 hours. They added no staff, switched no major systems, and use only four AI tools you already know. We sat down with them on a Tuesday morning over coffee to map what changed.

What 'admin' meant before

Marisol's old Sunday routine: 90 minutes writing Instagram captions for the week. Diego's old Tuesday routine: two hours sending invoice follow-ups to the three coffee shops they wholesale to. Staff meetings: one hour every other Friday, plus another 45 minutes typing up notes. Customer questions on social DMs and email: maybe 90 minutes a day, scattered.

Add it all up — 15 to 17 hours every week of work that was not making bread. They tried hiring help twice and both times the cost did not work out for a four-person shop.

Tool 1 — ChatGPT for customer messages and order writebacks

Marisol pays $20/month for ChatGPT Plus and uses it on her phone, often one-handed while running the front of house. She built a custom GPT trained on three things: Aurora's tone of voice (warm, specific, not corporate), the bakery's offerings, and a list of common customer questions.

Every customer DM or email now starts as a 30-second draft from the GPT. She reads it, edits one or two words, sends. Time saved: roughly 60-70 minutes a day.

Tool 2 — Canva Magic Studio for daily social posts

Sunday captions are now a 25-minute Sunday-morning job instead of 90 minutes. The combo: Marisol photographs the week's special on her phone, drops the photo into a saved Canva template, lets Magic Write generate three caption options, picks one, schedules in Buffer (next tool).

She told us the unexpected benefit was consistency, not time. 'Before, weeks where I was tired meant skipped posts. Now even on tired weeks I show up.'

Tool 3 — Otter for staff meeting notes

Diego now records every other-Friday staff meeting in Otter ($16.99/month for the Pro plan). It transcribes, summarises action items, and emails the summary to all four staff within 5 minutes of the meeting ending.

The 45 minutes Diego used to spend typing notes is gone. The unexpected benefit: action items get done more often, because the staff actually read a 200-word summary in a way they never read a typed-up 'minutes' document.

Tool 4 — QuickBooks AI for wholesale invoice follow-ups

Diego upgraded from QuickBooks Simple Start to Essentials ($65/month) specifically for the Intuit Assist AI. It now drafts polite-but-firm follow-up emails to the three wholesale accounts when invoices age past due dates. Diego still sends them, but he edits two sentences in 30 seconds instead of writing the whole email from scratch.

The two hours Tuesday morning is now about 20 minutes. They also got paid faster — average days-to-payment dropped from 18 to 11 — because the AI follow-ups went out on time every single week, which Diego had not been doing manually.

The hours saved, broken down

  • Customer DMs/emails (ChatGPT): 60-70 min/day x 6 days = ~6.5 hours/week saved
  • Social media posts (Canva + Buffer): 90 min - 25 min = ~1 hour/week saved
  • Staff meeting notes (Otter): 45 min every other week = ~22 min/week saved
  • Invoice follow-ups (QuickBooks AI): 2 hours - 20 min = ~1.6 hours/week saved
  • Total: ~9.4 hours/week, plus an extra ~1 hour from less context-switching

Total monthly AI cost: $20 (ChatGPT) + $14.99 (Canva Pro) + $6 x 3 channels (Buffer) + $16.99 (Otter) + $30/month upgrade delta on QuickBooks = $99.98/month. Total hours saved per month: roughly 42.

What did not work

Marisol tried two other tools that did not stick. An AI inventory forecasting tool (the bakery is too seasonal and too vibes-based for it to predict demand accurately) and an AI 'review responder' (their handwritten replies to Google reviews are part of the brand).

She also told us — and this is the most important sentence in this article — 'I had to make myself stop trying new tools after the first four worked. There is always another tool. The discipline is using the four you have well.'

Three lessons for your business

  • Start where you waste the most time. Aurora's biggest savings came from customer messages because customer messages were where they were leaking the most hours. Not from where the marketing said the savings would be.
  • Custom GPTs trained on your tone of voice beat generic AI drafts every time. The 30 minutes to set one up pays back in week one.
  • Stop adopting new tools after four. The compounding value of using a small stack well always beats a sprawling stack used poorly.

The bottom line

A four-person bakery making bread and serving coffee should not need to become an AI lab to run efficiently. They did not. They picked four tools that solved their four biggest time leaks, spent two weekends setting them up, and got back ten hours a week — which is what they now use to actually bake one new product per quarter and take a real day off.