The reason most small business AI efforts stall is not the technology. It is that "adopt AI" is not a task. Nobody can start it on a Monday. So here is the version you can start on a Monday: a 30-day sequence that takes exactly one workflow from manual to measured automation. By day 30 you will have either a working system with a number attached, or a cheap, fast "no" that saved you a year of subscription fees.
One workflow, four weeks, one decision at the end. Scale it or kill it; both are wins.
Week 1: Find the Workflow (Days 1-7)
Spend the week noticing, not buying. Keep a running list of every task that is repetitive, text-heavy, and annoying: inquiry replies, quote assembly, appointment reminders, invoice chasing, meeting notes into follow-ups. At the end of the week, score each candidate on three axes: hours per week it eats, how completely you can describe the steps, and the cost of a rare mistake. Pick the one with high hours, a describable process, and survivable errors. Resist the temptation to pick two. The most common way this plan fails is by becoming three plans.
McKinsey's State of AI research keeps landing on the same finding: value comes from redesigning specific workflows, not from spreading licenses across a team. Week 1 is you doing that redesign at small-business scale.
Week 2: Pilot It by Hand (Days 8-14)
Do the workflow with AI, manually, every day for a week. A chat window, your real inputs (minus anything on your red data list, per the data rules), and a prompt you refine each morning. Drafting inquiry replies? Paste each real inquiry in and grade the draft before you send it. By Friday you will know the honest automation rate: maybe 80% of these need no edits, or maybe the AI misses your tone constantly. This week costs a subscription at most, and it is the cheapest research you will ever run. If the manual pilot fails, stop here. You just saved the build cost, and you learned it in seven days instead of ninety.
Week 3: Wire It Up (Days 15-21)
Only what survived week 2 gets plumbing. Connect the trigger (form submission, inbox, calendar event) to the model to the destination (draft folder, CRM, task board) using a workflow tool like n8n, Zapier, or Make, the same options I priced out in the running-costs post. Two rules for the first version: outputs land as drafts for review, not as sent messages, and every run gets logged where you can read it. If the wiring is beyond your appetite, this is the one week where outside help compresses months into days; the workflow definition you built in weeks 1-2 is exactly the spec a builder needs. My first-agent guide covers the architecture choices in more detail.
Week 4: Measure and Decide (Days 22-30)
Run it all week, review every output, and tally three numbers: hours saved versus your week-1 baseline, percentage of outputs needing edits, and anything that went wrong. Then hold the day-30 meeting, even if it is a meeting with yourself, and make one of three calls:
- Scale it. The number is good. Remove the review step for the safest category, keep it for the rest, and pick workflow number two, which now takes two weeks instead of four because the muscle exists.
- Keep it supervised. The drafts save real time but need human eyes. That is still a win; most mature AI workflows live here permanently.
- Kill it. The number is not there. Cancel the tooling, write down why in three sentences, and keep the week-1 audit list; the next candidate is already on it. A fast kill is a success. The graveyard, as I wrote in the graveyard post, is built from decisions nobody made.
The Bottom Line
Thirty days, one workflow, one decision. No transformation initiative, no tool shopping spree, no committee. The businesses that are quietly ahead on AI in 2026 all started with a version of this month, and then simply ran it again.
Sources
- McKinsey & Company, "The State of AI" (workflow redesign as the primary driver of reported AI value). mckinsey.com
- Gartner, "Gartner Predicts 30% of Generative AI Projects Will Be Abandoned After Proof of Concept By End of 2025" (the failure pattern week 2 is designed to catch early). gartner.com
Want to run the 30 days with someone who has done it before?
We run this exact sequence with clients: pick the right workflow in week 1, and have the wiring done professionally by week 3.