The first 30 days with AI in a small business
One job, one number, four weeks. A plan that keeps the first AI project small enough to finish and clear enough to judge.
Contents
The first AI project in a small business fails for ordinary reasons: it tries to do too much, nobody measured the starting point, and nobody owns it after the first week. A 30-day plan with one job and one number avoids all three.
Why a month, and why one job
MIT's 2025 study of enterprise AI found that most generative AI pilots showed little measurable effect on profit and loss, and that the largest returns came from back-office automation1. A small business does not need a pilot programme to learn the same lesson. It needs a narrow job, a baseline and a decision date.
Week 1: choose the job and measure it
- Write down a week of recurring work, as in our guide to AI tools by job.
- Pick one job that costs time or money when it is late or missed: calls that go to voicemail, invoices sent late, supplier email, meeting notes.
- Measure it for the week: number of missed calls, days from job to invoice, hours spent on supplier follow-up.
- Name one owner.
Week 2: set up, with limits and data rules
Before the AI touches real work
- Choose the plan with the right data terms. Business plans from Anthropic and OpenAI state that business content is not used for training by default23.
- Write the limits down. What it may send, change, book or pay without a person, and what waits for approval.
- Decide where actions are recorded and who reads the record.
- Connect only what the job needs: one inbox, one calendar, one ledger.
Tools differ in how they express limits. OpenLazarus sets them per teammate, with examples such as purchase orders over $15,000 or credits over $500 that wait for a person, and logs every action where staff can see it4. With an automation platform, the limit is the design of the automation itself; Zapier's agents page describes monitoring activity but not approval steps5.
Week 3: run it, with a short daily check
Let the tool do the job, and have the owner spend ten minutes a day on what it did. Correct its instructions rather than its output where possible, so the same mistake does not recur. Keep a short list of what it got wrong and why.
Week 4: measure and decide
Measure the same number as in week one, in the same way. Then make one of three decisions in writing:
- Keep and widen the job slightly, if the number moved and errors were rare.
- Keep but tighten the limits, if the number moved but errors were costly.
- Stop, if the number did not move, and record why.
| Job | Number to measure | A good month looks like |
|---|---|---|
| Phone calls | Calls unanswered or sent to voicemail | Fewer missed calls, with messages logged |
| Invoicing | Days from completed job to invoice | Invoices out the same or next day |
| Supplier follow-up | Hours a week spent chasing | Hours down, deliveries tracked |
| Meetings | Time to send notes and actions | Notes out the same day |
What success looks like
After a month, one recurring job no longer depends on one person being at their desk, the number is measured the same way every week, and the owner knows what the AI may and may not do. The next job is chosen on that evidence.
For a business whose first job is operational, such as calls, invoices or purchasing, OpenLazarus's AI teammates work inside a system the business owns, each with an inbox, a phone number and set limits6. For a first job in writing or meetings, an assistant or the AI in the office suite is the simpler start; see our main guide and the cost guide.
Questions buyers ask
How should a small business start using AI?
Pick one recurring job that costs time or money when it is late, measure it for a week, choose a tool that fits that job, set written limits on what the AI may do alone, run it for three weeks with a short weekly review, and keep it only if the number moved.
How long does it take to see results from AI?
For a narrow job such as answering calls, chasing invoices or summarising meetings, four weeks is usually enough to see whether the number moves. Projects that change a whole process take longer and should be split into smaller steps.
Who should own AI in a small business?
One named person per tool, usually the person whose work it affects most. They set the limits, read the log or results each week and decide whether to keep it.
Sources
- MIT report: 95% of generative AI pilots at companies are failing. Fortune, 18 Aug 2025. Accessed 8 Nov 2026.
- Claude pricing: Team and Enterprise plans. Anthropic. Accessed 8 Nov 2026.
- ChatGPT Business. OpenAI. Accessed 8 Nov 2026.
- AI for wholesale and distribution. OpenLazarus. Accessed 8 Nov 2026.
- Zapier Agents. Zapier. Accessed 8 Nov 2026.
- The agent library. OpenLazarus. Accessed 8 Nov 2026.
Updated 8 Nov 2026: First published.