Where should a small business start with AI?

Not with a strategy, and not with a tool. Start with one annoying, repetitive job that happens every week.

Updated

The short answer

Pick one task that happens at least weekly, follows roughly the same pattern each time, and where a mistake is easy to catch before it does any harm. Run it with a person checking every result for a few weeks. If it saves real time, keep it and pick the next one. If it doesn’t, you’ve lost little.

Step 1: List the week’s repetitive work

Ask your team one question: “What did you do this week that felt like a computer should have done it?” You’ll get a better list than any consultant could produce from the outside. Common answers include:

  • Retyping details from emails or forms into another system
  • Writing nearly the same reply again and again
  • Chasing people for documents, payments or answers
  • Turning rough notes into a quote or a report
  • Pulling numbers together for a weekly summary

Step 2: Score each task

Ask three questions about each item:

  1. How often does it happen? Weekly or more is good. A once-a-year headache rarely pays back the setup.
  2. How similar is each instance? The more it follows a pattern, the better it’ll work.
  3. How bad is a mistake? A rough draft you review is low risk. Anything that sends money or gives legal, medical or safety advice isn’t.

Choose something that does well on all three. To put a dollar figure on it, try the busywork calculator.

Good first projects

  • Drafting replies to common email questions, for a person to send
  • Instantly acknowledging new leads and collecting their details
  • Turning voice notes or meeting notes into a written summary
  • Pulling line items from bills for bookkeeping to review
  • Sending reminders for overdue documents or invoices

Leave these for later

  • Anything unsupervised that speaks for you. Earn trust with drafts first.
  • Hiring or firing decisions. The legal and fairness risks are real.
  • Advice about health, safety, money or the law. A confident wrong answer can do real damage.
  • A process nobody agrees on. Sort out the process first, because automation only makes it fail faster.

Step 3: Do you even need AI?

Sometimes the answer is a simple rule. “When a form arrives, add it to this spreadsheet and send this email” needs no AI, and it’ll be cheaper and more predictable. AI earns its place when the input is messy and needs interpreting, like a rambling email, a photographed receipt or a voice note.

Step 4: Set simple ground rules

Before anyone pastes customer information into a chatbot, decide what’s allowed. It takes an hour, and our one-page AI policy template does most of the work. If you serve customers in Texas, read our summary of the rules that apply as well.

Step 5: Measure it honestly

Before you begin, note how long the task takes today. After a month, compare that with the time it takes now, including the time spent checking results. If there’s no real saving, stop. Dropping a tool that didn’t work is part of doing this properly.

Let’s talk.

Tell us a little about your business and what’s slowing you down. A real person reads every message.