Where should a small business start with AI?

What you will be able to do

You will pick a first AI project that fits a local service business, run it for a month, and know whether to keep it—without boiling the ocean.

Start here, not with a software tour

The best first project is:

• Something you already do often
• Something with a clear “done”
• Something where a wrong draft is easy to catch before customers see it
• Something that eats owner or admin time

Good first jobs: quote follow-up emails, review replies, checklist drafts from notes, weekly team update summaries, vendor compare briefs.

Weak first jobs: fully autonomous customer decisions, clinical advice, unsupervised refunds, complex multi-system automation on day one.

Step 1: Name the job in one sentence

Examples:

• “Draft a same-day follow-up after we send a quote.”
• “Turn Friday voice notes into a Monday opening checklist.”
• “Rewrite angry-but-fair responses to public reviews.”

If you need a paragraph to explain the job, narrow it.

Step 2: Capture a baseline for one week

Before any new tool:

• How many times does this job happen?
• How long does it take each time?
• What does a good result look like?
• Where do mistakes happen today?

Write the baseline on a sticky note or a single doc. No spreadsheet theater required.

Step 3: Choose one general assistant

For a first test, one reputable general LLM assistant is enough. Learn the habit in one place. Specialty tools (image, voice, niche ops) come after the writing/summarizing habit sticks.

Create a tiny prompt kit:

• Voice rules for your business
• Two example good outputs you already like
• A checklist of facts you must paste every time

Step 4: Run twenty real reps

Not demos. Real work.

After each use, spend thirty seconds noting: time spent, whether you sent it, what you fixed. Patterns appear fast—usually filler phrases, missing facts, or tone that is too stiff.

Step 5: Teach one other person

When your own workflow is stable, sit with one teammate for twenty minutes:

• Show the prompt kit
• Do two live examples together
• Agree what they must never paste
• Agree that they still review before send

Adoption fails when the owner has a secret prompt and everyone else improvises.

Step 6: Keep, kill, or narrow on day 30

Ask only:

• Did total time drop for acceptable quality?
• Did customer-facing mistakes go up or down?
• Will we still use this next month without nagging?

If yes, document the recipe and consider a second job. If no, cancel and pick a different job—not a bigger software package.

A starter roadmap for the next 90 days

Days 1–30: one drafting or summarizing job
Days 31–60: one internal SOP/checklist job
Days 61–90: one decision-brief or research-assist job for owner planning

Still one primary tool if possible. Depth beats breadth.

Common false starts

• Buying a bundle of AI logos “for marketing”
• Starting with the loudest vendor demo
• Automating a broken process
• Skipping the baseline so you cannot tell if anything improved

Fix the process notes first if nobody agrees how the job works today. AI amplifies clarity and confusion equally.

How to choose among three good first jobs

If several tasks qualify, score each 1–5 on:

• Frequency (happens every week)
• Ease of review (mistakes are obvious before send)
• Owner pain (you personally feel it)
• Data safety (low sensitivity)

Start with the highest total. Ties go to the job with the easiest review. A slightly less painful job you can check quickly beats a painful job that can quietly ship wrong prices.

What “ready” looks like on day one

You are ready to start when you have:

• The one-sentence job
• A baseline sticky note
• An approved tool login
• A voice card or example outputs
• A calendar block for the day-30 review

If any of those are missing, finish them before inviting the whole team into a new habit.

Short close

One job, one tool, one metric, thirty days—that is a real start.