Where Should AI Go First in a Small Business?

Most small businesses start with AI the same way: someone buys licenses, a few people try them, and a year later nobody can say whether it paid. The problem is not the tool. It is that nobody named the process first.

The better first question is not "which AI should we buy?" It is "where is our growth creating friction?"

Start with one sentence

Find one place where work is growing faster than your ability to handle it. An inbox that is always behind. Quotes that wait. The same employee questions answered every week. A report someone rebuilds by hand every Monday.

Write it down in one sentence, along with who feels it most. That sentence is your first process.

Then make a list, not a choice

One process is a starting point, not a decision. Before you commit, list every recurring process your company runs - work with a trigger and an outcome someone would miss. Aim for at least twenty rows. Most owners are surprised how much of the week is spent moving information from one place to another.

Take the decisions out before you score

Before scoring anything, remove the consequential human decisions: hiring, discipline, payment approvals, price concessions, contract commitments. These are not poor candidates for AI. They are not candidates at all.

What remains is usually plenty. Nobody wants AI approving a supplier payment, but reading the amount and purchase order number off a freight invoice is dull, frequent, and checkable.

Score value and readiness separately

The book's AI Opportunity Score rates each process from 1 to 5 on seven factors, in two halves:

• Value - repetition, time, cost, and impact. Is this worth improving at all?
• Readiness - feasibility, data readiness, and risk. Can AI improve it safely and soon?

Keep the halves apart. A process can be valuable but blocked by messy data, or easy but barely worth it. Plotting both gives you four groups: quick wins, strategic foundations, a watchlist, and do-not-automate.

Two habits keep the scoring honest. Write the evidence behind every 4 and 5 ("about 30 supplier invoices a week, from last month's records," not "happens constantly"). And when torn between two numbers, pick the lower one.

The frequent beats the exciting

The most common mistake is scoring the exciting process instead of the frequent one. Meeting follow-through, invoice reading, and message triage rarely make a keynote, but they happen dozens of times a month. That repetition is what turns a small improvement into real recovered capacity.

Your first AI project should be boring, frequent, checkable, and owned by someone who wants it fixed. Pick that, and you have a pilot. Pick the exciting one, and you have a demo.

The full scorecard, with the factor guide and evidence grids, is a free download on the Resources page.