Most small businesses approach AI in the wrong order. They see a demonstration, buy a subscription, and then look for a problem it solves. The predictable result is a tool nobody opens and a recurring charge nobody can defend at renewal.
Here is the order that works. Answer these twelve questions honestly before you spend anything. If you cannot answer most of them, you are not ready to buy — and that is useful information that costs nothing.
Part one: the work
1. What task does someone in your business do more than ten times a week?
Volume is the first condition. Automating something that happens twice a month almost never pays for the effort of automating it. Write down the three highest-frequency repeated tasks in your business. Ask the people doing them, not the people managing them — the answers differ more than owners expect.
2. Is it substantially the same each time?
Repetition is the second condition. A task that varies wildly case by case is a poor candidate regardless of how often it happens. Look for the shape: similar inputs, similar steps, similar outputs.
3. Could you write down the rules?
If you cannot describe the decision in writing to a competent new hire, you cannot describe it to software either. This question kills more candidate projects than any other, and it kills them cheaply — on paper, before anyone has spent money.
4. What does it currently cost you?
Hours per week multiplied by loaded hourly cost. Be honest about the loaded figure. If the answer is under two hours a week, be sceptical of any solution priced above a few hundred dollars.
Part two: the data
5. Where does the information live?
Any AI system needs access to the relevant information. If your pricing lives in one person's head and three inconsistent spreadsheets, the first project is not AI. It is writing the pricing down.
6. Is the data consistent enough to be useful?
Duplicate customer records, inconsistent field usage, a CRM where half the team invented their own conventions. This is the most common blocker we find, and the least exciting to fix. Fix it anyway — everything downstream is faster and better afterwards.
7. What in this data is sensitive, and who says so?
Client confidential material, protected health information, financial records, anything covered by a contractual confidentiality obligation. List it before you choose tools, because it determines which tools are permissible. Doing this after selecting a vendor is how businesses end up with an uncomfortable conversation.
8. Do your systems have a way to connect?
API, export, integration, or at minimum a scheduled file drop. Most modern software does. Some older systems do not, and finding out early changes the project.
Part three: the organisation
9. Who owns this internally?
Not the vendor — you. A named person with the authority to make decisions and the time to actually engage. Projects without an internal owner stall around week five, every time.
10. What is your team's actual current AI use?
Ask, without consequences attached. In our experience the honest answer in almost every business is “more than leadership thinks.” That is not a problem to punish, it is a starting point — and it tells you which workflows people already want help with.
11. Do you have a written policy about what may go into these tools?
If not, your staff are making individual judgment calls with your clients' information right now. A two-page policy is a weekend of work and removes an entire category of risk. See AI policy and governance.
12. How will you know whether it worked?
Define the metric before you build: hours saved per week, response time, error rate, throughput. “It feels faster” is not a metric, and it is what you will be left with if you skip this question. Agreeing the measure upfront also keeps your vendor honest — including us.
Scoring yourself
- Ten or more confident answers: you are ready to scope a specific project.
- Six to nine: you are ready for a structured assessment — the gaps are the assessment's job to close.
- Fewer than six: start with data hygiene and a written policy. That work is unglamorous, cheap, and makes everything after it work better.
If you want a second opinion on your answers, that is exactly what our free thirty-minute call is for. We will tell you which of the twelve you have actually answered and which you have skipped past. Call and book a time.