How to start with AI without wasting money

How AI money usually gets wasted

Most AI spending that goes nowhere follows one of a few patterns:

• Seats nobody uses. A business buys AI licences for the whole team. A few people try it, most don’t, and the bill renews anyway.
• A pilot with no owner. “Let’s see what AI can do” becomes a few experiments, no decisions and no measurable result.
• A tool that can’t see your information. A general assistant can’t answer questions about your orders, prices or procedures if it has no access to them, so people stop asking.
• Automating a broken process. If the steps are unclear or different every time, AI makes the confusion faster.
• Buying for the demo. A slick demo on someone else’s data doesn’t tell you whether it will work on yours.

Each of these is avoidable with a bit of structure. The SBA’s own advice to small businesses is to “start small,” noting that many AI tools offer basic services free or at lower cost, so you can test whether they add value before committing. ^[1]

Step 1: List the work that eats your week

Spend 20 minutes with the people who do the work and write down the ten tasks that take the most time or cause the most frustration. Be specific:

• Not “customer service,” but “answering ‘where’s my order?’ emails.”
• Not “admin,” but “typing invoice details into the accounting system.”
• Not “sales,” but “writing follow-up emails after quotes.”

Then score each one from 1 to 3 on how often it happens, how much time it takes, and how rule-based it is. Multiply the scores and pick the top one. (There’s more on this in How AI helps small businesses.)

Step 2: Define “working” before you start

Write down, in one or two lines, what success looks like and how you’ll know. For example:

• “Replies to common customer questions go out within an hour instead of the next day.”
• “Invoice details are entered without retyping, with fewer corrections.”
• “Follow-up emails go out the same day for every quote.”

Then measure the before. Time the task for a week, or count how many times it happens and roughly how long each one takes. Without a before, you’ll never know whether the after is better.

Step 3: Try the cheapest version first

Before buying anything new, look at what you already have. Many email, office, CRM, accounting and booking tools now include AI features such as drafting, summarising and suggesting categories. Turning on a feature you already pay for is often the cheapest test there is.

If you do need a new tool, use a free or low-cost tier on a small, real slice of the work, and treat the trial like an experiment:

• One task.
• A few people.
• Real examples from your business, with sensitive information kept out (see Step 4).
• A fixed trial period, for example two to four weeks.

Step 4: Keep a person in the loop, and set simple rules

Write a short list of what the AI may and may not do on this task. For example:

• It may draft replies. A person reviews and sends them.
• It may not send anything to customers on its own, change prices or promise refunds.
• Never paste in Social Security numbers, card numbers, passwords or health information.

The SBA recommends having another person review what free AI tools produce, and trying not to feed sensitive data or proprietary information into them. ^[1] Use business accounts rather than personal ones, so you control access and can remove it when someone leaves. Our guide to AI and your business data covers this in more detail.

Step 5: Decide with real numbers

At the end of the trial, compare against your before:

• Keep it if it saved real time or improved quality, and people actually used it.
• Fix it if the idea worked but something got in the way, such as missing information, an awkward step or unclear rules.
• Stop it if it didn’t help. That’s a good result too: you learned cheaply.

Then pick the next task. Small wins add up, and each one makes the next easier because your team knows what to expect.

When it makes sense to bring in help

Off-the-shelf tools go a long way. It’s usually worth getting help when:

• The task needs AI connected to your own systems or documents, such as your inbox, CRM, order system or file storage.
• You want AI as a step inside a workflow that runs on its own (read, decide, draft, route), not just a chat window.
• You need something built, such as a portal, internal tool or assistant that answers from your information.
• Nobody inside the business has time to own the setup.

What to get in writing from any AI partner

Whoever you work with, you should get these before work starts:

1. A written plan: the task, what will be built or set up, and what “working” means.
2. A quote: what it costs and what’s included.
3. Data handling: which AI tools will touch your information, where it’s stored, and who can see it.
4. Security expectations, in writing. The FTC advises businesses to put security expectations into contracts with service providers and then verify they’re met, not just take their word for it. ^[2]
5. Ownership: who owns what gets built, and what happens if you part ways.

A one-page checklist

• Ten time-eating tasks listed and scored
• One task picked, with a clear success line
• “Before” measured for at least a week
• Existing tools checked for built-in AI features
• Trial scoped: one task, a few people, fixed end date
• Rules written: what AI may do, may not do, and what never goes in
• Business accounts, not personal ones
• Results compared and a keep/fix/stop decision made
• Next task picked