Somebody in marketing signed up for an AI writing tool. The bookkeeper is trying something that reads invoices. The front desk wants a phone agent, and the owner has a Copilot license they opened twice. Six months later there are five subscriptions, nobody can say which one is working, and the general feeling in the building is that AI was oversold.
It usually wasn’t oversold. It was spread too thin. The businesses getting real value out of AI right now mostly did one thing first, did it properly, and only then moved on to the next.
Why This Keeps Happening
I recently sat in on a session with the AI and customer experience team at one of the larger technology distributors I work with. Their point was simple. AI used to be bought the way phone systems were bought: one department, one budget, one decision. That has changed. Now sales wants it, marketing wants it, finance wants it, and operations wants it, all at once and all for different reasons.
At larger companies that shows up as committees. More people in the room, more opinions, and buying decisions stretching out to six or nine months.
At a small business it shows up the other way around. There is no committee, so everyone buys their own. They call it AI sprawl, and it costs more than it looks like it should.
What Sprawl Actually Costs
- Overlapping subscriptions. Three tools that each do some version of writing help, billed to three different cards, renewing on three different dates.
- Company data in places you never approved. Every tool someone signed up for on their own is another place your customer list or pricing might now live.
- Nothing gets measured. When there are five experiments and no owner, none of them gets a before-and-after number. So none of them can prove anything.
- The good idea dies with the bad ones. One of those five might be saving real hours. It gets canceled alongside the others when the owner decides AI isn’t working.
How to Pick the First One
The right first project is rarely the most exciting one. It is the one you can prove. Look for a task that checks most of these boxes:
- It happens a lot. Dozens or hundreds of times a month, not twice a quarter.
- It follows a pattern. Same questions, same forms, same steps. AI handles repetition well and judgment calls poorly.
- You can count it today. Calls missed, hours spent, days to close an invoice. If you can’t measure it before, you won’t be able to show a difference after.
- A mistake is cheap. Something a person can catch and fix, rather than something that goes straight to a customer or a regulator.
- The information it needs is in decent shape. If the job depends on a customer list that is half duplicates, fix the list first or pick a different job.
- One person owns it. Someone whose job gets easier if it works, and who will notice if it doesn’t.
Good First Candidates
For most small and mid-sized businesses I talk to, the first win comes from one of a short list:
- Answering calls you are currently missing. After hours, at lunch, during the rush. An AI voice agent that books the appointment or takes the message is easy to measure, because you already know roughly how many calls go to voicemail. I wrote about how these work in AI voice agents for small businesses.
- Invoice and receipt entry. Reading vendor invoices and keying them into the accounting system is repetitive, high volume, and checked by a person anyway.
- Quote and estimate follow-up. A lot of businesses send a quote and never follow up. A tool that drafts and schedules that second and third email is low risk and shows up directly in closed work.
- Answering the same customer questions. Hours, directions, order status, what’s included. Common on Oahu with tour, rental and hospitality businesses fielding the same twenty questions from visitors all day.
Notice what isn’t on the list: overhauling the whole company, replacing a department, or building a custom agent that touches every system you own. Those projects can come later. They go much better when the first one has already taught you something.
Run It Like a Test
Treat the first project as a trial with an end date.
- Write down the number before you start. Missed calls per week, hours per month on invoice entry, quotes that never got a follow-up.
- Give it ninety days. Long enough to get past the setup and the learning curve, short enough that it can’t drift.
- Decide in advance what success means. If missed calls drop by half, it stays. If nobody can tell the difference, it goes.
- Keep the rest on hold. The other ideas are not canceled, they are next. Writing them down on a list is often enough to stop the side subscriptions.
The first project also tells you something you can’t learn any other way: how ready your data and your people actually are. If the invoice tool struggles because vendor names are entered six different ways, you have found the real project. That is worth knowing before you spend on anything bigger.
Then Expand
Once the first one works, the second is easier. You know who owns the result, you have a vendor relationship that either worked or didn’t, and the team has seen AI save them time instead of creating it. That is a much better position for bigger conversations about automating more of the business, and a much better place to have them from.
How I Fit In
I am a consultant and a partner with OTG Consulting, which is a consultancy that presents the services of other providers. I do not build AI tools, and I am not tied to any one platform, so I have no reason to steer you toward a particular product.
What I help with is the choosing: going through where your business loses the most time, picking the one task most likely to pay off, and setting up the before-and-after number so you can tell whether it did. When there is a fit, I connect you with the provider whose tool handles that job well, and that provider delivers and supports it.
Sometimes the most useful part of that conversation is deciding what to cancel.
Pick One
If you had to name the single most repetitive task in your business, the one everyone complains about, that is probably your first AI project. It is also a good place to start the conversation.
Schedule a free consultation and we will work out which task to start with, what to measure, and which tools are worth keeping.