A shoe store owner called me last month. She wanted an AI chatbot on her website. She had read an article about a boutique in Austin that put one up and “doubled conversions,” and she wanted to know what it would cost to build her something similar.
I asked her what happens after a customer clicks Add to Cart on her site. She laughed and said it goes to a spreadsheet she checks at the end of the day, and about once a week she forgets to look at it until Wednesday, which means Monday’s orders don’t ship until Thursday.
The chatbot would have been a $12,000 project. Fixing the order notification problem was a $400 project. Guess which one actually moved her business.
This is the first-workflow problem, and I see it constantly.
The flashiest workflow is almost never the right one
Every small business owner I talk to wants to start their AI project on the thing that would be most fun to show their friends. The chatbot on the homepage. The AI-generated marketing copy. The custom agent that “learns your business.” I understand the appeal. That stuff is fun, and it photographs well.
It is also, in almost every case I have seen, the wrong place to start.
The reason is simple: those workflows are externally facing. If they go wrong, they go wrong in front of a customer. Which means you can’t ship them until they’re close to perfect. Which means the engagement takes longer, costs more, and the payback is uncertain because you can’t quite point to what revenue it generated.
Meanwhile there’s a folder on your desktop called Paperwork. There are three people at your company whose job is partly or mostly filling out that paperwork. You know the number of hours they spend on it. You know what it costs you per hour. And nobody sees it except your team.
That is where the first AI project belongs. Every single time.
Two filters that work
When we scope a Nano-Pilot, we run every candidate workflow through two questions. They are not clever. They are just the two questions most people forget to ask.
Can you name the humans who currently do this work?
If you can, and you can say how much of their week it eats, the workflow is eligible. If the workflow is abstract, like “improve our marketing” or “get better insights from our data,” it is not eligible for a first project. Abstract workflows don’t have a baseline. You can’t measure what you did. A workflow that has three named humans and a measurable hour count has a baseline. You can measure what happened after.
If this breaks, does the customer see it?
If yes, the project gets harder by an order of magnitude. You need guardrails, testing, oversight, rollback plans. For a first engagement, that’s the wrong risk profile. You want a workflow where if the AI gets confused, it fails quietly inside the office, a human catches it, and nobody outside the team ever knows.
Those two filters eliminate about 80 percent of the candidates most business owners come in with. What’s left tends to look boring: intake forms, recall lists, quote preparation, document categorization, insurance verification, scheduling reconciliation, inventory reorders, claim denials. The stuff on the spreadsheet nobody talks about at dinner.
That is the list we work from.
Why boring wins
A workflow you pick for a first project should do three things. It should be something you can measure before and after. It should be something where failure is recoverable and invisible. And it should be something where the person currently doing it will actually like having the help.
That last one gets underrated. If the humans doing the work hate the AI when it arrives, the project dies inside the office regardless of how good the technology is. The work being automated has to be the work they already dread. Not the work they’re proud of. Not the work that makes them feel skilled. The work they complain about.
The shoe store owner, once we got past the chatbot conversation, told me the person who checks the orders spreadsheet also hated doing it. It was her least favorite part of the week. Fixing that workflow was a gift to her, not a threat. That’s the version of this that works.
What the right first project looks like
For the dental practice we worked with earlier this year, the first workflow was insurance verification. Not the patient-facing booking flow, not the recall marketing, not any of the glamorous things. Insurance verification. The front desk was spending around 14 hours a week on it. After the engagement, that number dropped to about 3. The coordinator who used to do it now does patient experience work she actually likes.
For the HVAC company, it was dispatch triage on emergency calls. Not a chatbot on the site. Not generated marketing. The three humans working the phones were answering the same twenty questions all day to figure out whether a call was actually an emergency or a maintenance request that could wait. We wired that triage into the intake flow. Same three humans, twice the calls handled.
Neither of those projects will make good Twitter content. They are not impressive in a demo. But they paid for themselves in under ninety days, and the businesses involved now trust the AI enough to expand into harder workflows.
That is the sequence that works. Start invisible. Measure the thing nobody else is measuring. Get the win. Then do the next one.
If you’re trying to figure out where to start
Pick the three workflows at your company where a named human spends the most hours on something boring. Write the hours down. Ask them which one they would pay to never do again. That’s your Nano-Pilot candidate.
If the answer is the thing on the website you wanted to show your friends, hire a different firm. That’s a fine second project. It is not a fine first project, and the firms that will happily take your money for it on day one are the same firms you’ll be frustrated with on day ninety.
The right first workflow is the one you’re slightly embarrassed to admit you still do by hand. That’s the one that pays for itself. That’s the one that builds the trust in the rest of what AI can do for you. And it is almost never the one you came into the conversation wanting to talk about.