Home/Guides/First steps

Where do I start with AI in my small business?

The short answer

Start with the work nobody sees. List the tasks your people do the same way every week while they're not talking to a customer, pick the one they hate most, and ask whether it should exist at all before you automate it. Then build that one thing, connecting the tools you already use, and take a sec to breathe with the hours it gives back. A comprehensive AI strategy can wait until one workflow is running automatically.

/ 01The Long Answer

Why the first project matters more than the first tool

Most owners who get stuck with AI picked a tool that somebody showed them, played with it for a week, and ended up with a subscription that feels like another to-do item. Instead, when AI gives you real hours back, the team stops being nervous about the whole idea, and then we can begin real AI transformation. A chatbot that takes more time from you does the opposite.

The question we ask on every first call

"Where are your people spending time where they're not visible to anyone?" Go department by department and ask two things: what are you doing internally that takes you time and nobody sees, and what's the thing you hate doing?

Write those down for your own business. You'll get a list of ten or fifteen things: re-keying data from one system into another, chasing documents, assembling the same report every month, answering the same questions, scheduling, sending the invoice and then the reminder, and so on.

Three checks before anything gets built

For each task on your list, ask whether it happens every week, whether you could write the steps down for a new hire, and whether the information it needs is in software, like your inbox, a CRM or a spreadsheet. A task that passes all three is ready for automation. A task that needs judgment every single time still needs a "human in the loop", and that's fine, because we can automate all the preparation and follow-up around it.

Before you automate it, try to delete it

This is the step most people skip, and it's the one I'd fight for. Automating a bad process just gives you a faster bad process. We work through a short sequence on every build: Challenge whether the task is really required, Remove every step you can, Improve what's left, Speed it up, and only then Program it. We call it CRISP.

When I took over operations at an importer in New York, the project tracking system was printed paper taped to a board and marked up with a Sharpie. Nobody automated anything. We moved the whole thing to a digital board, and compliance and visibility into every job improved from that change alone. The process had to be right before the software could be, and systems should be designed to evolve over time.

Pick one, build it, and let it pay for the next one

Resist starting three things at once. One task running on its own is worth more than three half-built ones, and you can build on the first success. A recent example: a firearms training academy was taking class bookings by text, then the owner typed each student into a spreadsheet and sent the invoice by hand. Now the form fills the sheet and the confirmation and reminder emails send themselves, and the owner's next question was: what else could work that way?

What if it seems like we can't automate anything?

Sometimes the honest answer is that every task depends on judgment, or the information lives in two people's heads. Then the first project is writing the steps down for the two or three processes the business depends on. This step is unglamorous but necessary, and it can enable us to build some amazing things on top - while increasing the value of your business.

If it sounds tricky, don't worry. We have some tricks of the trade to speed up the process.

What we'd do about it

This is the exercise we run in the free consultation. You show us where your week goes, we score the tasks with you, and you leave with a prioritized list of what to automate first, whether you hire us or not. If the right first step is a simple fix you can do yourself, we'll say so. If you'd rather see it than read about it, our case studies show what a first build looked like for other owners.

/ 02Questions That Usually Follow

Should I start with the biggest problem or the easiest one?

Neither, usually. Start with the task that happens most often, follows rules, and has its data in a tool already. That's where the hours are and the risk is low. The biggest problem is often the one that needs judgment, and the easiest is often not worth automating.

Do I need clean data before I start?

No. Finding out how messy the data really is happens during the first build, and cleaning only what that build needs is cheaper than cleaning everything first.

Is ChatGPT enough to start with?

It's a good way to find out what your team would automate, because the prompts they run every week are the candidates. It won't run anything on its own, and copying its answers into another system by hand is the job you're trying to get rid of.

How long does the first build take?

One task, in the tools you already use, can be a day of build time. A system that touches several tools and several people is a week or two.

/ 04Start Here

Ask us about your own business.

What questions do you have about AI that you are too afraid to ask? We'll give an impartial look into what's possible with automating your business, pointing out the pitfalls, and why you haven't been able to get the results from AI that you want yet.

joe@arowise.com