Over the past four weeks, we've covered what AI is, why you can't always trust it, what the job displacement headlines are actually saying, and what AI agents do that a chatbot doesn't. If you've made it this far, you're probably somewhere between "this is interesting but I still don't know what to do" and "I know I should do something but I don't know what." Both are reasonable places to land.

Most businesses start with the technology. They sign up for a tool, explore what it can do, and try to find problems that fit. This tends to go badly, because AI tools are general-purpose and your business is specific.

Start with the friction instead.

What friction actually looks like

Where does something break down repeatedly in your operation? What tasks sit on someone's desk not because they require that person's judgment, but simply because someone has to do them? Where does information get entered twice, moved manually, or copied from one system to another? Those are the seams in your operation, specifically the ones that make someone sigh quietly every Tuesday morning.

Friction hides well because it's normal. Nobody flags a process as broken if it's been broken since before they started working there. Say an office manager retypes numbers from a paper invoice into the accounting system every Friday afternoon. She's done it for six years. It's not a crisis, it's just Friday. Ask her what's inefficient about her week and she might not even mention it, because it stopped registering as a problem a long time ago. That's usually where the actual opportunity is hiding, not in the thing everyone already complains about, but in the thing that's been quietly accepted as the cost of doing business.

One way to find it: watch someone do their job for half a day and count how many times they stop doing the task in front of them to go get information from somewhere else. Every one of those interruptions is a seam. Not all of them are worth fixing with AI, but the list is usually longer than the owner expects, and shorter than the owner fears.

I'd also point out who tends to see the friction most clearly, and it's rarely the owner. Owners see the business from thirty thousand feet. The person doing the task every day sees exactly where it snags. If you want an honest list of where things break down, ask the people doing the work, not the person who signs off on the budget. That conversation alone, with no AI involved yet, usually surfaces more useful information than a demo from any vendor.

What a good starting point looks like

Good starting points tend to share a common shape: they're repetitive, they cost measurable time, and what "done" looks like is clear without needing years of experience to recognize it. Tasks that require relationship context, industry nuance, or judgment built over years are rarely the right place to begin, no matter how appealing the idea sounds.

A few markers I look for when I'm helping someone pick a starting project:

Say a landscaping company decides to look at its own operation this way. The obvious candidate, the one the owner keeps bringing up, is scheduling crews around weather and equipment availability. It's the most painful part of the week. It's also the worst starting point, because getting it wrong costs a missed job and an angry client, and judging whether the AI's schedule was actually good requires the same fifteen years of experience the owner already has. The better starting point turns out to be something the owner barely thinks about: turning field notes into client-facing invoices. It happens every day, everyone agrees on what a correct invoice looks like, and a mistake just means someone re-sends an email.

Six months later, once invoicing runs itself and the office manager trusts the output without checking every line, scheduling looks different. By then the business has a working relationship with what the tool gets right and what it still needs a human to catch. That's the order it usually has to happen in. Nobody starts a new employee on the hardest job in the shop either.

Why the technology is the easy part

PwC's research on this is worth sitting with. Technology delivers about 20 percent of the value in any AI initiative. The other 80 percent comes from redesigning the work around it. That's the part most businesses underestimate, and also the part that determines whether an investment compounds or just adds another subscription nobody uses six months later.

Here's what that actually means in practice. Buying the tool is a purchase order. Redesigning the work means deciding who reviews the AI's output, what happens when it's wrong, whether the old manual process stays as a backup, and who owns the decision to trust it a little more each month as it proves itself. None of that shows up on the vendor's pricing page, and none of it gets done automatically just because the software is installed.

I've watched two businesses buy the identical tool and get opposite results. One assigned an owner to the rollout, someone whose job included checking the output daily for the first month and adjusting the process based on what went wrong. The other bought the license, forwarded a login to the team, and moved on to the next priority. The first business is still using the tool a year later. The second cancelled the subscription in month four and told me AI "didn't work for them." The software was the same. What surrounded it wasn't.

The questions I get asked most

A few objections come up in almost every one of these conversations, so it's worth answering them here instead of waiting for the meeting.

"We don't have anyone technical on staff"

Good, mostly. The setup work I'm describing, cleaning up job files, deciding what "done" looks like, watching the output for a few weeks, isn't a technical skill. It's the same operational discipline that runs the rest of your business. The businesses that struggle most are usually the ones who hand the whole thing to whoever's most comfortable with computers and hope it works out, rather than the ones without a developer on the payroll.

"We tried something like this before and it didn't stick"

This is the one I hear most, and it's almost always a symptom of starting with the technology instead of the friction. Someone bought a tool because a competitor had one, rolled it out to the whole team at once, and never assigned anyone to watch whether it was actually helping. Six months later it's a line item nobody uses. That's not evidence AI doesn't work here. It's evidence the 80 percent got skipped.

"Isn't this just for bigger companies?"

The opposite, usually. A smaller operation has fewer approval layers, fewer competing priorities, and an owner who can see the whole business without asking six department heads for their opinion. The businesses I've seen move fastest on this are fifteen or twenty people, not fifteen hundred. Size isn't the obstacle. An unclear starting point is.

Where this goes sideways

The right starting point often isn't obvious from the inside. What feels like your biggest inefficiency may not be the most tractable one, and what looks like a simple handoff to AI often has more complexity underneath than it appears. Getting that read right at the beginning saves a lot of expensive backtracking later.

"But our biggest problem is obviously X, why would we start somewhere smaller?" is a fair question, and one I get almost every time. X is usually big precisely because it's tangled up with judgment, relationships, and edge cases that took years to learn. Starting there means debugging the hardest problem in your business with a tool nobody on your team has used before. That's an expensive way to learn what AI is actually good at.

I've watched this play out with a sales demo more than once. An owner sits through a pitch, gets excited about the flashiest capability on the screen, and decides that's the project. The demo was built to impress a room, not to solve your Tuesday. By the time the contract is signed, the flashy capability turns out to depend on data the business doesn't have in a usable format, or a workflow change nobody agreed to. The fix isn't to distrust every demo. It's to ask, before signing anything, whether the friction it solves is actually one of yours, or just one that looks good on a screen.

The other trap is picking something that looks small but isn't. A task that seems like straightforward data entry might actually require someone to catch inconsistent supplier codes, or reconcile two systems that were never designed to talk to each other. Those hidden dependencies don't show up until you're three weeks into the project and the timeline has quietly doubled. This is usually where an outside read helps, not because the business owner doesn't understand their own operation, but because they're too close to it to see which parts are simple and which parts only look simple.

What happens after the first one works

Starting with something small and well-defined builds the knowledge to take the next step without throwing money at the wrong problem. The businesses that get lasting value from AI tend to start with one thing that works, learn from it, and expand. The ones that lead with ambition and skip the foundation tend to have interesting stories to tell about what went sideways.

The compounding part is real, and it's the part people underestimate most. The first project teaches the team what "AI actually helping" feels like versus what "AI creating more work" feels like. That distinction is hard to explain in the abstract and obvious once you've lived through both. The second project moves faster because the review habits, the trust calibration, and the internal champion are already in place. By the third or fourth, the friction-finding itself gets easier, because the team has learned to recognize the shape of a good candidate instead of relying on one person's gut feeling.

None of this happens on a fixed schedule. Some businesses get from first project to fourth in six months. Others take two years, because the first project surfaced bigger data or process problems than expected, and fixing those came first. Both are fine outcomes. What doesn't work is skipping straight to project four.

There's also a quieter benefit that doesn't show up until later: your team stops being afraid of the technology. The first rollout, done carefully, teaches people that AI is a tool they can question, correct, and push back on, not a black box handed down from head office. That confidence is worth more than any single project's time savings, because it's what lets the next ten decisions get made faster, by people closer to the work than you are.

Where to actually look first

I wrote this series to give you enough grounding to ask better questions, not to hand you a finished plan. If you're the type to want a first move today, here's a reasonable one: pick the task in your business that's repetitive, has a clear right answer, and would embarrass nobody if it went wrong for a week while you figured it out. That's rarely the biggest problem in the building. It's usually the most boring one, sitting somewhere nobody's bothered to look because it's never been the fire that's burning loudest.

If you're ready to figure out where your actual starting point is, that's exactly the kind of conversation I have every day. The free assessment is a low-effort way to get a second set of eyes on it before you spend a dollar.

You know where to find me.