How easy are you to do business with?
We're buying a house in San Diego, and we're handling it all from a distance. So last week we lined up contractors to come look at the place. Flooring, some carpenter work, the stuff you sort out before you move in.
Since we can't be there in person for most of it, how someone handles the back and forth tells me a lot before they ever pick up a tool.
Three contractors came to us through our realtor. All three had the same house to look at and the same shot at the job.
Here's how it went.
The first guy moved his own schedule around so he could be at the house during the inspection. He took measurements and sat with us to talk through the work. Then he followed up that same day. Easy. I knew where he stood and I knew what came next.
The second one showed up too. We had a nice talk. And then quiet. It's been five days and we still don't have a quote. Am I supposed to chase him now?
The third one never even got that far. The realtor made the intro and he just never wrote back. As far as I know he never came to the house at all. If he did, I never heard about it.
So I keep pondering one question. How easy are you making it for people to do business with you?
We haven't closed on the house yet, so I wasn't cutting anybody a check this week. But I'm about as close to picking someone as a person can get. I was ready to say yes. Two of them made me work for the chance, and when the time comes, I'm probably going to spend that money somewhere else.
The same lesson, from the other chair
That same week I was on the other side of the counter, and I got a good look at my own version of this.
I've been doing Duolingo for years and I wasn't getting where I wanted. So I signed up for a different app and paid for the full year. Right away it kept telling me I hadn't paid. That one was on me. I'd signed up under a different email than the one showing in my profile. My own mix-up.
Then I went to cancel Duolingo the same day, and I still got charged. When I opened the app to sort it out, there was nowhere to ask about a billing question. No help section at all. I found out later there is one on the desktop version, but I learn on my phone, so I never touch the desktop. On the one door I actually use, there was no handle.
So, before I go further, some of this was on me, and none of us are immune to it.
I have no doubt I've cost myself a customer somewhere without ever knowing it. This isn't me claiming to be perfect. I'm just noticing how often the path to hand someone money, or fix a small problem, is harder than it needs to be.
Nobody woke up trying to lose a customer
Before I make anybody the bad guy here, let me climb into their day for a minute.
Take that second contractor. He's up early, with three job sites before noon, a supply run in the middle, and a customer at the second stop who's upset about something that isn't his fault. The day is all measuring and hauling and driving from stop to stop.
By the time he sits down, it's nine at night, his back hurts, and there are four quotes waiting to be written. Ours is one of them. He means to get to it, and the man wants the work.
But the quote is the part that always slides, because the day ate the time he needed to write it. So it sits. And a good customer, one who was ready to pay, just moves on. He never sees the money he lost, because he never sees the customer walk away.
Now Duolingo. Bad Bunny plays the Super Bowl halftime, half the country decides this is the year they finally learn a language, and sign-ups pour in. That's a great problem to have.
But every one of those new people brings a question, like a charge that looks off or a cancel that didn't stick, like mine. Maybe support is buried under a wave they never staffed for. Or maybe the help section just never made it into the phone app the way it did on desktop.
I don't know which, and I won't guess. What I know is that when I went looking for help where I actually was, there was no door. Nobody there is a villain. Something in how the work is set up is costing them customers.
It's an operations issue before it's an AI issue
This is where my work comes in, and I want to be clear about what it is. It isn't only an AI thing. Both of these are operations problems first.
When I step into a business, I don't start with tools. I look across the whole Trifecta, the operations, the marketing, and the people, all at once, watching how the work really moves, talking to the folks doing it, and reading the numbers.
Losing customers at the door is one thread I can pull, and there are others. The point is to find the actual problem before you fix anything, because what they came in worried about is often not what's hurting them.
A lot of what I find gets fixed with plain, stronger operations and no AI at all. Here's what I mean.
On the contractor's side, it might turn out nobody ever decided whose job the quote is. He assumes the office will send it, the office assumes he will, and it drops in the crack between them. No tool fixes that. You just need one clear rule about who owns that step.
Then, for the pieces that repeat the same way over and over, AI becomes one good tool for the job.
Where AI comes in, and what a playbook even is
So how do you know which pieces can be handed off? You look for the work that repeats.
The contractor might argue with me here. Every job is custom. The flooring and the layout change from one house to the next, and so do the needs. No two projects are the same.
And that's true. But the way he quotes them is the same every single time. He measures the space, checks the material against his pricing, and adds his labor. Then he drops it into the format he always uses and sends it.
The details change from job to job, but the steps never do. That repeating shape is what lets you hand a piece of it off.
Once you can see the shape, you write it down as a playbook.
If the word is new to you, a playbook is the written version of a job you do over and over. It spells out what kicks the job off, what you need on hand to start, the steps in order, what you hand back at the end, and what good looks like. All of it on paper instead of trapped in your head.
This way of thinking comes from Rachel Woods, who teaches it in her AI Operator work, and she's glad to have it passed on.
For the contractor, that part's simple. A finished measurement kicks it off, his price list and the room specs are what he needs, and out comes the quote, the kind he'd be glad to send. Once it's written down, AI can pick it up and run it, and a job that used to sit on him comes off his plate for good.
There's a catch, and it's the part most people miss. The steps are the easy part. What takes actual work is writing down the thinking behind them, the way he decides in the moment.
A playbook can hold every right step and still miss that judgment, and then you get the flat, generic version that sounds like everybody else.
This is where that word context comes in, the one everybody's throwing around about AI. Context is everything the tool can see while it works, sometimes called its context window, and it only knows what you put in front of it.
Hand it the steps alone and it guesses at the rest. Give it your judgment too, your price rules and your sense of what good looks like, and it runs the work the way you would.
And the finished tool isn't always a chatbot. How you build it comes down to a couple of plain questions.
Does something kick the job off on its own, or does a person decide when it's time? A quote that should fire the moment measurements are logged can run as an automation in the background. Something a person chooses to start, and that leans on judgment each time, works better as a chat-based playbook that walks them through it.
The form follows the work.
How you build one without it swallowing your month
The fear I hear most is that this turns into some giant, six-month project. It doesn't have to, and it shouldn't.
There's a way to build one that stays small, and it has a name. It's called the CRAFT cycle, five stages that carry you from a messy problem to a working tool. Here's each one, using that same-day quote as the example.
C is for Clear Picture. First you get clear on the true problem, the actual spot where he's losing work. That means sitting with the contractor, watching how he really prices a job, and pulling a few quotes he's proud of. The prize here is how he decides, the judgment behind each step, because that's the part that's easy to lose.
R is for Realistic Design. Next you figure out how the thing will actually get built, and you draw a hard line around it. You decide what's in and what's out, and anything that doesn't fit goes on a list for later. Then you get a yes on that scope before a single thing gets built, so the project can't balloon into something else.
A is for AI-ify and Automate. Now you build it. You hand the playbook to the AI, then test what it gives back against the example quotes from stage one. Where the draft misses, you fix it, until the tool turns out a quote he'd be glad to send.
F is for Feedback. Then it goes in front of the person who does the job every day. He runs it on the next few quotes and tells you where it's wrong. You fix the things that matter most, the ones standing between this and something he trusts, and you grade how it's doing.
T is for Team Rollout. Last, you roll it out so everyone who needs it can use it. A tool only you understand isn't much of a tool. The quote now goes out the same day, every time, no matter who starts it or how tired anyone is at nine at night.
The whole cycle runs on an appetite, a set amount of time you're willing to spend, not an open tab. For a piece like this, that might be a few weeks start to finish.
You keep it to one thing so you actually finish it, then build the next one on top. Done is better than perfect. A small win you can use beats a perfect plan you never ship.
The goal is small
Find the one or two moments where good people keep losing customers, and close them, one at a time.
For Duolingo, maybe it's one clear way to reach a human about a bill, right there on the phone where the customer already is. The contractor's is the same-day quote.
They're small fixes, and they're the difference between the person you say yes to and the two you walk away from.
So this weekend, be your own customer. Buy something from yourself. Ask yourself for help and time how long it takes. Then try to cancel and see if you get charged anyway.
Whatever made you sigh is the first thing to fix.
-
Because the know-how lives in your head, not on paper. You are the only one who knows how the important work really gets done, so it all routes back through you. The way out is to get that thinking out of your head and written down, step by step, so someone else, or an AI tool, can run the task the way you would. That is what starts to pull you out of the middle of everything.
-
Start with the tasks you repeat every week, the ones only you seem to get right. Write down the steps and what good looks like, so the work lives on paper instead of in your head. That written version is a playbook, and once it exists, another person or an AI tool can run the task without you in the middle of it. Then have someone outside check that it holds up in real use.
-
Pick one task that eats your time every week. Get the thinking behind it out of your head and onto paper first, step by step, including how you decide. Then build a small AI tool around that playbook, a chat-based helper or an automation, and test it against a few examples of your own work before you lean on it. Starting small and checking your work beats trying to hand your whole business to AI at once.
-
Most of the time the tool is fine. What is missing is the setup. If you hand AI a task without writing down how you actually do it, you get a generic result that does not fit your business, or sounds like every other AI post out there. When you write down your own steps first, and have someone outside check the setup, AI has something solid to work from and the results start to fit.
-
Generic output comes from generic input. If you hand AI a task with no sense of how you think or what good looks like to you, you get the same flat result everyone else gets. Write down your own steps and your own judgment first, feed the AI that, and check what comes back against examples of your own work. The more of your thinking it has to work from, the more it sounds like you.