Identify an AI opportunity in a business you know nothing about, using only first principles.

Fennrose Labs builds AI tools for local service businesses. Nyamekye Bakhtiari, a senior AI PM there, is running Tigist Marasigan's final interview. Nyamekye says one line: Dapplecoat Grooming Co, three locations, pet grooming, you know nothing about it. Find an AI opportunity, right now. Saphira Kwapong runs the front desk at Dapplecoat's busiest location and has asked the same two questions on every phone booking for four years.

How to answer this, stage by stage

Nobody is grading whether Tigist happens to know pet grooming. They're grading whether she can build a real chain of reasoning, out loud, from a business name and nothing else, and catch herself before she guesses.

Let's learn

Dapplecoat Grooming Co runs an online booking form where a customer picks a groomer and a time, the same way you'd book a haircut. It also takes phone bookings, handled by Saphira, at its busiest location.

This is the judgment call the online form quietly has no way to make. Not a lack of skill from Saphira. A lack of anywhere to put it.

Two years ago, about a fifth of bookings came through the online form. Saphira's phone questions covered almost everyone else. As the form got easier to use, that share climbed past a third, then past half. It now sits at 65 percent of all bookings, and none of those ever get Saphira's two questions.

An overrun of more than twenty minutes doesn't just cost one dog's appointment. On any day with three or more unflagged hard grooms, an average of four later appointments get pushed back or rebooked. Across one recent quarter, of roughly 280 written reviews, 38 percent of the negative ones used a word like late, rushed, or waited.

The nearest sounding comparison and the wrong one. A table doesn't change shape depending on who sits at it. A dog's coat does.

At its worst, this costs more than a few refunds. Dapplecoat's online form is the main way it's been growing, and 65 percent of new bookings already pass through it with no judgment call attached. A slow, quiet run of bad afternoons at one location was starting to show up in its own review score, right as the owners were weighing whether to open a fourth location.

The actual thing Tigist would build. Not a chatbot. A guess about difficulty, made at the exact moment it can still change the schedule.

What I would leave alone: a bath only booking, or a nail trim, barely varies from dog to dog. Running a difficulty guess on those adds a question with almost no payoff behind it. Leave them on the flat default.

A newer, trendier mixed breed the model has seen less of is exactly where this should fail safe, not fail quietly.

The lesson: a business you know nothing about isn't actually a mystery. It's a set of constraints you can reason through, if you ask what varies and who's already judging that by hand. The flashy guess and the borrowed comparison both skip that question. That's exactly why they're both usually wrong, and it's a trap you fall into right when you feel like you need to sound sure of yourself fast.

Now here is the same thing as a story

The short version above is what you'd actually say in the room. Read this one for why the two wrong guesses would have been so easy to make, and why the honest question was harder.

Saphira Kwapong has worked Dapplecoat's front desk for four years. Ask her two questions and she already knows which groomer to book. "When was he last fully brushed out?" "Does he mat up between grooms, or is he pretty easy?" She learned which of the fourteen groomers had the patience for a bad de-matting job and which ones would rather not, and she routed around it without writing any of it down.

The online form arrived a little over a year and a half ago. For the first few months it mostly picked up simple, easy bookings, a bath here, a quick trim there, and Saphira barely noticed it. She still fielded most of the harder calls herself.

Then the share crept. A fifth of bookings, then a third, then past half, without anyone deciding that on purpose. Nobody turned off Saphira's question. It just stopped being asked, one booking at a time, as more of them skipped the phone entirely.

The trigger was small. One Tuesday, three badly matted dogs landed back to back on the same groomer's calendar, all booked online, all given the same flat 45 minutes. Nobody had flagged any of them, because nobody had asked. By early afternoon four other families were sitting in the lobby, an hour past their own slot.

Saphira couldn't fix that afternoon after the fact. What she could do, and did, was start padding every phone booking with fifteen extra minutes, just in case, whether or not the dog actually needed it. It was the sensible move available to her. It also quietly wasted real groomer time on ordinary dogs that never needed the buffer at all.

Months later, in a different room, Nyamekye Bakhtiari handed Tigist Marasigan one line: Dapplecoat Grooming Co, three locations, pet grooming, you know nothing about it. The short version of what came next is above, the reasoning about a restaurant table, then the real question about who judges difficulty today, then the anchor.

Five short questions, in order, built her whole answer. She never once claimed to know how pet grooming actually works.

Nyamekye let her finish, then said the part Tigist hadn't been told: Dapplecoat was a real Fennrose account, and its real intake notes were sitting behind the question the whole time. A year and a half earlier, in a sprint planning meeting, a five person team had picked the flat 45 minute default because modeling variable time felt like more than a first version needed. Sensible, at the time, when most bookings still passed through Saphira's phone.

Fennrose had already quietly piloted something close to Tigist's answer with Dapplecoat, six weeks before this interview, to see if the idea would actually hold up. Before it started, the busiest location was averaging four rescheduled appointments a day from overruns. By week two that was down to about 2.6. By week four, 1.6. By week six, it had settled at about 1 a day, and "ran late" complaints over the same stretch had roughly halved.

One design assumed every dog was the same size problem. The other let the booking form ask the one question Saphira had been asking by ear the whole time, just at a moment when it could still change the schedule instead of only explaining it afterward.

What Tigist would say to herself, walking back into that interview room: the point was never to already know pet grooming. It was to trust that the business would tell her what mattered, if she asked the kind of question that finds it, instead of the kind that just sounds good in the room.

SPARK, five moves for a business you've never seenNot a script for sounding thorough cold. SPARK is what forces the one concrete anchor out, and proves it survives the first time the guess is wrong.

The line Tigist held under pressure. Not because the chatbot is a bad idea forever, because nothing in the reasoning pointed at it.

The recap, one line per letter: situation is a business name and nothing else, payoff is one repeatable question asked before any feature, anchor is a difficulty and time guess made at booking and routed to the right groomer, risk is guessing low costing a whole afternoon so the model leans cautious, and keep out draws the line at claimed knowledge and at picking the idea that just sounds better.

And if you want to be sure it really works, try it somewhere elseSame five letters, a translation agency instead of a grooming salon. This time the pattern hiding in plain sight isn't a matted coat, it's a document nobody's routed to the right person.

Quillbridge Linguistics runs a self serve upload form: a client submits a document and picks a turnaround time. A few weeks after Tigist's loop, Nyamekye ran the same blind prompt on another candidate, Ferdows Chikuma, who has never worked in translation. Mapped onto SPARK: situation is one line, Quillbridge Linguistics, document translation, nothing else. Payoff is the same question asked again, what's the highest volume, repeated judgment call here. Ferdows reasons that a project manager today must be guessing which of dozens of freelance translators actually knows a given document's subject, medical, legal, marketing, and how long a genuinely dense one will really take, the same way Saphira judged coat difficulty by ear. The borrowed comparison he catches himself reaching for is a ride hailing app, nearest driver gets the job, and he sets it aside the same way Tigist set aside the restaurant: nearest available doesn't matter here, subject fit does, since a mistranslated medical document is a different order of risk than a late one. The anchor: at intake, a model reads the document and estimates its subject area, its real complexity, and an honest turnaround, then routes it to a translator with both the right expertise and the open time. Risk: guess the subject wrong and a document lands with someone who doesn't know the terms, so uncertain matches get flagged for a person instead of auto assigned. Keep out: no claim about how translation agencies actually price rush jobs, something Ferdows plainly doesn't know and says so.

Same five questions, a document instead of a dog. The method didn't change. Only what it found did.

Swap the trigger and it still runs.
Speed: an interviewer caps you at sixty seconds. Skip straight to the anchor, the difficulty guess and the route.
Cost: no budget to build the photo path at all. Ship the three question fallback alone first, it still beats a flat default, and add the photo later.
The model got better, for real: say a future version reads photos far more accurately than today's. The method still matters, because "what's the highest volume judgment call here" is a question about the business, not about how good any one model happens to be that quarter.

Where people run it wrong.
They reach for the first idea that sounds like AI, a chatbot, a filter, before asking what's actually broken.
They borrow a comparison from a business they do know, and never say the comparison out loud to check it against what they can actually reason about.
They treat "I don't know this industry" as a reason to stall, instead of a reason to reason from constraints they can see.

How to use it live. The moment you're handed a blind business, don't answer yet. Ask yourself one thing first: what's the one task here that probably happens hundreds of times a week, where a person has to size up something that varies. Naming that, out loud, buys you real thinking time and gives the interviewer something concrete before you've said the actual answer.

Flashcards (tap any card to flip it)

Check yourself Score: 0 / 0