Over one long working session I took Daring Strategy (my AI products for small business venture) from "has a landing page" to a working acquisition funnel: a hosted AI product, a free lead magnet, nine industry reports, and a live Google Ads test — all deployed and capturing leads by the end of the day.

A few days later I found out it had been throwing every one of those leads away.

I'll get to that, because it turned out to be the best argument I have for the thing this piece is actually about. At every fork in this build, I tried to measure instead of guess. The funnel looked like it worked. The form returned success. The only number that told the truth was the one I made myself go check.

From "sell a prompt" to "sell a system" — and how I priced it

The original plan was to sell a single really good prompt for $49. I killed it about an hour in. Nobody pays $49 for a prompt — you paste it once, it works or it doesn't, and there's nothing to come back to.

So the product became the thing that runs the prompt for you: a guided 20-minute interview that turns a conversation into a set of structured files a business can drop into any AI — brand, voice, customers, offers, the works. Same underlying idea, except now it's a real product instead of a text file.

You're buying a system, not a prompt.

That reframe also changed how I priced it. I landed on $50, and I want to be honest about how I got there — because this is the kind of thing I'd normally walk through with someone in a live coaching call.

The cost to run one interview is about twenty cents — I'll show you how I know that below. So I'm not pricing to margin. I'm pricing to value. A business that gets a genuinely personalized AI context system out of a 20-minute session is getting something that would otherwise take hours of YouTube, trial-and-error, and piecing things together from five different sources. Time is money. For a business, $50 is not cost-prohibitive for that.

Now, I've been critical of SaaS companies on this site — the recurring lock-in model, the drip-fed features, the pricing designed to make cancellation feel like a loss. I'm aware of the tension, and I want to be upfront about it. But a one-time $50 product that gives you a permanent context system for your business is a different thing than a subscription that owns you. It's not a platform. It's a tool you paid for once and keep.

And there's no usage cap. Run it as many times as you want. I do want to find a way to prevent abuse eventually — one license getting shared across twenty businesses isn't really the spirit of it — but right now that's not a real concern, and I'm not going to punish early users with artificial limits to solve a problem I don't have yet.

The payment process is finished and it's live now at https://daringstrategy.com/business-brain.

What a run costs (I didn't guess)

At some point I wondered what one of these interviews actually costs me to run. The easy move is to estimate — ballpark the tokens, multiply, move on.

Instead I instrumented it. Logged the actual token usage from the model, ran two real interviews through the thing, and read the numbers straight off the logs.

About twenty cents a run. Twenty-two if it reads the customer's website first.

That number settled a few things at once. Cost isn't the thing I need to protect — a purchase covers multiple runs, and the economics are nowhere near tight. Any usage limits I add in the future will be there to prevent one code from being shared across twenty businesses, not to protect a quarter. Knowing the real figure meant I spent my next hour on the right problem instead of an imaginary one.

If your product is a conversation, simulate the other side of it

Here's the part I'm most proud of. The product is a 20-minute interview, and testing a 20-minute interview by hand is brutally slow. Sit through it, note what's clunky, change one line, sit through it again. A whole afternoon gets you four runs.

So I built a small army of fake customers. Agents that each role-play a different small-business owner — different talking styles, some with a website, some without — sit through the real interview end to end, then grade the files it produced. Did it capture how they actually work? Did it invent anything? Did it get weirdly biographical?

It caught things I'd have missed for days. One question read four decision-making styles out loud like a multiple-choice quiz, and every single simulated owner flagged it as leading them. The interviewer also confidently made up a business partner's name from a website. I would have shipped both.

I don't think you can test a conversational product any other way at a reasonable speed. And the grader is where the actual product spec turned out to live — writing down what a good result looks like forced me to decide what the thing is even for.

The ad test, and the "required" step that wasn't

The longest, messiest stretch was Google Ads. I put $200 behind one industry — coaching, because it's closest to my world and the clicks are cheaper — pointed at the free check rather than the paid product, on its own message-matched landing page.

Google fought me the whole way. It quietly switched me into its Performance Max product instead of plain Search; I caught it and switched back. Its recommended daily budget would have burned the whole $200 in under three days, so I capped it hard. Its suggested keywords were the wrong intent entirely — people who want to hire a consultant, not coaches who want to use AI themselves.

Then it tried to route me through a conversion-tracking setup that needed events I hadn't built, which was going to cost me another day.

This is the decision I want to flag. Because I'd chosen "maximize clicks" as the bid strategy, Google didn't actually need conversion data to run the campaign — that tracking was only for measurement. And I already measure every signup, because each lead lands in Ghost the moment they submit. So I skipped Google's tag completely and built my own: the landing page drops a small tag in the browser, the check reads it, and the signup gets labeled as coming from the ad. Cost per lead is spend divided by those labeled signups. I read it in Ghost.

The lesson I keep running into: find out what a platform's step actually requires before you do the work it tells you to. Half of "required" is "required for the path they'd prefer you take."

67 clicks, and one of them was me

The ad cleared Google's verification and traffic showed up: 6,070 impressions, 67 clicks, sixty-two cents a click. Real people, clicking a real ad, landing on the check.

Then I opened Ghost to count the leads. There was one member. It was me.

Sixty-seven people went through the funnel and not one of them got captured. The form looked fine — it submitted, it returned success, it handed back the report. Everything you'd verify by clicking through it yourself worked.

The problem was underneath. I'd wired signups through Ghost's built-in flow, which sends a confirmation email and only creates the member once they click it. But I was handing over the report the instant they submitted, so nobody had any reason to go dig that email out of their inbox. The funnel collected everyone and kept no one.

Here's how close this came to going unnoticed. If I'd trusted the green checkmark on the form, I'd have let that ad run a week and then wondered why a campaign with fine click numbers produced nothing. I caught it on the first day of real traffic for one reason: I went and counted the actual members in Ghost instead of trusting that "submitted successfully" meant "captured."

The fix was small once I understood it — create the member on submit, skip the confirmation email. For a free download where the report is the whole payoff, asking people to confirm by email just guarantees you lose almost all of them. That went out the same day.

The honest part

A one-day build invites a fair criticism, so let me make it for you: speed usually means corners cut, tools chased instead of fundamentals respected, a demo that falls over the first time a real person touches it. That's a reasonable worry, and I've shipped things that earned it.

What made this one hold up wasn't going fast. It was checking the rendered thing instead of trusting that it worked. The model invented a clinic's name out of thin air — caught because I read the output instead of assuming. A stale logo from an old concept almost went onto nine PDFs — caught because I opened the file instead of trusting the filename. My own name leaked into reports that were supposed to be about the reader — caught the same way. The lead leak is the same lesson with higher stakes: the system told me it was fine, and it wasn't.

None of that is clever. It's just looking at what you actually made, every time, before you believe it.

The speed came from the tools being ready and from measuring at each fork, so I never had to backtrack far. It did not come from skipping the boring checks. Those are the part that lets you move quickly without it quietly falling apart on you.

What's not done

The ad test has a check-in on the calendar before the budget runs out. Proper conversion tracking is on the list now that the launch pressure is off. And the simulated-customer testing approach I described above is something I'd genuinely like to compare notes on — whether you've found your own version of the leak you didn't know was there.

I suspect more of us are running funnels that quietly keep no one than would care to admit it.