Offer testing case study: three rounds, and what stayed unknown

One offer, three rounds of testing, and a decision at the end. The business, the people and every number below are invented. What is real is the sequence, because it is the sequence most first offers go through, and the point of following it is to see what changed between rounds and why.

The founder in the case is a bookkeeper who wants to sell a fixed-fee month-end close to independent restaurants. She has never sold anything before.

What to take away

The offer as first written

For independent restaurant owners, we deliver clean month-end numbers within five working days, replacing the shoebox and the annual panic, for a fixed monthly fee, and here is why you should believe it: the bookkeeper has done this for restaurants for twelve years.

The last clause of that sentence is a claim, and it was one she could support, which is the standard the FTC applies to small business advertising: a reasonable basis before the claim runs rather than a defense assembled afterwards.

She underlined who, because she was unsure whether owners cared enough to pay monthly rather than once a year at tax time. The five-slot method says the underlined slot is the one being tested and the others hold still.

Round one: the page

Design. A single page with the sentence above, a price, and a button to book a fifteen-minute call. Traffic from paid advertising aimed at restaurant owners in her city. Two weeks.

Prediction, sealed. "Of roughly 300 visitors I expect six booked calls and two people who agree to a paid first month."

Result. 340 visitors. Four calls booked. Zero paid.

Reading. Before she blamed the offer she listed who the four callers were. Two were accountants asking whether she was hiring. One was a franchise manager with no authority over spending. One was an owner who wanted a one-off cleanup, not a monthly service. The advertising platform had matched "restaurant" to anyone in the industry. The who slot had not been tested at all, because the people the offer named had not arrived.

Decision. Throw the result away. Change the channel, not the offer.

Round two: direct outreach

Design. She walked into forty restaurants between two and four in the afternoon, asked for the owner, and said the sentence out loud. She asked for a fifteen-minute conversation, then, at the end of it, for a refundable deposit toward the first month.

Prediction, sealed. "Of forty, I expect to speak to fifteen owners, book eight conversations, and take two deposits."

Result. Twenty-two owners reached. Nine conversations. Two deposits. Five owners said the price was fine but they already had "a guy". Two said the price was too high against what the guy charged.

Reading. The offer sentence worked when said to the right person. The substitute was a part-time bookkeeper, usually a relative or a friend, whose fee was low and whose output was late. The two deposits came from owners whose guy had recently quit. So the who was narrower than "restaurant owners": it was owners whose current arrangement had just failed.

Decision. Continue. Rewrite the who slot to "independent restaurant owners between bookkeepers" and put that sentence in front of the next group. Also note the price signal for later; two "too expensive" against a named substitute is a reading to follow up in a proper price test, not something to act on yet.

Round three: the narrowed offer, and a cap

Design. Twenty more restaurants, chosen by asking suppliers which owners had recently lost a bookkeeper. Same script. This time she asked for a paid first month, not a deposit, with a stated start date and a refund if the numbers were not delivered within five working days. Where money is taken before delivery, that window is an obligation rather than an aspiration, and the order and shipment rule sets out the notice and refund a seller owes when it cannot be met.

Prediction, sealed. "Of twenty, I expect ten conversations and three paid first months."

Result. Twelve conversations. Three paid. One asked for weekly reporting instead of monthly, which she declined and wrote down.

Reading. Three strangers paid a stated price for a stated outcome, having been told plainly that she would be doing the work herself. The number was small and she did not turn it into a rate. She asked all three why they had said yes; all three named the five-day promise, not the fee.

Decision. Five customers total, delivered by hand for three months, no new sales meanwhile, with a written stop rule: if she missed the five-day promise for two customers in the same month, she would pause sales and fix the delivery before taking anyone else. The cap and the stop rule came from the minimum offer discipline of designing for ten rather than one.

What the three rounds cost

Two weeks and a small advertising budget for round one. Four afternoons for round two. Three afternoons and some phone calls to suppliers for round three. No software was built. The offer sentence changed once, in one slot, and each change had a written reason.

What was still unknown

Whether the price held for owners who were not in a crisis. Whether delivery survived at ten customers. Whether "between bookkeepers" was a segment she could find repeatedly without walking the streets. Those became the next three tests, in that order, and the third one is a launch question rather than an offer question.

Common questions

Why did she not fix the page after round one?

Because the page had not been shown to the right people, so nothing about it had been tested. Fixing it would have meant iterating on a result that contained no information.

Was walking in unrealistic?

It was the channel she could control and observe. The point of round two was to test the offer on the right who at any cost; whether the channel could be repeated was a later question, and she treated it as one.

Should she have taken the deposits in round two as proof?

Two deposits from owners in a crisis proved that the sentence was legible and the price was not absurd for that situation. She wrote down what it did not prove, and the third round was designed around that list.

What would have made her stop?

A round three with twelve conversations and no payment. The prediction named three; one would have been a pause and a set of conversations about why; zero would have sent her back to the substitute question.