MKT-10 · SEC. 08 Market Research

Pre-Registered Demand Test: Commit the Pass/Fail Numbers First

Design a one-week demand test where the kill threshold is written down before you see a single result.

FORMAT: goal
DIFFICULTY: beginner
TIME: 15 min
TOOLS: universal
MODELS: any
COPIES: 0 so far

When to use this

You're about to run a landing page, waitlist, or fake-door test for a product idea, and you know from experience that once the numbers come in, you'll rationalize whatever they say. You want the pass/fail line committed before launch.

The pattern

Pastes as plain text

Act as an experiment designer. My product idea and the audience I can
realistically reach this week are on the two labeled lines at the bottom
of this message.

Your goal: a one-page demand test plan I can run in 7 days or less, with
pass/fail thresholds committed BEFORE launch. The plan is done when every
item below is filled in with a specific number or a specific artifact,
no "it depends" left anywhere.

1. TEST FORMAT: pick ONE cheapest-sufficient format for this idea (landing
   page + waitlist, fake-door button in an existing product, concierge
   offer in a community, pre-order link). Say why the others are overkill
   or under-powered for this specific idea.

2. THE COSTLY ACTION: define what a visitor must do for it to count as
   demand. An email address is weak, pick the most costly action this
   format allows (a card-required preorder, a booked call, a filled intake
   form) and say what each level of costliness would prove.

3. PRE-REGISTERED THRESHOLDS: given the audience size I can reach, commit
   three numbers now:
   - KILL: below this conversion, I stop and don't relitigate
   - CONTINUE: above this, I run the next bigger test
   - ZONE OF SELF-DECEPTION: the range in between, and what single
     follow-up question decides it (e.g. 5 replies to a "why did you sign
     up" email)
   Base these on realistic conversion norms for the chosen format, and say
   which norm you're using. If I could talk myself into calling any result
   a win, the thresholds are too soft, tighten them.

4. CONTAMINATION LIST: 3 ways this test could false-positive (friends
   clicking, a misleading headline overselling, traffic from people who
   can't buy) and one concrete guard for each.

5. THE COPY: draft the actual headline, subhead, and call-to-action button
   text for the test. Plain claims only, nothing the real product wouldn't
   deliver, an oversold test validates the wrong product.

End with the plan as a checklist I can execute Monday morning, in order,
with the three committed numbers at the top.

My product idea:
Audience I can reach this week (channel and rough size):

Real example output

COMMITTED NUMBERS (audience: ~2,000 newsletter readers, indie Shopify store
owners)
KILL: under 1.5% click-to-intake (fewer than 8 completed intake forms per
~550 expected visitors). Stop. Do not rerun with "better copy."
CONTINUE: 4%+ completed intake forms (22+). Run a $200 paid-traffic
replication next.
SELF-DECEPTION ZONE: 1.5-4%. Decider: email all signups "what would you
expect this to cost per month?" - 5+ replies naming a number = continue,
mostly silence = kill.

TEST: landing page + intake form (not just email: form asks store URL and
current monthly returns volume, a 2-minute cost that filters tourists).

CONTAMINATION GUARDS: no personal-network shares until day 3; headline says
"reduce return fraud" not "eliminate returns"; intake rejects stores under
50 orders/month since they can't buy at the planned price.

COPY: headline "Catch serial returners before you refund them." CTA:
"Check my store's exposure" (not "Join waitlist").

MONDAY CHECKLIST: 1. page live, 2. intake wired to sheet, 3. thresholds
pinned in team channel BEFORE the newsletter goes out, 4. send, 5. tally
Friday against the pinned numbers.

Why it works

Committing kill/continue numbers before seeing data is the same mechanism as pre-registration in research: it removes the option to move the goalposts after the fact. Naming the self-deception zone and its tiebreaker question in advance handles the most common real outcome, an ambiguous middle result, without a second round of motivated reasoning.