Skip to main content
Each playbook below follows the same shape: the question you are trying to answer, how you would set it up, what the output tells you, and the real-world result from validation. The setup steps in every playbook happen in the product: you describe the study in the intake chat, build and screen the cohort, upload any stimuli, then launch and read the report.

​
Playbook 1: Should we change our pricing?

Question: If we raise the price or add a cheaper tier, how many customers stay, switch, or churn?

Screen for the right population

Add a screening question so you only interview people the decision affects, such as “Do you personally pay for this subscription?” Getting this wrong is the most common way a pricing study goes sideways.

Describe the change neutrally

Give replicas the scenario (current price, the new options) without editorializing. Avoid lines like “unfortunately we have to raise prices.”

Offer symmetric choices

Present accept, downgrade, and cancel as balanced options, and ask for the reasoning behind the choice.
What the output tells you: the predicted distribution across choices, plus the reasoning. Usually it comes down to a mix of inertia, value perception, and individual budget constraints.
Validated: on the Disney+ 2022 price increase, replicas predicted the accept/switch/cancel distribution with 91% accuracy and screened the paying population within 4.3% of ground truth. They correctly captured that about 94% of subscribers simply absorbed the increase.

​
Playbook 2: Which ad creative should we run?

Question: Of two or more creatives, which will actually perform?

Target your real audience

Build a cohort that matches who the ad is for, for example “women 22 to 35 earning $100K+”.

Show the creatives as stimuli

Upload the ads and let the researcher present them during the interview. Use the Compare Ad Creatives template.

Ask reaction-first questions

Get unaided reactions (“what would you do after seeing this?”) before any forced comparison, and never name the weakness you suspect.
What the output tells you: which creative wins, by how much, and the real engagement drivers behind it. These are often things that surface-level best practices miss.
Validated: across near-identical ad pairs, replicas picked the higher-reach ad 89% of the time versus 65% for a general AI. In a paid-ads case study with Mera NYC, replicas predicted the winner 72% to 28%, matching the real result: the winning ad earned a much higher return on ad spend. The general AI predicted the opposite winner.

​
Playbook 3: Which product will the market choose?

Question: Given a lineup of options at different price points, what does the market pick?

Use a representative cohort

For market-share questions, weight the cohort to your target population so the distribution is meaningful.

Lay out the full lineup

Present each option and its trade-offs, then ask which they would choose and what drove it: price, features, brand, or social signaling.
What the output tells you: the predicted share for each option, and the decision drivers that separate them.
Validated: on the iPhone 17 lineup, replicas matched real Q4 2025 sales distribution with 97.6% accuracy, reaching the same conclusion through simulated individual decisions. That held up even against models that had market-share data in their training set.

​
Playbook 4: What is this worth to customers?

Question: What will people actually pay?

Use the Test Pricing template

It asks a proven set of pricing questions that find the range people will pay, instead of the leading “how much would you pay?”

Anchor to a real product

Ask about a specific item the replica was already considering, and keep explicit prices out of the question text.
What the output tells you: the price range people will actually pay, and the point where it tips into too expensive.
Validated: across about 300 products ($20 to $2,200), replicas predicted what people would pay about as closely as that same person would land on a different day.

​
Before You Launch Any of These

Write neutral questions

Framing distorts results more for AI than for humans. Get this right first.

Plan how you'll read it

Know what a surprising result means before you see one.