CHAMPPS
Write the expected number down before you run it. That is the only stage that matters.
What is CHAMPPS?
CHAMPPS is seven stages: Challenge, Hypothesis, Action, Measure, Predict, Prove, Scale. Most of it is ordinary experiment design. One stage is not, and it is the reason to use this rather than a test plan template: Predict makes you commit to the expected number before the test runs.
Experiments without a written prediction are almost impossible to lose. The result comes back at 4.6% against a 4.1% baseline and the team decides — after seeing it — whether that counts. If it is positive it validates the hypothesis; if it is negative it was under-powered and deserves a bigger sample. Both readings are available and the one chosen is whichever the team already preferred. Writing the falsification threshold down in advance removes that freedom, and the discomfort of doing so is the signal that it is working. A growth programme that cannot kill its own hypotheses is not learning; it is accumulating.
Where CHAMPPS Came From
A modern growth convention with no author of record
CHAMPPS has no documented inventor. It sits in the growth-practice tradition that emerged from the mid-2010s — the build-measure-learn lineage, the growth-team playbooks, the experiment-backlog culture — compressed into seven letters. Any page naming a creator is inventing one.
Two of the seven are about not fooling yourself
The interesting structural point. Challenge, Action, Measure and Scale are what any test plan has. Predict and Prove are the additions, and both exist to constrain interpretation rather than execution — one before the result arrives and one after. That is a more honest picture of where growth experiments actually go wrong than any amount of advice about sample size.
Where the Prove stage comes from
Prove asks what a positive result would and would not license you to conclude, which is the question correlational data quietly begs. If teams that invite a second user convert at 11.3% against 4.1%, the tempting reading is that inviting causes conversion. The other reading is that teams already committed enough to convert are also the ones who invite. An experiment designed without that distinction in mind will confirm itself.
The 7 Slots, One at a Time
Each slot is a decision. Leave it out and the model still makes it — just without you.
Not a feature idea, a problem. "Trial-to-paid is 4.1% on 9,000 trials a month" is a challenge; "we should add an invite flow" is a solution that has skipped it. Starting from the solution is how growth backlogs fill with tests nobody can interpret.
We believe X causes Y, and if we do Z then metric M moves from A to B. The test is whether it could be wrong — "we believe improving onboarding will increase conversion" cannot be, which is why it is the most common thing written in this slot and the least useful.
Sized explicitly. An action that would take three weeks in a one-week slot becomes a smaller action badly, discovered halfway through. Stating the constraint in the prompt — one week of two engineers — makes the model design something that fits rather than something aspirational.
Three things, and the guardrail is the one people omit. A test that lifts trial conversion while raising first-week churn has not succeeded, and without a stated stopping metric nobody notices until the cohort matures. Compute the sample from your real traffic, not from a rule of thumb.
The stage that distinguishes this framework, and the uncomfortable one. Commit to what you expect and, crucially, to the result that would make you abandon the hypothesis rather than re-run it with more traffic. If you cannot name a number that would change your mind, the experiment is a formality.
The correlational trap, handled explicitly. State what would have to be true for the effect to be causal rather than selection, and what a positive result still does not establish. This is the stage that stops a promising test becoming a company-wide belief it cannot support.
Rolling a winning variant to everything is where a good experiment turns into a bad decision. Name the conditions — sample maturity, guardrail stability, whether the effect held in the segments you did not test — before the result creates its own momentum.
One Task, Before and After
The task: A one-week growth experiment on trial-to-paid conversion. Both prompts below are scored by our free prompt checker — paste either in and you will get the same number, because the scoring is deterministic.
Design an experiment to improve our trial conversion.
Eight words. You get a test plan with an unfalsifiable hypothesis, no guardrail, and no threshold - so any result can be read as encouraging.
Design a growth experiment using Challenge, Hypothesis, Action, Measure, Predict, Prove, Scale. The situation, and the only data available: 4.1% of free trials convert to paid. 9,000 trials a month. Trials last 14 days. 71% of trials never invite a second user, and among those that do, conversion is 11.3%. We have no data on why. Two engineers can spend a week on this. Challenge - state the business problem in one sentence, with the number. Hypothesis - a single falsifiable statement in the form: we believe X causes Y, and if we do Z then M will move from A to B. Not a goal, a claim that could be wrong. Action - what you will actually build or change, sized to one week of two engineers. Measure - the primary metric, the sample needed, and the guardrail metric that would stop the test. Predict - and this is the stage that makes CHAMPPS different from ordinary A/B testing. Commit, before running it, to what result you expect and what result would falsify the hypothesis. Write the number down. Prove - what a positive result would and would not license you to conclude. The 11.3% figure is correlational; say what it would take to treat it as causal. Scale - what happens if it works, and specifically what would have to be true before rolling it to all 9,000 trials. Write the experiment plan as a one-page brief, 700 to 900 words, for a product lead and two engineers, in seven short headed sections. For example, write the Predict stage like this: ``` Predict: trial-to-paid among the treated group rises from 4.1% to at least 5.5%. Below 4.6% we treat the hypothesis as falsified rather than under-powered, and we do not re-run it with a bigger sample. ``` Weigh the selection effect before you write: teams that invite a second user may convert better because they were already more committed, not because inviting causes conversion, and an experiment that ignores that will confirm itself. Ensure the sample calculation uses the 9,000 monthly trials. Do not invent industry benchmarks, competitor conversion rates or statistical figures not derivable from the numbers above.
Ninety-two. The Predict stage is the whole difference: without a falsification number written first, a weak positive validates and a weak negative is under-powered, and the team picks.
Ninety-two, and nothing checks whether the hypothesis could be wrong
One check is out of reach, and the two properties that make an experiment plan worth writing are unmeasured:
No persona slot. Eight points for adding one and no reason to expect a better experiment from it — this is a plan two engineers will execute, not a piece of writing.
The core of the method, and invisible. "We believe better onboarding will improve conversion" scores exactly as well as a claim with a threshold attached — and it cannot be wrong, which means the experiment cannot fail, which means it will teach you nothing.
A prediction added after the result is indistinguishable from one committed to beforehand. Only the timestamp knows, which is a good argument for writing the plan somewhere with one.
The line worth taking from CHAMPPS into any test: name the result that would make you abandon the hypothesis rather than re-run it. It takes one sentence, it is genuinely uncomfortable to write, and it is the difference between a growth programme that learns and one that accumulates tests it interpreted generously.
Copy-Paste Prompt Template
Replace the bracketed placeholders with your specific details.
[The situation, and the ONLY data available — including which figures are correlational] Challenge: [the business problem in one sentence, with the number] Hypothesis: [we believe X causes Y; if we do Z then M moves from A to B. It must be able to be WRONG] Action: [what you will build, sized to the time you actually have] Measure: [primary metric, sample needed, and the GUARDRAIL that stops the test] Predict: [the expected number AND the result that falsifies the hypothesis — written BEFORE it runs, and not re-run with a bigger sample] Prove: [what a positive result would and would NOT license you to conclude] Scale: [what must be true before rolling it out] [Watch the selection effect. No invented benchmarks or statistics]
When CHAMPPS Fits — and When It Does Not
- Growth experiments where the result will inform a real decision.
- Teams with an experiment backlog and no shared standard for calling a test.
- Conversion, onboarding and activation work with enough traffic to measure.
- Any test where a correlational signal is being treated as a causal one.
- Situations where a previous test was interpreted generously and nobody wants to say so.
- Low-traffic products where no honest sample is reachable — the plan will be theatre.
- Qualitative research. Use SCOPE, which is built for investigations rather than tests.
- Obvious fixes. A broken button does not need seven stages of experiment design.
- Strategic decisions that cannot be A/B tested. Use ToT or SWOT.
- Anywhere the organisation will not accept a negative result — the framework will not save you.
10 Ready-Made CHAMPPS Prompts
Every prompt below was produced by the Frompting generator with CHAMPPS selected — not written by hand for this page. Each is scored by our prompt checker; the median is 90/100. Click one to open it, then copy.
An experiment to improve trial conversion 90
You are a growth strategist tasked with crafting a detailed experiment to boost trial-to-paid conversion. Your output should be a concise, step-by-step plan that walks through the problem, a testable idea, the exact actions to run, how to track results, expected outcomes, validation methods, and a roadmap for scaling if successful. The plan is for **[PRODUCT TYPE]:** a brief description of the product or service offering the trial, and **[TARGET USER PERSONA]:** the primary segment you aim to convert. Structure your response as follows, keeping each section to under 1200 words total (≈ 700 words overall): 1. **Problem statement** - Describe the specific conversion bottleneck you are addressing, based on the provided product and persona. 2. **Testable proposition** - Formulate a clear hypothesis linking a change to an expected lift in conversion. 3. **Experiment design** - List the precise actions (e.g., messaging tweaks, onboarding flow adjustments, incentive offers) and the experimental setup (control vs. variant, sample size, duration). 4. **Metrics & data collection** - Identify the primary metric(s) to measure (e.g., conversion rate, time to upgrade) and any secondary signals, specifying how data will be captured. 5. **Projected impact** - Estimate the potential uplift and its business relevance, stating any assumptions made. 6. **Validation criteria** - Define the statistical thresholds or success criteria that will prove the hypothesis. 7. **Scaling roadmap** - Outline next steps for rolling out the winning variant, including any required resources or further tests. **Quality criteria:** - Each step must be actionable and directly tied to the conversion goal. - Use concrete, measurable language; avoid vague statements. - Highlight any assumptions and note where additional information is needed. **Boundary:** Do not include any marketing copy or creative assets; focus solely on the experimental framework and execution plan. Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader. If any bracketed detail above is left unfilled, choose a sensible value from the context, state that assumption in one line before you begin, and continue - do not ask for it and stop.
Reducing checkout abandonment 83
You are a growth experimentation strategist. Your task is to design a systematic test aimed at reducing checkout abandonment. First, describe the specific problem you are addressing, including any known metrics such as the current abandonment rate and the checkout flow steps where drop‑offs are observed. Next, propose a clear, testable hypothesis that explains why users abandon the checkout and how a change could improve completion. Then, outline a detailed experimental plan: the exact variation to implement, required resources, and the steps for execution. After that, specify the key metrics you will track to evaluate the experiment’s impact, including primary and secondary indicators. Provide a short forecast of the expected outcome if the hypothesis holds true, expressed as a percentage change or absolute number. Explain how you will validate the results, covering statistical significance criteria, sample size considerations, and any required analysis methods. Finally, suggest how the winning variation could be rolled out at scale, noting any potential constraints or additional steps needed for broader implementation. Your response should be a concise, structured document of **300–400 words**, using clear headings for each section (Problem, Hypothesis, Plan, Metrics, Forecast, Validation, Scaling). Quality criteria: 1. Each section directly follows the logical flow described above. 2. All recommendations are actionable and measurable. 3. Language is precise, avoiding vague statements. Assume any missing details are unknown; explicitly state those assumptions and ask up to three clarifying questions (e.g., current checkout funnel steps, target user segment, available analytics tools) before finalizing the plan. [CURRENT_ABANDONMENT_RATE]: supply the present checkout abandonment percentage. [CHECKOUT_STEPS]: list the key steps in the checkout process you wish to target. [AVAILABLE_RESOURCES]: describe the team, tools, or budget allocated for the experiment. Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
A test for a new onboarding flow 96
You are a growth‑experiment design specialist. Your task is to create a detailed test plan for a new onboarding flow. First, identify the core problem the onboarding flow aims to solve and the desired outcome. Next, formulate a concise hypothesis that links a specific change in the onboarding experience to an expected improvement. Then, outline the exact actions required to implement the test, including any variations, user segments, and tools. Specify the primary metrics you will track to evaluate the test, and define how you will collect and analyze the data. Predict the expected impact on the key metric, providing a realistic range. Describe the evidence needed to confirm the hypothesis, including statistical methods and confidence levels. Finally, propose how the test could be scaled if successful, noting any additional resources or adjustments required. Deliver the test plan in a structured markdown document with the following sections: - Problem statement (max 80 words) - Hypothesis (max 60 words) - Test actions (bullet list, each ≤ 20 words) - Measurement plan (table with metric, collection method, frequency) - Predicted outcome (max 40 words) - Validation criteria (bullet list, each ≤ 15 words) - Scaling strategy (max 80 words) The plan should be clear, actionable, and grounded in data‑driven reasoning. Ensure all recommendations are feasible and avoid speculative claims beyond the provided information. If any essential details are missing, state your assumptions explicitly and ask up to three clarifying questions before finalizing the plan. [ONBOARDING_GOAL]: brief description of the intended user behavior or conversion you want to achieve. [KEY_METRIC]: primary metric to measure success (e.g., completion rate, activation rate). [TARGET_USER_SEGMENT]: specific user group the onboarding flow targets. [TEST_DURATION]: proposed length of the experiment (e.g., 2 weeks). [SUCCESS_THRESHOLD]: numeric target that defines a successful outcome (e.g., 5% lift). Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader. Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
An experiment on pricing page layout 96
You are a growth‑experiment strategist. Your task is to design a complete pricing‑page layout test that can be executed by a product team. First, describe the core problem the current pricing page presents and why it matters for conversion. Next, formulate a clear, testable prediction about how a specific layout change will affect the key metric. Then, outline the exact steps the team should take to implement the variation, including any required copy, design elements, and placement details. After that, specify how to collect data—what primary metric to track, any secondary signals, and the measurement method (e.g., analytics event, funnel step). Provide an estimate of the expected impact based on the prediction, noting any assumptions made. Detail how to validate the result, including statistical criteria (confidence level, minimum sample size) and the decision rule for success. Finally, suggest how to expand the test if it succeeds, covering rollout strategy, additional variations, and longer‑term monitoring. The output should be a concise, structured plan in plain text, using short headings (no more than four words each) followed by bullet points. Keep the entire response between 250–350 words. Quality criteria: 1. Each section flows logically to the next, reflecting the implicit framework order. 2. All recommendations are actionable with concrete details. 3. Assumptions are explicitly stated, and any unknowns are marked as placeholders in the form **[PLACEHOLDER: brief hint]**. Boundary: do not include code, mockups, or visual designs; focus solely on the experimental design and execution steps. If any critical information is missing, state your assumptions and ask up to three clarifying questions before finalizing the plan. Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader. Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
Improving email open rates 89
You are a growth‑experiment strategist. Your task is to devise a concrete, data‑driven test aimed at increasing email open rates. First, describe the specific problem you are trying to solve, including any known baseline performance. Next, formulate a clear, testable hypothesis about what change will boost opens. Then, outline a step‑by‑step experimental plan, specifying the variation(s) to be created, the audience segment to receive each version, and the method of delivery. Identify the primary metric you will track, how you will collect the data, and the statistical criteria for determining success. Provide a short forecast of the expected impact if the hypothesis holds, based on reasonable assumptions. Explain how you would validate the results and what evidence would constitute proof. Finally, suggest how the winning approach could be rolled out at scale, noting any operational considerations. Deliver the response in a concise markdown format with the following sections in order: problem statement, hypothesis, experimental design, measurement plan, impact prediction, validation criteria, scaling plan. Keep the total length between 300 and 400 words. Quality criteria: - Each section must be logically connected and actionable. - Use concrete, measurable language; avoid vague statements. - Include at least one example of the email subject line or preheader you would test. Boundary: Do not propose changes unrelated to the email subject line, preheader, or send timing. If any key details are missing, state your assumptions explicitly and ask up to three clarifying questions (e.g., current open‑rate baseline, target audience profile, testing duration). Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
An experiment on referral incentives 88
You are a growth‑experiment strategist. Your task is to design a complete, data‑driven experiment that tests a new referral incentive aimed at increasing user acquisition. First, define the specific problem you are trying to solve and the desired outcome. Next, formulate a clear, testable hypothesis about how the referral incentive will impact acquisition. Then, outline the exact experimental setup: the incentive mechanics, user segments, delivery method, and any required tools or platforms. After that, specify the primary and secondary metrics you will track, the data collection method, and the statistical significance threshold. Provide a short forecast of expected results based on reasonable assumptions, and describe how you will validate those predictions with the collected data. Finally, detail a step‑by‑step plan for scaling the incentive if the experiment succeeds, including any adjustments to targeting, incentive value, or rollout cadence. Your response should be a concise, structured plan of **approximately 350 words**, using clear headings for each section (Problem, Hypothesis, Experiment Design, Metrics, Prediction, Validation, Scaling). Quality criteria: 1. Each section must be logically connected and actionable. 2. Metrics and success criteria must be quantifiable and appropriate for referral programs. 3. The scaling plan should address potential risks and resource considerations. Exclude any discussion of alternative growth tactics unrelated to referral incentives. If any essential details are missing, state your assumptions explicitly and ask up to three clarifying questions (e.g., target user segment, incentive budget, desired time frame) before finalizing the plan. Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
A test for a new signup flow 96
You are a growth‑experiment design specialist. Your task is to create a detailed test plan for a new signup flow. First, identify the core problem the signup flow aims to solve and describe the expected user behavior change. Next, formulate a concise hypothesis that links a specific change in the flow to an improvement in a key metric. Then, outline a concrete experiment, including the exact variation to be implemented, the user segment to target, and the steps required to launch the test. Specify the primary and secondary metrics you will track, the data collection method, and the statistical confidence level needed. Provide a short forecast of the expected impact if the hypothesis holds, including a range of possible outcomes. Detail how you will validate the results, describing the analysis techniques and criteria for confirming the hypothesis. Finally, suggest a plan for scaling the successful variation, noting any additional resources, integration steps, or follow‑up experiments required. Deliver the test plan in a structured markdown format with the following sections: - Problem statement (≈50 words) - Hypothesis (≈30 words) - Experiment design (≈120 words) - Metrics & measurement (≈80 words) - Expected impact (≈60 words) - Validation approach (≈70 words) - Scaling strategy (≈80 words) Quality criteria: 1. All steps are actionable and include specific details (e.g., sample size, duration). 2. Metrics are clearly defined, measurable, and aligned with the hypothesis. 3. The scaling plan addresses potential constraints and resource needs. Exclude any assumptions about the product’s industry, user demographics, or existing data unless you state them as assumptions. If any critical information is missing, explicitly state your assumptions and ask up to three clarifying questions before finalizing the plan. Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader. Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
Increasing feature adoption 90
You are a growth experiment strategist. Your task is to design a concise, step-by-step plan to boost adoption of a specific product feature. First, describe the core problem you are trying to solve and why increasing adoption matters. Next, formulate a testable hypothesis that links a concrete change to the expected lift in adoption. Then, outline the exact action(s) you will implement to test the hypothesis, including any variations, targeting rules, or user flows. After that, specify the primary metric(s) you will track, the data sources you will use, and the success threshold that will indicate a meaningful lift. Following the measurement plan, predict the expected outcome range based on reasonable assumptions. Provide a brief validation approach that explains how you will confirm the results are reliable and not due to random variation. Finally, suggest how the experiment could be scaled if the results meet or exceed the success threshold, noting any additional resources or adjustments needed. Deliver the plan in a clear, numbered list, keeping the total length under 495 words. Quality criteria: 1. Each step must be actionable and include enough detail to be executed without further clarification. 2. The hypothesis and success metric must be directly tied to the adoption goal. 3. The scaling suggestion should consider feasibility and potential impact. Assume any missing details are unknown; state those assumptions explicitly and ask up to three clarifying questions before finalizing the plan. [FEATURE_DESCRIPTION]: brief description of the feature to be adopted. [TARGET_USER_SEGMENT]: specific user group or persona you aim to influence. [ADOPTION_GOAL]: numeric target or percentage increase you seek. [TIMELINE]: duration for running the experiment. [RESOURCE_LIMITS]: any constraints on budget, tooling, or personnel. Write this for [AUDIENCE: who will read the output, and how much they already know]. Match the depth, vocabulary and examples to that reader.
A test for paywall placement 88
You are a growth experiment designer. Your task is to create a detailed plan for testing the optimal placement of a paywall on a digital product. First, describe the specific problem you are trying to solve, including the current user journey and where the paywall might be introduced. Next, formulate a clear, testable hypothesis about how the chosen placement will affect user behavior. Then, outline a step‑by‑step experimental procedure, specifying the variations to be tested, the audience segments, and the tools or platforms to be used for implementation. Identify the primary metric(s) you will track to evaluate the experiment’s success, and explain how you will collect and analyze the data. Provide a short forecast of the expected outcome if the hypothesis holds true, including any quantitative targets. Detail the method you will use to validate the results, describing statistical checks or confidence thresholds. Finally, propose a scalable rollout plan that leverages the findings to expand the paywall strategy across the product, noting any additional resources or adjustments required. Deliver the plan in a structured markdown format with clear headings for each section, keeping the total length between 300 and 450 words. Quality criteria: - Logical flow that mirrors the experimental design steps. - Concrete, actionable instructions without vague language. - Clear criteria for success and validation. If any essential details are missing, state your assumptions explicitly and ask up to three clarifying questions before finalizing the plan. [CONTENT TYPE]: specify the type of content (e.g., article, video, tool) the paywall will guard. [PAYWALL MODEL]: specify the paywall model (e.g., hard, soft, metered). [TARGET AUDIENCE]: describe the primary user segment to be tested. [KEY METRIC]: indicate the main performance metric (e.g., conversion rate, churn reduction). [TEST DURATION]: define the intended length of the experiment. Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
Reducing churn in month two 90
You are a growth experiment strategist. Your task is to design a concise, data‑driven experiment aimed at reducing user churn during the second month after acquisition. First, describe the specific problem you are addressing, including the current churn metric you aim to improve. Next, formulate a clear, testable hypothesis that explains why churn is occurring in month two and how a targeted change could lower it. Then, outline a single, actionable intervention you will implement to test the hypothesis, specifying the exact changes to the product, messaging, or user experience. Identify the key metric(s) you will track to evaluate the experiment’s impact, and define the success threshold that would confirm the hypothesis. Provide a brief forecast of the expected outcome if the hypothesis holds true, including an approximate percentage reduction in churn. Explain how you will validate the results, detailing the analysis method and any statistical checks required. Finally, describe the steps you would take to scale the successful intervention across the broader user base, noting any additional resources or adjustments needed. [PRODUCT TYPE]: brief description of the product or service being tested. [CURRENT CHURN RATE]: current percentage of users who churn in month two. [TARGET USER SEGMENT]: specific user group the experiment will focus on. [RESOURCE BUDGET]: approximate budget or resources allocated for the experiment. [TIMEFRAME]: duration of the test period (e.g., weeks). Deliver the plan in under 350 words, using clear headings for each section, and ensure the language is precise, actionable, and grounded in realistic assumptions. Include any assumptions you make and ask up to three clarifying questions if needed. Before writing the final answer, work through the problem step by step and weigh the main trade-offs; present only the reasoned conclusion, not your working notes.
Scores range from 83 to 96. They are shown as generated rather than cherry-picked — a library where every entry scores in the nineties tells you it was curated, not measured.
CHAMPPS vs the Alternatives
The research counterpart, and both commit to evaluation criteria in advance. SCOPE plans an investigation; CHAMPPS designs a controlled test.
What tells you which stage to experiment on. REAN diagnoses the funnel; CHAMPPS tests a fix for the stage it identified.
For goals rather than tests. SMART makes a target checkable; CHAMPPS makes a belief falsifiable, which is a stricter thing.
The same instinct applied to reasoning rather than experiments — both build in a stage whose job is to attack the conclusion you were about to reach.
For generating the hypothesis in conversion work. LIFT's factor scores are hypotheses with metrics attached; CHAMPPS is the experiment around one.
Five Ways People Get CHAMPPS Wrong
The defining CHAMPPS failure. "Improving onboarding will increase conversion" cannot be wrong, so the test cannot fail and nothing is learned. Use the form: we believe X causes Y, and if we do Z then M moves from A to B.
Without a number written in advance, a weak positive validates and a weak negative is under-powered. Both readings are available and the team picks the one it preferred.
A Challenge section that describes a feature idea has skipped the problem. Backlogs fill this way with tests nobody can interpret afterwards.
Teams who invite a second user convert better — possibly because inviting helps, possibly because committed teams do both. An experiment that does not separate them will confirm whichever story it started with.
A test that lifts conversion while raising early churn has not worked. Without a stated stopping metric this is discovered when the cohort matures, which is after it shipped.
A win creates momentum, and momentum is not evidence that the effect holds in segments you did not test. Name the conditions for rolling out before the result exists.
CHAMPPS Questions
What does CHAMPPS stand for?
Challenge, Hypothesis, Action, Measure, Predict, Prove, Scale — a seven-stage structure for designing a growth experiment.
What makes it different from a normal A/B test plan?
Two stages. Predict commits you to the expected result and the falsification threshold before the test runs; Prove asks what a positive result would and would not license you to conclude. Both constrain interpretation rather than execution.
What is a falsifiable hypothesis?
One that could turn out to be wrong. "We believe X causes Y, and if we do Z then metric M moves from A to B" can fail. "Improving onboarding will help conversion" cannot, which is why it is the most common and least useful thing written in that slot.
Why write the prediction down beforehand?
Because after the result arrives, a weak positive validates the hypothesis and a weak negative was under-powered — and both readings are genuinely available. Committing in advance removes the freedom to pick the one you preferred.
What is the Prove stage for?
Separating what a result shows from what it suggests. If the signal you started from was correlational, a positive test may reflect selection rather than causation, and saying so in advance stops a promising result becoming a company belief.
What if we do not have enough traffic?
Say so and do not run it. A test with no honest sample produces a number that will be interpreted anyway, which is worse than no number at all.
Generate a CHAMPPS Prompt Instantly
Skip the manual template — Frompting applies CHAMPPS to your topic in one click.
Try it FreeFramework Details
| Name | CHAMPPS |
| Stands for | Challenge-Hypothesis-Action-Measure-Predict-Prove-Scale |
| Domain | Persuasion & Conversion |
| Steps | 7 |
| Access | Pro |