Use cases

Query

If we move the mid tier from £9 to £12, who leaves, who upgrades, and who quietly stops spending without cancelling?

Result

MEDIAN −4.1PP · SPREAD −9.2PP TO −0.6PP

CONFIDENCE: MODERATE-HIGH

Churn and upgrade distributions at each candidate price, split by segment, with the point where the churn curve bends.

What changes

You pick the price with the tolerable tail rather than the best median.

Query

We are shipping a change our heaviest users will hate. How much do they actually hate it, and do they leave or just complain?

Result

MEDIAN −7.4PP · SPREAD −19.0PP TO +1.2PP

CONFIDENCE: MODERATE

Reaction distribution by usage tier, and the split between vocal objection and actual departure.

What changes

You stop treating the loudest cohort as the representative one, or you find out they are and delay.

Query

Three framings for the same announcement. Which one moves people, and which one starts a fire?

Result

FRAMING A +3.1PP · FRAMING B +5.8PP · FRAMING C +6.9PP

CONFIDENCE: HIGH

Per-framing response by segment plus a backlash tail estimate for each.

What changes

You kill the option that wins on average and loses on the tail.

Query

If we change this rule, who complies, who leaves, and who organises against it?

Result

COMPLIANCE 74% · SPREAD 61% TO 83%

CONFIDENCE: MODERATE

Compliance, exit and coordination rates, with the conditions under which organised resistance forms.

What changes

You know before publication whether you need an enforcement plan or a communications plan.

Query

Apologise, clarify, or say nothing. Which does least damage?

Result

APOLOGY −2.6PP · CLARIFY −4.9PP · SILENCE −11.8PP

CONFIDENCE: MODERATE-HIGH

Three simulated responses, each with a damage distribution and a tail-risk figure.

What changes

You choose against worst case instead of against instinct, in hours rather than after the fact.

Query

We have four ideas for keeping people. Which of them actually work?

Result

BEST LIFT +6.2PP · SPREAD +3.4PP TO +8.1PP

CONFIDENCE: HIGH

Retention lift per intervention, plus which ones only delay exit and which ones accelerate it by naming the problem.

What changes

You stop funding the intervention that looks good in a deck and reminds people why they were leaving.

Query

We have never sold to this population. Do our assumptions transfer?

Result

MEDIAN +1.8PP · SPREAD −12.4PP TO +14.0PP

CONFIDENCE: LOW-MODERATE

Predicted response with an explicit confidence penalty on the parts furthest from observed behaviour, and a list of which assumptions break.

What changes

You spend the market-entry budget on the uncertainty that matters instead of all of it at once.

Query

Will this policy stop the behaviour or just move it somewhere we cannot see?

Result

SUPPRESSED 68% · SPREAD 52% TO 79%

CONFIDENCE: MODERATE

Suppression versus displacement rates, and where the displaced behaviour surfaces.

What changes

You avoid shipping a policy that makes your metrics look better and your platform worse.

Query

What does the bad version look like, and how likely is it?

Result

MEDIAN −5.0PP · P05 −31.0PP

CONFIDENCE: MODERATE

The tail of the outcome distribution, the conditions that produce it, and how far from the median it sits.

What changes

You size the downside before you commit, which is the cheapest thing on this page to simulate.