Pricing
Audit my A/B test
Causal-decision infrastructure for consumer apps

The decision layer for AI-first apps. We decide paywalls.

Plug into Statsig, Klaviyo, GrowthBook, Iterable, Braze, Shopify, BigQuery or Snowflake and get a doubly-robust readout on your own data in a day. Upgrade individual experiments to our SDK when you want billable lift. We sit on top of your stack, we don’t replace it.

Audit my last paywall test — freeSee how it works →

One CSV. One experiment config. Same-shape readout back in three business days.

5 SDKsPython · TS · Kotlin · Swift · Flutter
<30msp99 decide latency
DR + ESSbillable lift gate
1 dayto integrate

One decision engine, many surfaces

Metapolicy controls the parameters behind your website, push, mobile app and lifecycle email. Switch from one-size-fits-all to a contextual bandit that serves each user the experience predicted to convert them — and measures the lift it causes.

Banditeach user is served their own experience
First-time visitor
shop.example.com
Free trial
Cart abandoner
shop.example.com
50% OFF
Comparison shopper
shop.example.com
25% OFF
Loyal returner
shop.example.com
10% OFF
Website1,842 live
Dormant 14 days
We miss you
50% OFF — tap
Trial ending soon
Your trial ends tomorrow
30% OFF — tap
Daily streak
Keep your 12-day streak
10% OFF — tap
Weekend lapser
Weekend deal, just for you
40% OFF — tap
Push notification2,106 live
MetapolicyPER-USER
parameters → experiences
paywall.discountPct
Free trial
Discount
app.gridColumns
3
Layout
email.sendHour
9am
Send time
New install
3-col
Power user
2-col
Casual browser
3-col
Churned · back
2-col
Content explorer
3-col
Focused shopper
2-col
Mobile app3,540 live
Onboarding · d3
✉Finish setup — 2 steps left9am
Lapsed subscriber
✉We saved your spot — 40% to return7pm
Weekly digest
✉Your week in review is ready8am
Win-back · 60d
✉It's been a while — 50% off inside9pm
Re-engagement
✉3 new features since you left5pm
Lifecycle email1,488 live

● each cable is one user's decision · every surface personalised per segment, thousands in flight

Trust & compliance
SOC 2 Type II certified
ISO/IEC 27001 certified
GDPR compliant

SOC 2 Type II certified. ISO 27001 certified. GDPR compliant. Hosted in the EU. Your data stays where it belongs.

WHY METAPOLICY

Beyond
A/B tests.

A/B tests report one number: the average lift across your whole base. That average quietly hides the cohort the winner is prompting into churn, and the cohort that would have paid more. Doubly-robust evaluation finds both — on your own logs, without rerunning anything.

Built for consumer subscription apps in
  • Travel
  • Fintech
  • Fitness
  • Mental Health
  • Dating
  • EdTech
  • Media
01

Causal, not correlated.

T-tests average lift across the whole base, hiding the cohorts where the winner actually wins (or loses). Our doubly-robust off-policy evaluation surfaces per-segment CATE with bootstrap confidence intervals and an ESS guardrail &mdash; the readout your CFO will sign off on.

DR + ESSbillable lift gate
02

Outcome-priced. Always.

We are paid a share of the incremental revenue we cause, not a software fee. No lift number that fails the ESS guardrail is invoiced. Your CFO sleeps because the math itself caps how much we can over-claim.

% liftthe only thing we bill
03

One-day integration.

A single SDK call: decide(experiment, user_id, context). Five native clients, Python, TypeScript, Kotlin, Swift, Flutter, same wire format, same six methods. Your engineers ship the integration in an afternoon.

5 SDKsPython · TS · Kotlin · Swift · Flutter
USE CASES · NO-BRAINERS PER VERTICAL

What the bandit decides on every surface.

Pick your vertical. Each use case is a decision your team is making today with a static rule or a global A/B test. Each one is one SDK call and one logged propensity away from a contextual-bandit policy with a doubly-robust readout.

Booking apps, OTAs, mobility platforms.

ICP · Klook · GetYourGuide · Trainline · Hopper · Booking

Headline impact · Aerial+$2.1M / yrARR · paywall pricing per routeRead the worked audit →
Surface · Checkout paywall

Peak-season paywall pricing per route

Today

Hard-coded fare per region, sometimes a seasonal multiplier.

With Metapolicy

Contextual bandit over (route popularity × seasonality × user LTV decile × device). Propensity logged per assignment.

Outcome

+8–15% trial-start lift on high-elasticity routes, no margin erosion on low-elasticity ones.

Who feels it · VP Growth, the paywall is their P&L.

We've already re-analysed this exact decision in a worked audit.

Read the worked audit

None of these is hypothetical, each is a live decision your team is already making with a static rule. We just log the propensity and return a doubly-robust readout.

Audit my last 3 A/B tests, free
" " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " "
WORKED AUDIT EXAMPLES

Five audits we've already handed back.

“
Aerial's Q4 paywall A/B looked dead on a t-test (p = 0.34). Per-cohort doubly-robust found +9.3% lift on top-quartile-route × peak season, a slice carrying 23% of trial-start volume. Shipped just to that cohort, +$2.1M / yr.
W

Worked audit

Travel & mobility · AERIAL-2026Q1-AUDIT-001, Aerial · travel

+$2.1Mannualised ARR · worked audit example
Featured companies
HOW IT WORKS

Decide.Log.Reward. Read.

Idempotent on (tenant, experiment, user)OPE replays your last 90 days on day one
Integrate in a day
DAY ONE · NO COLD START

Value before the bandit has learned anything.

The online bandit policy takes 4–8 weeks to converge. The lift readouts work from day zero on data you already have. Propensities are deterministically reconstructible from your experiment configs, no logging upgrade required to get a first answer.

010 days · 1 CSV

Free 1-hour A/B audit

A 30-minute video walkthrough of one of your recent paywall or onboarding A/B tests, re-analysed through doubly-robust math. Original t-test number, our DR number, ESS, per-segment CATE.

What you walk away with

A defensible number for your last A/B test, plus the cohort the average hid, usually worth a recoverable six-figure ARR slice.

021–3 days

90-day historical replay

A 3–6 page report of your entire experiment program re-evaluated under DR + ESS. One row per experiment. Per-segment CATE where overlap is defensible.

What you walk away with

A one-page register of which past “wins” hold, which “no effects” actually had a per-segment win, and which decisions to roll back this quarter.

031 week joint work

Schema upgrade plan

1-page recommendation to add propensity logging into your existing experimentation framework. Free, even if you never sign with us, every future test becomes replay-able.

What you walk away with

An exact column-level diff against your current experiment schema. Ship it once and every future decision becomes auditable, with us or without us.

044–6 weeks

One bandit, one surface

LinUCB on 3–5 arms for one decision (typically paywall or push). Propensity logged. Sticky-by-user 5% holdout. Weekly doubly-robust readout.

What you walk away with

A live, doubly-robust lift number on real users within six weeks, gated by a 5% holdout your CFO can verify against revenue.

050 SDK · data only

CATE-only discovery report

For regulated buyers (fintech, healthcare) who can't move fast on a new SDK. X-learner / T-learner / Causal Forest on your existing experiment data, “your hidden cohort effects.”

What you walk away with

A heterogeneity report your growth team can act on internally next sprint, no SDK, no engineering ticket, no procurement cycle in the way.

Needs time

Online bandit convergence4–8 weeks for paywall-class decisions; 8–16 weeks for retention with long outcome windows.

Works day zero

OPE · CATE · replayDoubly-robust readouts and heterogeneity analysis run on your existing assignment + outcome data.

The unlock

Propensity from configsA 50/50 split is p = 0.5; a stratified test is the stratum weight. Reconstruct, replay, report.

Send your last paywall test.

We'll send back a doubly-robust readout with ESS, bootstrap CIs, and per-segment CATE in three business days. Free.

Audit my last 3 A/B tests, free
THE NUMBERS

Causal,
not correlated.

Every assignment is propensity-logged. Every reward is idempotent. Every lift readout is doubly-robust and ESS-guardrailed. The math itself caps how much we can over-claim.

5
Language SDKs
Python · TS · Kotlin · Swift · Flutter
<30ms
p99 decide latency
in-region, propensity-logged
DR + ESS
Billable lift gate
no overlap-limited readout invoiced
MIT
Open-source estimators
offpolicy.py on PyPI
IPS · SNIPS · DM · DR · SDRthe off-policy estimator suite, open-sourced
SDKs and ecosystem

Works with
your stack.

Five native SDKs with identical wire formats. Plus drop-in adapters for the mobile-growth tools you already run.

SDK
py
Python
SDK
ts
TypeScript
SDK
kt
Kotlin
SDK
sw
Swift
SDK
fl
Flutter
Billing
$
Stripe
Subscriptions
rc
RevenueCat
CDP
sg
Segment
Analytics
am
Amplitude
Analytics
mx
Mixpanel
Analytics
ph
PostHog
Warehouse
sn
Snowflake
Experimentation
st
Statsig
Experimentation
gb
GrowthBook
Lifecycle
kl
Klaviyo
Lifecycle
it
Iterable
Lifecycle
bz
Braze
5Language SDKs
OpenMIT-licensed estimators
<30msp99 decide latency
View on GitHub→
PRICING

Outcome-priced.
DR + ESS gated.

We charge a share of measured incremental revenue. Doubly-robust off-policy evaluation gates every invoice, no DR + ESS guardrail, no bill.

AUDIT-MODE

Audit-mode

Connect-in-an-hour. We read your existing Statsig / Klaviyo / GrowthBook / Iterable / Braze / Snowflake stack and produce doubly-robust readouts on your own data. Advisory, never billable.

$24–60K/ year
  • Six Tier-1 connectors + CSV upload
  • DR + ESS readout on imported experiments
  • CATE grid the incumbent doesn't ship
  • Disclosed propensity-quality badge
  • Cut-over-to-decision-mode CTA per experiment
  • Quarterly billing, fixed fee
Connect a source

Floor

Decision-mode SDK with logged propensity. Infrastructure cost; caps customer downside if measured lift is small.

$24K/ year
  • Full SDK access (5 languages)
  • Doubly-robust OPE worker
  • ESS-guardrailed readouts
  • Per-tenant API key issuance
  • Append-only decision logs
  • Mailto-grade support
Get in touch
DESIGN PARTNERS

Typical

Series-B/C subscription app, 5–10% of incremental ARR. We grow as you grow.

$50–220KACV
  • Everything in Floor
  • Contextual bandits (LinUCB / TS)
  • CATE refresh + bandit-prior seeding
  • Bootstrap CIs on every readout
  • Replayable policy snapshots
  • Founder-direct support
Become a design partner

Ceiling

Caps our share so the customer's CFO sleeps. Outcome-priced, never more.

$1M/ year
  • Everything in Typical
  • Dedicated OPE windows
  • Custom estimator pinning
  • Compliance and audit support
  • Quarterly causal-readout review
  • Direct line to the founder
Talk to us
No software feeHoldout-validatedAudit on your own logs via offpolicy.py
Email the founder →

Every consumer app
needs a decision layer.

Send me your last three A/B tests. I will tell you which ones survive a doubly-robust readout on your own logs, free, no obligation.

Email the founderSee how it works

No pitch deck. No NDA. A real conversation about the math.