Skip to content

Stop losing users.

Analytics tools show where people leave. Anku tells you what to fix.

No account. No key. No card.See the recorded case, recovered elsewhere →

The stack you already ship

Stripe
BillingTest event
PostHog
Product analyticsConnected, read-only
GitHub
Code and pull requestsNot connected
Resend
SenderNot connected
Supabase or Clerk
IdentityNot connected

PostHog and Stripe are read-only. GitHub opens one branch and one pull request. Resend sends inside your policy. Identity is read-only. A human merges. Every key is revoked in one click.

Diagnosis before dashboards

Charts flag the dip. Anku fixes the screen.

Stop scrubbing fifty session replays on a Sunday night.

WITHOUT ANKU

Scattered clues

  • Fifty replays on a Sunday
  • Raw SQL guesses
  • A fix that waits for the weekend
WITH ANKU · [EXAMPLE]

One grounded fix

61% vs 3% return
  • 5 named of 281 people in the read
  • One staged play, no live send
  • Two books, never blended
01 · DIAGNOSE281 people in the read · [EXAMPLE]
TELEMETRY DROP · [EXAMPLE]The aha gap
People in the read
281
Reached the aha
111
Did not reach
170
Return split61% reached vs 3% missed

111 reached the aha event. 170 did not. The gap is recorded, not modeled.

02 · PRESCRIBE & SHIPOne recorded run · a human merges
"Take a 5-min drill" on /onboarding/onboarding
Make the first click change the control's state and start the next step. If the control is inert, wire it. Do not add a second click target.
8 rage clicks · 5 people · 5 sessions · 30 daysOne instructionOne assertionNo repo connected. No PR opened.

The recorded case has no repo connected, so its artifact is the spec, not a PR.

03 · PROVEVerified: $0 · Recovered: $29
SEPARATED PROOF BOOKS72h window
Book 1 · Verified$02 source ids required
Book 2 · Recovered$291 source id · in_test_1Q4x7kR2

The recorded $29 is recovered, not attributed. Never a verified receipt.

01 · ISOLATE THE DROP

Pinpoints the broken step

Joins the three events you named with the billing facts to name the step that leaks.

281 people in the read · [EXAMPLE]

02 · BOUNDED EXECUTION

Opens one tested fix, or one note

The code writer grounds a defect to a file, writes the fix and one test, and opens a draft PR. A human merges. The send path stays inside your policy.

One recorded run · a human merges

03 · TWO-EVENT LOCK

Proves what came back, or stops

A verified receipt needs the Anku link and the same invoice paid inside 72 hours. One event is recovered, not attributed. The books never blend.

Verified: $0 · Recovered: $29

Diagnosis before dashboards

Charts flag the dip. Anku fixes the screen.

Stop scrubbing fifty session replays on a Sunday night.

WITHOUT ANKU

Scattered clues

  • Fifty replays on a Sunday
  • Raw SQL guesses
  • A fix that waits for the weekend
WITH ANKU · [EXAMPLE]

One grounded fix

61% vs 3% return
  • 5 named of 281 people in the read
  • One staged play, no live send
  • Two books, never blended
01 · DIAGNOSE281 people in the read · [EXAMPLE]

Pinpoints the broken step

Joins the three events you named with the billing facts to name the step that leaks.

TELEMETRY DROP · [EXAMPLE]The aha gap
People in the read
281
Reached the aha
111
Did not reach
170
Return split61% reached vs 3% missed

111 reached the aha event. 170 did not. The gap is recorded, not modeled.

02 · PRESCRIBE & SHIPOne recorded run · a human merges

Opens one tested fix, or one note

The code writer grounds a defect to a file, writes the fix and one test, and opens a draft PR. A human merges. The send path stays inside your policy.

"Take a 5-min drill" on /onboarding/onboarding
Make the first click change the control's state and start the next step. If the control is inert, wire it. Do not add a second click target.
8 rage clicks · 5 people · 5 sessions · 30 daysOne instructionOne assertionNo repo connected. No PR opened.

The recorded case has no repo connected, so its artifact is the spec, not a PR.

03 · PROVEVerified: $0 · Recovered: $29

Proves what came back, or stops

A verified receipt needs the Anku link and the same invoice paid inside 72 hours. One event is recovered, not attributed. The books never blend.

SEPARATED PROOF BOOKS72h window
Book 1 · Verified$02 source ids required
Book 2 · Recovered$291 source id · in_test_1Q4x7kR2

The recorded $29 is recovered, not attributed. Never a verified receipt.

Built around restraint

Runs on your stack, not on guesswork.

01 · 3-EVENT MAP

The stream you already emit

1 · signup_completed2 · scan_completed3 · invoice.paid
No wrapper library · read-only PostHog and Stripe
02 · SEND ENVELOPE

The cap and the gate

Play A2 queued for n•••@example.com. Waiting on your click.

Mode: human approve · cap ok
The policy check runs before every outbound message
03 · RULE 14

Scoped paths, green checks

+ allow: app/** components/**
- forbid: .github/workflows/**
- forbid: supabase/migrations/**
- forbid: .env* lib/policy/** lib/agent/**
= merge_gate: human_click_only
Zero self-merges · the diff stays inside the allowlist
04 · HONEST STOP

Stops cleanly when retention holds

61% vs 3%The stop is a deliverable

Recorded gap · [EXAMPLE]

Never fabricates lift when churn is outside the funnel
01Three events, zero SDK bloatName signup, the aha event, and renewal once. Anku reads them from the tools you already run.
01 · 3-EVENT MAP

The stream you already emit

1 · signup_completed2 · scan_completed3 · invoice.paid
No wrapper library · read-only PostHog and Stripe
02Policy caps before every sendRequire a click for every send, or set a daily cap and a quiet window and let the policy run inside it. The kill switch stops the loop.
02 · SEND ENVELOPE

The cap and the gate

Play A2 queued for n•••@example.com. Waiting on your click.

Mode: human approve · cap ok
The policy check runs before every outbound message
03Rule 14 boundary on your repoAnku edits only the paths you allow, opens a branch and a draft PR, and never merges. The diff stays inside the allowlist.
03 · RULE 14

Scoped paths, green checks

+ allow: app/** components/**
- forbid: .github/workflows/**
- forbid: supabase/migrations/**
- forbid: .env* lib/policy/** lib/agent/**
= merge_gate: human_click_only
Zero self-merges · the diff stays inside the allowlist
04Honest stops on flat cohortsWhen activation climbs and retention holds, Anku says it is not a funnel problem and stops.
04 · HONEST STOP

Stops cleanly when retention holds

61% vs 3%The stop is a deliverable

Recorded gap · [EXAMPLE]

Never fabricates lift when churn is outside the funnel

Contract ledger

What ran, what is labeled, and what is not.

Scroll or select a card to inspect the source.

Product specimen · Lamplight dark

One calm workspace from leak to receipt.

Cohorts, the grounded fix, the capped note, and the ledger on one sheet.

Workspace / Example / Diagnose

The verdict

A paying user's card expired and the last payment failed.

Failed Cards1Fake Churn ≤3 actionsnot measuredPre-Cancel Driftnot measuredJob Done / Wrong Fitnot measured

n•••@example.com · Growth $29/mo

A paying user's card expired and the last payment failed.

Recorded case

Evidence

  1. analytics
    Active in the recorded window
    6 events · longest session 47s · never reached the aha event · distinct_id distinct_0b8e2f1a · recorded aggregate window
  2. 18 Oct · consented test event
    Invoice failed (card_declined: expired_card)
    $29.00 due · invoice in_test_1Q4x7kR2 · customer cus_test_8f3a2b7c
  3. 25 Oct · consented test event
    Smart retry #3 failed
    3 attempts, card_declined (expired_card) each time
  4. 28 Oct
    Play staged · payment recovery
    The note is written. It needs a sender.
  5. 29 Oct
    Channel connected · ready
    Resend · Recorded before the sender link existed
Plan · Growth $29/moOpen the case
NO EVENT BLOATReads your existing three-event map and the billing factsZoju · consented demo data · Recorded before the sender link existed

Hard questions and current limits

Clear answers before you connect anything.

Select a question

The stop is a deliverable

What if my problem is not the funnel?

Then the case says so. The acquisition stop, the PMF gate, the no-dominant-leak check, and the not-a-funnel stop all stage nothing and hand over the work that can still be done. A no is a deliverable.

No case staged
The case stays open

The limits that stay

  • The receipt proves a payment happened, not that Anku caused it.
  • Only failed cards are measured. The other three cause buckets are not.
  • The queue surfaces one case. The ranking is designed, not exercised.
  • A founder cannot connect a sender on this page. The code has landed; a verified domain and a live send have not been demonstrated.

Simple flat pricing

$49 a month. Flat.

Flat workspace price. No cut of recovered revenue. Pro is not on sale yet; request access and we email you when your data can join.

Public diagnostic

Free

$0your case, free · no card

Paste any public SaaS URL. Open Case #0001 across four read-only checks.

  • 4 public surface reads
  • A stated stage diagnosis or an honest stop
  • No account, no key, no card
Autonomous workspace

Pro

$49/mothe recovery loop · opening when your data joins

Connect PostHog, Stripe, and GitHub for grounded draft PRs, capped sends, and the proof ledger.

  • The full play catalog across acquisition, activation and billing
  • Draft PRs under Rule 14; a human merges every diff
  • Daily send caps, 0% discounts, and the two-source lock
  • Merchant of record via Polar; a flat rate regardless of recovered MRR

Pro is not on sale yet. Request access and we email you when your data can join. No card is taken while Pro is closed; when it opens, cards are entered only on Polar-hosted pages.

Compare plans →

Ready when your funnel is

Find the leak before another cohort slips through.

Start with your public URL for a free read-only check. No keys, no card, no discount sequences.

No account. No key. No card.

Anku: find the leak in four reads