Retrics

AUTOPILOT · AD SPEND, ANSWERABLE

Ad spend, answerable
to verified orders.

Retrics prices every campaign on store-verified orders — and on whether the cohort it bought ever came back. It recommends the move, with the case attached, for you to apply in your ad platform.
See pricing →
EARLY ACCESSCONCEPT PREVIEW · NUMBERS ILLUSTRATIVE
RETRICS · AD DIAGNOSIS7-DAY WINDOW
CHANNELCLAIMED → VERIFIED ROAS
META4.8x 2.1x
GOOGLE3.9x 1.6x
AUTOPILOT SUGGESTS

Cut budget — Prospecting, Broad

META · $180 → $153 / DAY · −15%

Apply in your ad platformDismiss

CONCEPT PREVIEW · NUMBERS ILLUSTRATIVE · META AND GOOGLE ARE THE PLATFORMS IT READS

ONE VERIFIED LEDGERPLATFORMS DON’T GRADE THEMSELVESSUGGEST → APPROVE → ARMEVERY RECOMMENDATION HAS A RECEIPTKILL SWITCH ALWAYS LIVEVERIFIED OVER CLAIMED

THE CONTROL LOOP

Verified, proposed, applied by you, bounded.

01 IT VERIFIES

Platforms grade themselves.

Retrics prices every campaign on store-verified orders — and on whether the cohort it bought ever came back.

02 IT PROPOSES

With the case attached.

Verified CAC against target, cohort size, attribution quality, and a confidence grade — stored verbatim.

03 YOU APPLY IT

Retrics touches no ad account.

Retrics executing on your ad account is switched off in code. Approving records your decision; you make the change in your ad platform, from the proposal and its case.

04 IT STAYS BOUNDED

Limits you set. Floors you can’t.

Caps and targets are yours. The rate cap, the drift ceiling, the circuit breaker, and the kill switch are not editable away.

01 — DIAGNOSE

Two platforms. One ledger.

Every campaign is priced on store-verified orders — and on the share of the cohort it bought that ever came back. Meta and Google are the platforms Retrics reads. There is no TikTok client, so no TikTok spend is read, and none of it is graded here.

Verified ledgerVERIFIED AGAINST STORE ORDERS
CHANNELCAMPAIGNVERIFIED2ND ORD
METAProspecting — Broad1.9x9%
METARetargeting — 30d3.4x34%
METAAdvantage+ — Catalog2.8x28%
GOOGLEBrand search5.1x41%
GOOGLEPMax — Catalog1.2x6%
7-DAY WINDOW · STORE-VERIFIED REVENUEOVER TARGET CAC · $412/day

CONCEPT PREVIEW · NUMBERS ILLUSTRATIVE

SECOND-ORDER RATEPER CAMPAIGN
META · PROSPECTING9%
META · RETARGETING34%
COHORT PRICED ON REPEAT REVENUELTV : CAC
ATTRIBUTION MATCH7-DAY WINDOW
MATCHED BY ORDER ID86%
ACQUIRED IN WINDOW20
DECISION CONFIDENCEHIGH

WATCH IT REBALANCE

The channels, graded against
store-verified orders.

Retrics reads Meta and Google — the two platforms it has a client for — then prices each on store-verified orders, not on the returns each platform reports for itself. It recommends the shift, with the case attached, for you to apply in your ad platform.

CONCEPT PREVIEW · SUGGEST-ONLY, EXECUTION SWITCHED OFF IN CODE · KILL SWITCH ALWAYS LIVE

META
4.6xSTORE-VERIFIED ROAS
GOOGLE
1.8xSTORE-VERIFIED ROAS
AUTOPILOT RECOMMENDS

Shift $120/day from Google to Meta

RECOMMEND-ONLY · YOU APPLY IT IN YOUR AD ACCOUNT

See the caseDismiss

CONCEPT PREVIEW · NUMBERS ILLUSTRATIVE

WHAT IT READS

It reads your channels. It answers to your store.

Autopilot reads spend and delivery from your ad platforms and grades all of it against store-verified Shopify orders — read-only, with no write access to your store required to run.

Shopify orders

Read-only sync of orders and customers — the store-verified ledger every channel is graded against. No write access to your store is needed to run.

Meta

Live today. Spend, delivery, and campaign structure read from your connected ad account.

Google

Built, and waiting on developer-token approval for production accounts before it reads live.

TikTok

Not read. There is no TikTok client — the gateway refuses the provider, there is no connection to make, and a TikTok sync would finish as an explicit no-op. It is a reserved slot in the schema and nothing more.

Order-ID match

Orders matched back to the campaign that earned them by order ID, inside the attribution window.

Only what it can attribute

A channel is graded on the orders it can be tied to. What can’t be matched is never counted.

02 — DECIDE

It makes the case. You make the call.

Proposal, decision, outcome — one auditable trail, scored against the numbers the decision was made on. The change itself is for you to apply, in your ad platform.

AUTOPILOT SUGGESTS

Cut budget — Prospecting, Broad

META · $180 → $153 / DAY · −15%

VERIFIED CAC$63
TARGET CAC$42
SPEND · ACQUIRED$1,260 · 20
Apply in your ad platformDismiss
APPROVAL QUEUE3 PENDING
CUT BUDGETPENDING
SCALE BUDGETPENDING
PAUSE CAMPAIGNPENDING
RETRICS EXECUTIONOFF · YOU APPLY IT
OUTCOME7 DAYS AFTER
VERIFIED CAC AT DECISION$63
VERIFIED CAC NOW$46
VERDICTIMPROVED · −27%

OBSERVATIONAL BEFORE / AFTER — NOT A CONTROLLED TEST

CONCEPT PREVIEW · NUMBERS ILLUSTRATIVE

03 — THE AUTONOMY LADDER

Autonomy is earned. One rung at a time.

Rung one is where Autopilot is today: it recommends, and you apply it yourself in your ad platform. Rungs two and three are the roadmap. Retrics executing on your ad account is switched off in code, and no plan, setting, or approval turns it on.

SuggestACTIVE

Recommends. Touches no ad account.

RUNG 01 · SUGGEST · LIVE TODAY

It recommends. You apply it.

  • · PROPOSALS → APPROVAL QUEUE
  • · SPEND UNTOUCHED
  • · YOU APPLY IT IN YOUR AD PLATFORM
ApproveLOCKED

Not enabled. Retrics would apply it.

RUNG 02 · APPROVE · NOT ENABLED

One tap — once it is built.

  • · APPROVAL WOULD APPLY THE CHANGE
  • · RECEIPT WITH ITS EVIDENCE
  • · OUTCOME SCORED AFTERWARDS
Narrowly armedLOCKED

Not enabled. Would act per class, inside caps.

RUNG 03 · NARROWLY ARMED · NOT ENABLED

Earned one action class at a time.

  • · PER-CLASS ARMING
  • · LAUNCHES WOULD BE CREATED PAUSED
  • · RAILS NEVER RELAX

ARMING BY ACTION CLASS · RUNG 03 DESIGN

Cut budgetFIRST TO ARM
Scale budgetAPPROVAL
Pause campaignAPPROVAL
Launch campaignPAUSED ON CREATE

CONCEPT PREVIEW · NUMBERS ILLUSTRATIVE · ARMING IS SWITCHED OFF IN CODE

THE RAILS UNDER EVERY RUNG

BUDGET LIMITS · YOUR POLICY

MAX SINGLE CHANGE15%
FLOOR · CEILING$60 · $400
ACCOUNT DAILY CAP$2,400
MIN SPEND TO RECOMMEND$20
≤ 10 AUTO ACTIONS / DAY · ARMED RUNGS+50% WEEKLY DRIFT CAPSPEND CIRCUIT BREAKER

YOU SET THE TARGETS AND CAPS · THE FLOORS BENEATH THEM ARE NOT MERCHANT-EDITABLE

KILL SWITCH — ALWAYS LIVE

KILL SWITCH STOPS EVERY PROPOSAL — AND EVERY AUTONOMOUS ACTION AT THE LATER RUNGS → ONE TAP · NOT EDITABLE AWAY

WHAT’S TRUE TODAY

01

Early access

Every surface above is concept UI. Autopilot recommends today and applies nothing — approval-based application and narrow arming are the next two rungs, switched off in code. The suggest → approve → arm model, the rails, and the floors are the engine’s actual control design.

02

Meta first, Google next

The Meta client runs today. The Google client is built and waits on developer-token approval for production accounts. TikTok has no client at all — no connection to make, and a sync that finishes as a no-op — so no TikTok row appears above and no TikTok spend is read.

03

Audiences, then launches

Winback, second-order, VIP-seed, and replenishment segments push to Meta as hashed custom audiences — Meta is the only platform that receives an upload. Suppression lists are never a target.

See pricing →

QUESTIONS

The honest answers.

What is Retrics Autopilot?

Autopilot is the ad-spend layer of Retrics. It reads your ad platforms — Meta today, Google once its developer token is approved for production accounts — and grades every campaign against store-verified Shopify orders, not the returns each platform reports for itself. TikTok has no client, so nothing from TikTok is read. It recommends the move, with the case attached, for you to apply in your ad platform. It is in early access.

What does Autopilot diagnose?

For each campaign it sets the ROAS the platform claims against the ROAS verified from store orders, measures how much of that spend can be matched back to real orders by order ID, and looks at whether the customers a campaign bought ever came back to order again. Every figure is store-verified revenue — Retrics holds no product-cost data, so nothing is graded on cost.

Does Autopilot change my ad spend on its own?

No — and not on your approval either. Autopilot is suggest-only: it puts a recommendation in a queue with the numbers behind it, and it does not create, edit, pause, or budget a campaign in your ad account. Approving one records your decision; you make the change in your ad platform. Retrics making the change for you, and later acting alone inside limits you set, are the next two rungs, and both are switched off in code.

How does the suggest → approve → arm model work?

Three rungs, in order, and only the first is built. Suggest, live today: it recommends the move, stores the case verbatim, and touches no ad account — you apply the change yourself in your ad platform. Approve, not enabled: Retrics would make the approved change for you and score the result afterwards against the numbers the decision was made on. Narrowly armed, not enabled: a single action class — a budget cut, say — could run inside caps while everything else waited for your approval. Autonomy is earned one rung at a time, never assumed.

What limits and safety rails are there?

You set the targets and caps — a maximum single change, a budget floor and ceiling, an account-wide daily cap, and a minimum spend before it will recommend a change. Beneath those are floors that are not merchant-editable: a cap on auto actions per day, a weekly drift ceiling, and a spend circuit breaker. A kill switch is always live: one tap stops the engine, so it proposes nothing now and, at the later rungs, does nothing.

What data does Autopilot read?

Read-only spend, delivery, and campaign structure from your connected ad platforms — Meta today, Google once its token clears, and nothing else — and read-only orders and customers from Shopify, the store-verified ledger every channel is graded against. It needs no write access to your store to run, and a channel is only graded on the orders it can attribute.

What can’t Autopilot do?

It can’t grade a campaign on cost — Retrics holds no product-cost data, so everything is measured on store-verified revenue, never on what a sale earned. It can’t touch your ad account: it does not create, edit, pause, or budget a campaign, and no plan, setting, or approval turns that on — the change is yours to apply. It can’t relax its own non-editable floors or switch off the kill switch.

Is Autopilot available yet, and which platforms are supported?

Autopilot is in early access. The Meta client runs today; the Google client is built and waiting on developer-token approval for production accounts. TikTok has no client — there is no TikTok connection to make, and a TikTok sync finishes as an explicit no-op — so no TikTok spend is read, graded, or attributed anywhere in Retrics. Request early access to join the first cohort.

Every ad dollar, reporting to the ledger.

Suggest today. Approval next. Autonomy last, and only inside the rails.

See pricing →