Revformance
How it works The Case Outcomes Trust Pricing Privacy policy Terms and conditions
Read-only access. Revformance never changes your campaigns.

Something moved in the account. Know why before the client asks.

Revformance connects Google Ads and Meta Ads, finds the changes that matter, and explains each one with the evidence behind it. You accept, edit, defer or reject — then it measures what actually happened.

Early access, no card Shape what we build first Read-only OAuth scopes
Reads from Google Ads Meta Ads Google Sheets CSV upload Never writes back

The hard part was never finding the number. It was explaining it well enough that someone would act.

Every Monday, a strategist covering fifteen accounts spends two hours reconstructing what moved.
Revformance does the reconstruction, and shows its working.

How it works

Six steps, and the honest limit of each one.

Revformance is a loop, not a dashboard. Everything below is what the product does today — not a roadmap, and not a demo that quietly needs a human behind it.

01 — Connect

Read-only, in minutes

Google Ads and Meta Ads through read-only OAuth, or a CSV with column mapping if you cannot authorise platform access. Ninety days of history backfills on first connection.

The limit: a manual CSV carries no attribution settings or delivery status, so four of the fourteen checks cannot run on it. We say which, on the account.

02 — Detect

Fourteen deterministic checks

Cost per acquisition, return on ad spend, conversion volume, spend without matching conversions, click-through, cost per click, frequency, delivery, budget constraint, tracking integrity, stale data, missing data, variance.

The limit: a check that cannot run is reported as a check that did not run — below minimum volume, missing data, or a mute you set. The detection log lists every run.

03 — Explain

Arithmetic ours, prose the model's

Where a check fires, Revformance assembles the evidence deterministically and asks a language model to explain the pattern over figures it was handed. Nothing is calculated by the model.

The limit: a model can be wrong about cause. That is why observations and hypotheses are visually separate, every hypothesis carries its contradicting evidence, and confidence is a word rather than a false-precision percentage.

04 — Decide

You accept, edit, defer or reject.
There is no fifth option.

Rejections take a structured reason, which is what teaches the system what your team does not want to see again. There is no bulk accept and no automatic execution — a recommendation only becomes work when a person with the right role approves it.

Accept Edit before accepting Defer Reject Already known
05 — Act

A tracked action, with a name on it

An accepted recommendation becomes an action item with an owner, a due date and an audit trail of who approved what. You make the change in the platform. We record that you did.

06 — Measure

Seven, fourteen, thirty days later

Target metrics compared against the fourteen days before, alongside a trend control for the rest of the account. Then we tell you what happened, including when the honest answer is that we cannot tell.

The artefact

Everything lands in one place, called a Case.

Alert, evidence, diagnosis, recommendation, decision and outcome — one object, one audit trail, one vocabulary across the app, the API, the emails and the client report.

Evidence is never more than one interaction away

Every figure in a diagnosis is a citation that opens the stored evidence row it came from, with its period, its source and its definition. If you have to trust the product, the design has failed.

Every comparison states its period in visible text
Not a tooltip. "15–28 Aug against the preceding fourteen days, matching weekdays."
Every metric carries its definition
Reachable from the label anywhere it appears, so two people never argue about what CPA means.
Every silence has a reason
"Eleven of fourteen checks ran, three could not — here is why." Never a bare "all good."

How Revformance reached this

A disclosure sits on every Case, permanently. Not a legal footnote — the actual working.

Rule and threshold
Cost-per-acquisition increase, v1.3. Effective threshold 25% over fourteen days — a client override set by Arjun Rao on 4 July, against a 30% workspace default.
Calculated by Revformance
CPA as spend ÷ conversions for both windows, period-over-period deltas, frequency and CTR deltas, and every severity input.
Sent to the model
Metric names and values, comparison windows, entity names, attribution settings, data-quality flags. No credentials. No client personal data. The field list is printed, not summarised.
Constraints

Four things we will not do.

Most of what makes this product trustworthy is what it refuses to do — and refuses to add later because a competitor shipped it.

Change your campaigns

We request read-only scopes. There is no execute button anywhere in the product, and no roadmap item to add one without explicit, per-action consent.

Assert a cause we cannot prove

A correlation is presented as a correlation. Every hypothesis carries its contradicting evidence and a way for you to check it yourself.

Hide what we do not know

Missing days, partial data, low conversion volume and attribution changes sit next to the number, not in a footnote nobody reads.

Promise a result

We estimate impact and label it an estimate. Then we measure the actual change and publish it — including the times it was inconclusive.

Outcomes

The number most tools will not show you.

Anyone can report their wins. Revformance reports inconclusive and confounded results beside them, because a success rate that hides them is a marketing number, not a measurement.

Acme Retail · last 90 days · 10 implemented actions
4
Improved
3
No material change
1
Declined
2
Inconclusive
1
Confounded

Measured, not asserted

Each action is compared against the fourteen days before it shipped, at seven, fourteen and thirty days, alongside a trend control for the rest of the account. Where another change lands in the same window, the result is marked confounded and names it.

Stated on every outcome: this is a before-and-after comparison over a fixed window. It does not establish that the action caused the change.

ACT-297 · Shopping budget increased 20% on 18 Aug
Baseline CPA₹904
7-day after₹831
14-day after₹795
Trend control−2.4%

Lower is better for CPA. The rest of the account moved −2.4% over the same window, so most — but not all — of the 8.1% improvement sits outside the general trend.

Trust

What we access, and what we have not done yet.

Connecting an ad account is a decision your client is trusting you with. Here is everything you need in order to answer for it.

In place today

Read-only by design
The narrowest read scope each platform offers. Revoking at Google or Meta stops all syncing immediately.
Per-tenant encryption
OAuth tokens encrypted with keys scoped to your workspace, never exposed in the interface, exports or logs.
Five roles, enforced server-side
Owner, admin, manager, analyst, viewer — scopable to individual clients, not just hidden in the UI.
Immutable audit log
Every consequential action, with before and after, exportable. Support impersonation needs your consent and shows a banner.
Data residency in India and the EU
Region pinned per workspace, with subject-access and deletion runbooks for DPDP and GDPR.

Not yet, and we will say so

A young product that claims a certification it does not hold is not a trustworthy one. This list changes when the state changes, not when the marketing calendar wants it to.

In progress
SOC 2 Type II
Controls implemented, observation window under way.
Planned
Third-party penetration test
Scheduled for Q4 2026, with the summary shared on request.
Partial
WCAG 2.2 AA
A defined subset committed today, full audit scheduled — the accessibility statement names exactly which criteria.
Available
Single sign-on, SAML
On the Scale plan.
Pricing

Priced on connected accounts.

No per-seat tax on the person who only reads the report. Every plan includes all fourteen checks, the full evidence trail and outcome measurement.

Studio
₹14,000/ month

For a small team running a handful of accounts.

Up to 10 ad accounts
5 team members
Daily detection, all 14 checks
Weekly client-safe reports
Email digests, 90-day history
Join the waitlist
Scale
Talk to us

For larger portfolios and in-house teams with review requirements.

25+ accounts, unlimited members
Single sign-on (SAML)
Data residency choice
Custom retention
Onboarding and quarterly review
Join the waitlist

Pricing is indicative until we open paid plans. It is published now so you can judge whether this is built for a team your size, and waitlist members will be told before anything is charged.

When you reach a limit, nothing is silently truncated. Connecting an account beyond your plan is refused with the limit and your current usage stated, and an upgrade path. If your monthly AI diagnosis quota runs out, detection, evidence and reports keep running — only new written diagnoses pause, and the product says so on the Case.

We are opening early access
a few teams at a time.

Tell us which platforms you run and roughly what you spend. We are sequencing invitations by the connectors people actually need first, so what you tell us changes what we build next.

Read-only OAuth · India and EU data residency · We will only email you about early access