01
Defined windows
A baseline and a test window agreed in advance, so the comparison is not chosen after the fact.
The framework
How an engagement actually runs.
Six stages, in sequence. The purpose of writing them down is that both sides know what happens next, what has to be true before the next stage starts, and who signs off.
Test cadence
STAGE 01
Discovery & offer fit
We start with what exists rather than what is aspirational: your role in the chain, the offers or traffic you already hold, capacity, compliance posture, and the outcome you are being measured on.
In practice
- Intake call covering role, verticals, geos, and constraints
- Review of current offers, traffic sources, or both
- An early, direct read on whether there is a fit worth building
Some discovery calls end with us saying no. That is a successful call.
STAGE 02
Partner and geo alignment
We match the opportunity against partner terms and the rules of each market in play. Availability, payout structure, and creative restrictions differ by geo, and those differences decide what is actually runnable.
In practice
- Offer availability confirmed with the partner, per geo
- Category and platform-policy constraints documented up front
- Payout structure, caps, and restrictions written down before work starts
We work globally. Offer availability depends on partner terms and local rules.
STAGE 03
Tracking & creative setup
Measurement is built before spend. Pixels, postbacks, parameters, and naming conventions are configured and validated, and creative is produced for the destination environment rather than adapted to it later.
In practice
- Pixel, postback, and server-to-server configuration with your platforms
- Naming and parameter conventions so sources stay separable
- Native-friendly creative and offer pages reviewed against offer terms
Advertorial-style pages are labelled as advertisement or sponsored content.
STAGE 04
Controlled launch
The first push is deliberately small and time-boxed. A defined test window with a defined read is worth more than an open-ended launch that nobody can interpret afterwards.
In practice
- Limited initial volume across a defined window
- Event validation confirmed live before scaling anything
- Agreed checkpoints and a stated definition of a readable result
Timelines vary by partner, platform review, and market.
STAGE 05
AI-assisted test suggestions, human approval
Model-supported analysis reviews creative, landing experience, and conversion patterns and returns a ranked queue of suggested tests with reasoning attached. An operator checks each against offer terms and compliance rules and decides what ships.
In practice
- Suggested tests ranked, with rationale and a measurement window
- Operator review against offer terms, platform policy, and claim limits
- A written record of what was tested, what shipped, and what was rejected
AI-assisted recommendations are decision support. Operators review changes before they go live.
STAGE 06
Reporting and next decisions
Reporting arrives with its context: the window, the traffic mix, the known variables, and an honest statement of where the read is weak. Then we decide together what the next iteration is.
In practice
- Reporting on an agreed cadence with definitions written down
- Interpretation separated clearly from measurement
- A recommended next action, including stopping when that is the right call
Results vary by offer, geo, traffic source, creative, competition, and budget.
Reporting
What a before-and-after review looks like.
At the reporting stage we compare a defined baseline window against a defined test window, metric by metric, and state what changed and what we cannot yet attribute. The layout below is an example of that structure.
Sample layout — not client results
02
Stated variables
Traffic mix, creative changes, seasonality, and platform changes noted alongside the numbers.
03
Honest confidence
Where the sample is thin or the attribution is contested, the report says so rather than rounding up.
Governance
Where the human sits in the loop.
The review queue is the control point. Nothing reaches a live offer path without an operator reading the suggestion, checking it against the offer terms and platform policy, and approving it by name.
- Model output is a ranked queue of suggestions, never an executed change.
- Operators check claims, restrictions, and category rules before approval.
- Rejected suggestions are logged with the reason, which improves the next round of review.
- Compliance and legal questions escalate to people, not to a model.
Interface illustration of the operator review queue. AI tools support analysis only.
Next step
Ready to run stage one?
Discovery is a conversation, not a pitch. Bring your role, your markets, and your constraints, and we will tell you what stage two would need to look like.
AI-assistedHuman-approvedCompliance first
- Tell us whether you are an advertiser, publisher, affiliate, or agency.
- Share the verticals, geos, and volume or budget range you are working with.
- We reply with an honest read on fit — including when there is not one.
Offers and campaigns are subject to partner, publisher, and platform review. We do not guarantee volume, EPC, approvals, or results.