Attribution tells you what got credit.
We tell you what caused it.
Causal measurement, standing experimentation, and a first-party data foundation built for the deadlines that are actually coming.
The shift
The deadline was cancelled.
The crisis wasn’t.
The industry spent years preparing for the end of third-party tracking. Then the deadline was called off, and a lot of marketers read that as a reprieve.
It wasn’t. Several major browsers had already stopped playing along. Consent enforcement tightened. And people simply started saying no more often than they used to. The deadline died; the measurement problem arrived anyway, through the side door.
So the industry re-platformed regardless — onto modelling for allocation, experiments for validation, and platform-reported numbers demoted to running the day rather than deciding the year.
Adoption is broad and shallow. Most marketing teams now run tests. A large share of them don’t trust their own results. And only a small minority say they can turn a modelling insight into an actual decision. That last gap is the whole opportunity. It isn’t a tooling problem — the good tools are free. It’s a decision loop that somebody has to make mandatory.
The credible open-source modelling field has narrowed considerably, and the tool still being actively developed is built by one of the largest sellers of the media it measures. That’s a genuine concentration risk. It belongs in your decision rather than buried in ours, so we build in a way that lets the model be rebuilt elsewhere.
Client challenges
SOUND FAMILIAR?“Every platform claims the same conversion and our CFO has stopped believing all of them.”
Correctly. We rebuild measurement as triangulation — a model for allocation with its uncertainty stated, experiments to validate it, and platform numbers demoted to operations. The CFO gets one number and an honest range around it.
“We test constantly and we’ve never changed a budget because of a test.”
Then you have experiments, not an experimentation programme. The difference is a standing share of budget set aside for it, hypotheses registered before the test runs, a fixed cadence, a decision log, and a hard rule that the model gets updated by the results. It’s a governance practice with good plumbing — not a product.
“We have first-party data in six systems and consent records in none of them.”
A unified customer view, identity resolution, and a consent architecture built to what the law actually requires — including notices in the languages your customers use. This one has a clock on it, so it goes to the front of the queue.
“Data protection is on the risk register and marketing owns it.”
India’s data protection regime tightens through the back half of this year and lands fully next year, with penalties large enough to be a board conversation. We build the consent and governance layer as part of the data foundation rather than as a compliance project bolted on afterwards — which is cheaper, and the only version that survives an audit.
“Our acquisition cost has doubled and we can’t tell which half is working.”
Common right now, and getting worse — paid acquisition economics across Indian D2C have deteriorated sharply, with returns on cash-on-delivery orders adding a second squeeze on working capital. We measure incrementally, reallocate against what the evidence supports, and build the lifecycle system that makes retained customers carry more of the load.
What we build
Measurement Health Check
The way in, and deliberately small. An audit of the inputs — data completeness, weekly granularity, spend and outcome hygiene, consent coverage — and a plain answer to one question: what can your current setup actually prove, and what is it only claiming? Sometimes the finding is that modelling should wait. We’d rather say that in week three than in month four.
Measurement Architecture
Design of the full stack: modelling for allocation, geographic experiments and holdouts for validation, platform attribution demoted to in-flight operations. Follows the health check, because the inputs decide whether it’s worth building.
Marketing Mix Modelling
A model built on your own history, reported with its uncertainty stated rather than as a single confident number, with a planner attached so you can ask what-if questions. Genuinely within reach for a mid-sized brand — the constraint is data hygiene, not budget.
Incrementality Programme
The operational form of always-on experimentation. A standing budget carve-out, a backlog of hypotheses registered in advance, a fixed cadence of geographic and holdout tests, automated readout into a decision log, and the rule that the model gets updated by what the tests find.
First-Party Data Foundation
A unified customer view, identity resolution, consent architecture and governance — built to clear the enforcement dates rather than arrive just after them, with notices in the languages you operate in.
Lifecycle & Retention Systems
CRM, loyalty and lifecycle built as predictive systems rather than batch campaigns — investing in customers based on what they’re likely to be worth, not what they last bought.
Command Centre
A live decision surface instead of a monthly deck. The causal read, the state of every running experiment, the visibility trend, content throughput and cost per asset, in one place.
How we report
Ranges, not single numbers
pretending to be facts.
The buyers for this work are sceptical by training, and they should be. So we do three things differently.
We publish the method, in full, including its limits.
In a field where a large share of practitioners say they don’t trust the data available to them, that’s worth more than any claim about a result.
We report with uncertainty attached.
A model output is a range. Presenting it as a point estimate is the most common way measurement work gets discredited about six months in.
And we say what we don’t know.
If your data can’t support a conclusion, that’s the finding. We’d rather deliver an uncomfortable answer than a confident one we can’t stand behind.
FAQ
Frequently asked
questions.
Yes, if you have a couple of years of clean weekly spend and outcome data. The tooling is open and free. The constraint is data hygiene and the discipline to read a range as a range.
Then we start with geographic experiments and a decision log, and build the modelling later. Running a weak model on thin data is worse than running none, because people believe it.
No. We guarantee you’ll know which decisions the evidence supports and which it doesn’t. Anyone guaranteeing causal lift is either funding that guarantee out of your own budget, or hasn’t thought about it.
We build the marketing-side data and consent architecture to meet what the law requires, and we work alongside your legal counsel. We aren’t lawyers and we don’t sign off on compliance. We’d rather say that here than later.
Find out
what your data can actually prove.
Two to three weeks. Fixed fee. An honest answer about what you can measure, and what you’ve only been claiming.