The hidden barrier to bringing measurement in-house

Every year a wave of executives leaves Cannes with the same resolution: we need to fix our measurement. Half spent the week being pitched the same converged stack by six vendors. The other half quietly realized they should own this themselves. The second instinct is the right one.

What gets undersold is that most who act on it will fail, and almost never because the model, the systems, or the integrations were too hard to build. The build is the last mile. Knowing what to build, how to sell in the results, and how to create the environment for good measurement to survive are the long poles.

The build is the 20 percent. The talent, the resourcing, and the cultural muscle to act on a hard number are the 80 percent.

The barrier shows up in a predictable order

It rarely appears all at once. It comes apart in three stages, and almost always in the same sequence.

First, talent

The system needs someone who can do the statistics, defend the number to a skeptical CFO or a skeptical client, and recognize when the model is lying because it contradicts an experiment. That person is rare and expensive, and most already sit inside elite in-house teams or the vendors. When you cannot fill the seat, the program stalls before it starts.

Second, resourcing

A measurement system is not a deliverable. It is a living thing. Understaff it and it quietly goes stale: priors drift, integrations break, the taxonomy rots, and the calibration tests stop running. Within a year you are making decisions off a system that describes a business you no longer run.

Third, culture

This is the hardest one. A flawless system returns nothing if the organization cannot absorb its answer. The real test is whether your team can tell an executive that their favorite channel, or a flagship ad product, does nothing, after they have staked years of credibility on it, and get the org to act instead of shooting the messenger.

The prerequisites that kill it before it starts

Underneath all of that sits a layer of unglamorous technical and process work: clean data, disciplined campaign classification, reliable cost aggregation. Any one of them, done poorly, can kill an in-house model before it produces a single number. None of it shows up in a build estimate or a vendor demo, and all of it decides whether your in-house bet compounds or rots.

And no, an AI measurement agent will not fix this. AI can accelerate parts of the build. It does not supply the judgment to defend a number, catch a model contradicting an experiment, or move an organization to act on an answer it does not want to hear.

What actually makes it work

That 80 percent is the center of what we do at Growth By Science. We design the in-house measurement function itself, the roles, the stack, and the operating approach, then carry you through the transition to running it yourself, buy side or sell side. When you want hands-on help, we build the models and wire the integrations with you. We have done this in our own careers, and with enterprise clients on both the buy and the sell side.

The resolution you flew home with is the right one. Making it survive contact with your own organization is the hard part, and that is what we build for.

Own your measurement.