Measurement Matters

Welcome to the first edition of Measurement Matters: a regular look at how businesses can take better actions from data.

I’m starting with a topic that sits at the heart of that: triangulation. Because the biggest mistake in measurement is often assuming there should be one definitive answer.

Siloed results: the case for triangulation

Most marketing teams do not have a measurement problem, they have a trust problem.

Last click says one thing, MMM says another, a lift test lands somewhere else again. Finance sees the disagreement, decides none of it is reliable, then goes back to the one number it can access every day.

Usually, that number is last click.

Speaking to a few clients recently, I think multi-touch attribution is on its way out. Not because last click is more accurate (MTA tends to be c. 10% more accurate than last click) but because MTA is often hard to explain, hard to maintain and still unable to answer the core commercial question: what was genuinely incremental?

That leaves a familiar gap.

Marketing has models. Finance has reports. Neither side has a single view they are prepared to use to make a decision.

Last click is wrong, but it has a job

Last click gives credit to the final recorded interaction before a conversion.

That means paid search often looks great. It is close to the sale, easy to see in the data and available at the right level of detail to manage activity daily.

It is also consistently misleading.

Last click misses the brand activity that created demand in the first place. It gives little credit to channels that work over time. It cannot deal well with one channel making another more effective, such as social activity lifting search demand. It also tells you nothing about whether the sale would have happened anyway.

Still, it has strengths that should not be dismissed:

It is granular It is frequent It is already embedded in most organisations It is easy to explain It helps teams spot what is changing now Its free

• It is granular

• It is frequent

• It is already embedded in most organisations

• It is easy to explain

• It helps teams spot what is changing now

• Its free

The mistake is not using last click, the mistake is treating it as the answer.

MTA has become an expensive middle ground

MTA promised a more sensible way to divide credit across the customer journey.

In theory, that makes sense. In practice, the underlying data has become less complete and less dependable. Walled gardens limit what can be observed, cookie deprecation has reduced visibility, iOS changes have altered tracking, and the real world rarely follows a tidy sequence of touchpoints.

A customer may see a brand campaign, hear about a product from a friend, search two weeks later, receive an email, then buy after clicking a paid search ad. Trying to assign neat fractions of credit to every interaction can give the appearance of precision without resolving the underlying uncertainty.

More importantly, MTA is not incremental. It can describe paths to conversion, but it cannot reliably tell you which activity created sales that would not otherwise have happened.

That is the number a CFO needs.

So rather than spending time and budget building another attribution model that people will struggle to trust, I would keep the useful part of last click and add the evidence it cannot provide.

Different methods should do different jobs

The answer is not to choose a winner between last click, econometrics and lift testing.

It’s to use each for the job it is best placed to do, then bring the results into one commercial view.

Last click gives you a fast, detailed read of what is happening across campaigns, audiences and keywords.

Econometrics, or MMM, gives you the wider picture. It measures incremental sales across brand and performance activity, accounts for the effects that linger beyond the week of spend, and can expose interactions between channels. Done properly, it also places media alongside price, promotions, distribution, seasonality and the other things that move sales.

Lift tests give you direct evidence. They test whether activity changes outcomes in a controlled way, which makes them especially useful for resolving the areas where people have the least confidence.

Recommended by LinkedIn

Each approach has weaknesses on its own.

Last click is fast but biased.

MMM is broad and incremental, but is traditionally seen as slow and blunt: cannot tell you what happened to a specific campaign yesterday morning.

Lift tests are persuasive, but they take time to build into a useful bank of evidence across channels, markets and campaigns.

Pull these techniques together, the weaknesses cancel each other out.

Start with a calibrated daily read

A practical approach is to use econometrics and lift tests to estimate the gap between observed last-click results and incremental value.

You may find that paid search still performs well, but not to the extent shown in the platform data. You may find that brand activity creates a meaningful share of the search demand that last click awards to performance media. You may find that a campaign with a weak last-click return is creating sales over a longer period, or improving the return from other activities.

Those findings should not live in a model readout or a separate dashboard.

They should adjust the numbers people already use.

That gives the business a calibrated daily read rather than a choice between simplistic reporting and an analysis that arrives too late to act on.

It also makes the conversation with finance much easier. Instead of asking them to accept a black box, you can show the link between the familiar daily data, the broader econometric view and the test evidence used to verify it.

Triangulation executed practically

Bringing the different measures together should focus on not just the measures, but also the confidence of the read.  You want to lean the result to your more confident measure.

One way to do this is to create a weighted average between your most reliable measures, lift test and econometrics / MMM, using the confidence level as the weight between the two ROIs.  This way the result is tending to the more reliable and trusted measure.

Then compare this confidence weighted, consolidated ROI with the last click read to get to a calibration, which then can be passed back to the platform to get a more realistic ongoing view of ROI.

The aim is fewer arguments, not more models

Many businesses already have enough measurement.

What they lack is a way to reconcile it.

When different techniques produce separate reports, each team defends its own numbers. Marketing quotes the platform, Analytics quotes the model, Finance uses last click because it is visible and repeatable. Decisions slow down while everyone debates whose figure is right.

Triangulation changes the question.

You are no longer asking, “Which model should we believe?”

You are asking, “What does each source tell us, and how do we combine that evidence into the best commercial decision?”

A much more useful discussion.

It reduces duplicated effort, removes the need for another login and gives teams a shared basis for setting budgets, defending brand investment and planning the next move.

So what?

Do not replace last click with another model that creates another argument.

Keep last click for the speed and detail it provides, use econometrics to establish the incremental, cross-channel picture and use lift tests to verify the areas that matter most.

Then triangulate the outputs into one view the business can understand and use.

If your reports currently disagree, the answer is not more dashboards. It is a measurement framework that makes the numbers non-conflicting, commercially credible and useful for action.