Looking Back to Go Forward – How Measurement Can Fuel Prediction

As marketers, we’re constantly trying to navigate a reliable course through an increasingly unpredictable media sea. Conditions are continually shifting as we encounter new channels, fragmented audiences and changing signals of attention and consent. In that turbulence, measurement has traditionally been our compass: it tells us where we’ve been and what has worked.

Traditionally, research has been backward-looking: we measure, we analyse, we report. But what if measurement could help us anticipate performance? If we could move from ‘measurement as proof’ to ‘measurement as signal’ – helping brands and publishers steer campaigns while they’re still in motion?

The shift from retrospection to active prediction

We’re not the first industry to wonder about this. Wherever there is data in sufficient volumes, we’ve seen the same ambition forming. Self-driving cars learn from millions of hours of driving data to anticipate what’s coming next on the road. In healthcare, AI predicts disease risks by analysing patient histories. Finance models forecast risk and opportunity from decades of market data.

Today’s advances show how unstructured data can be harnessed at scale. Yet, when scale meets consistency, the results are even more powerful. Our industry has been quietly building a structured data set that doesn’t just grow in volume, but in reliability, making future predictions sharper and more credible.

The power of the data we hold

Fortunately, we now have structured data sets that catalogue many thousands of campaigns across dozens of markets and, crucially, have been accumulated consistently over the years, using the same framework and taxonomy, capturing not only results but the full context around them.

Effectively, data of this kind gives marketers access to a detailed historical chart of how different media environments impact brand outcomes. And in addition to outcomes, every campaign can yield opportunities to collect a wealth of passive data relating to ad exposure, engagement and brand context.

When you gather data in a consistent way over time, patterns begin to emerge. You begin to see how formats, contexts, and creative choices interact with brand goals. That consistency is what turns data from something descriptive into something that is actually capable of prediction.

When such data is cleaned, structured, analysed and tested, it offers the possibility of building predictive models at two levels:

• Campaign-level models that forecast brand outcomes, when enough is known about the campaign set-up
• Ad-impression-level models that aim to predict the brand effect of individual exposures in real time

Given continuous refinement and validation, these predictive models can in turn be applied at three stages of a campaign:

• Before a campaign – to recommend the most effective plan or mix
• During a campaign – to help decisioning systems prioritise impressions likely to drive brand lift
• After a campaign – to estimate directional impact even without new survey data

Each use case points toward a world in which brand outcomes are live, not retrospective. To make this work in the real world, the predictive signal needs to be integrated with the platforms that already power ad decisions: DSPs, SSPs, verification and optimisation partners.

The aim of this needn’t be to sell data, but to make insights available within existing workflows, safely and at scale – a collaborative ecosystem approach, built on open standards and privacy-first principles.

Marketing that learns and adapts

Data of this kind, properly handled, can play a part in creating a new outcomes-focused future. Consistent measurement shows us where we’ve been; prediction points to where we should go next. Combined, they create a much fuller system for marketing investment – one that learns and adapts with every campaign.

Of course, no model is perfect. Predictions are most accurate when we are in familiar waters – known brands, known formats, known publishers. When we move too far from that territory, uncertainty grows. That’s why continually increasing the scale, range and granularity of the data set is critical to better navigation. This is a continual process of improvement.

And for media owners, this opens up new possibilities. Their own inventory data enriches the shared picture, improving predictive accuracy for everyone. In return, it provides them with the ability to sell inventory that is objectively optimised for brand outcomes, not just impressions or clicks.

So the challenge is this: can we, as an industry, move from retrospective measurement to predictive insights? Can we turn brand outcomes into a live signal that shapes how we buy and sell media? The potential is certainly there if we choose to build it together.

After all, you can’t change the direction of the wind, but you can adjust your sails. With the right data and the right collaboration, we can all do exactly that and navigate the choppy waters ahead.

Also published in: Advertising Week