By Anders Lithner, CEO, Brand Metrics
AI is transforming our world. Self-driving cars, accurate medical diagnoses and many other previously inconceivable things are now very possible. And provided you train your systems on enough data, there are few limitations: LLMs that write convincing copy and create impressive images; the software on your phone that calculates arrival time when travelling long distances through traffic chaos; the app that guesses what unheard songs you will like, or predicts what you will look like when you’re 80.
But what about advertising? Could AI trained on outcomes accurately predict future campaign performance? Surely it can’t be much harder than making a correct medical diagnosis only from a description of symptoms, or sending a car into traffic without a driver? And like those other examples, it is possible – as long as the predictive model is trained on sufficiently large quantities of relevant and consistent data.
We are qualified to say this, because this is data we hold, having consistently measured the brand lift outcomes across more than 50,000 campaigns – always in the same way and using the same four metrics: awareness, consideration, preference and action intent.
Within these categories, we have captured a lot of other data too – more than 50 different campaign signals, including impression levels, unique frequencies, time in view, formats, target segments and industry categories. In all, we hold over 10 million survey responses, with more than 50 billion ad impressions and multidimensional metadata on every campaign.
Individually, these pieces of data may not tell us much, but when you feed the touchpoints from all 50,000 campaigns into your proprietary AI for modelling, it starts to get a lot more interesting.
Predicting campaign outcomes
When we ask our model to predict the outcome of a campaign that has not yet started, and then compare our predictions with the actual measurement results weeks later, our customers’ jaws drop, whispering, ‘how could it know?’ But that’s the logic of large data.
Such data creates a whole new opportunity for media research, transforming retrospective measurement data into a predictive signal that can influence future buying strategies and campaign outcomes.
So, what can it tell us? No, we cannot predict whether an individual knows a brand or not, or the impact of campaign exposure on that person’s brand awareness, consideration, preference, or purchase intent. But we can predict how much a campaign will influence these things on an aggregate level. And we can predict how much a specific ad impression is likely to contribute to this compared to another ad impression.
This means three things:
- Before a campaign we can provide a recipe on how to maximize the outcome by picking the best contexts, placement, formats, targeting, pacing, timing etc.
- During a campaign we score available inventory, to be used as a signal in the programmatic buying and selling of ads.
- After a campaign we can provide strong directional reporting of the campaign impact without actually measuring it.
Outcomes or media quality?
So, the data we have allows us to score according to predicted outcomes. But is that all we need? Or are advertisers better off, as some might suggest, saving themselves this trouble and just investing in the most premium inventory?
In practice, asking whether outcomes or media quality is more important is like asking an athlete to choose between winning the race and prior training. You can’t really have one without the other.
Should I execute my campaign towards the highest media quality inventory per dollar, hoping it will lead to the best outcomes? Or should I execute it towards the inventory with the highest outcome scores per dollar, regardless of the inventory quality?
It depends entirely on how outcomes are defined in the available tools. Is it clicks? Is it some kind of conversion attribution? Often it is these things, and if that’s all that is available, those arguing for quality over outcomes are not wrong. Bidding for low-quality inventory with high clicks is a less profitable strategy than bidding for high-quality inventory with no outcome prediction attached.
But as someone once wisely said, “the only advertising space worth paying for is the space between people’s ears”. Advertising can influence or reinforce people’s awareness about something. It can get into our heads and claim a slot on the shortlist of considered options, or just remind us to hold onto a thought.
It can create or maintain positive attitudes and preference over competing brands or products. And it can create purchase intent – a plan to go and buy. All these things are in the minds of consumers.
Outside of the walled gardens, we provide the world’s only scaled outcomes classification that covers this, because ours is the only company with a sufficiently large and consistent dataset.
The signals we provide are the only outcomes signals on the open web that provides a better bidding tactic than bidding for high quality – the only signal that lets you go beyond training and actually optimise for winning. And right now, this signal is being embedded in key ad tech platforms across the digital supply path.
