In the past there’s been an assumption that econometrics is great at evaluating traditional ATL channels, but weak at understanding the role of digital in the media mix.

This has been for a couple of reasons:


1. Treating all digital channels as a single ‘Digital’ channel

We have seen instances of people combining spend (and/or impression data) for ALL digital channels and adding them to econometric models. Unsurprisingly this leads to misleading (read wrong) results.

All impressions are not created equal. Is one person searching for a relevant term on Google the same as a small display ad served in a poor position on an irrelevant site? Of course not. In above-the-line terms, this would be like adding up all the money you spend on TV, out of home, radio, press and so on, across all campaigns, and calling it ‘advertising’ in your model. You can see why it doesn’t work.

Same goes for Digital. It needs to be treated as a collection of channels, not just a single mega-channel. Obvious, really.


2. Placing our bets too widely

When testing and first deploying digital, clients tends to do so at low amounts of spend, spread over a large number of weeks. One of the selling points of many digital channels is the ability to do this, reducing the amount of marketing budget being put at risk. However, it can cause problems when we’re trying to get a read on their impact using econometrics. Basically, it doesn’t move the sales needle. And if it doesn’t move the sales needle, then no amount of econometric modelling, of any degree of sophistication, is going to register a blip.


So what can we do to get Digital & Econometrics on talking terms?

At MetaMetrics we take a staged, structured approach to deploying and testing digital with our clients, to ensure that they end up in the right place, even if it does take a little longer than taking £100k out of the TV budget and spreading it, wafer-like, across every conceivable digital channel.

What does that mean? Essentially, we work with our clients to devise digital ‘experiments’ which we measure using econometrics. Each experiment comprises a controlled mix of concentrated digital activity, giving each channel enough weight to cut through and, crucially, giving the econometric model the best chance to register an effect. Think of it as test and learn meets econometrics.

The first step is to understand digital at a channel level, by which we mean:

  • Display – broken down into all its myriad forms
  • Paid search – likely split between brand and product/generic
  • Video on Demand (VoD)
  • Social – usually split by platform (e.g. Facebook, Twitter, Instagram) and into paid/owned activity
  • structuring the media plan so that we have a good shot of isolating the role of each channel, so not all deployed for the exact same 6 weeks at the same weight.

Once we have tested at this channel level and discarded the poor performing channels, budget can them be freed up to test more detailed digital tactics. Over the years, we’ve tested a number of different digital areas, including:

  • Programmatic vs. non-programmatic, for both display and video. This was a hot topic in years past. We have typically seen programmatic perform 20 to 30% better than non-programmatic, although there are still some good direct-buy partnerships out there
  • Algorithmically varying budgets in response to external factors. This could be as simple as increasing daily budgets when the weather changes (the age-old, ice-cream-in-the-summer example) through to more advanced trackers of economic factors. These may not boost your ROIs but may allow you to increase your budget without reducing ROI, no mean feat.
  • Different types of programmatic tactics. We have structured and analysed testing plans to understand the role of tactics such as private marketplace, whitelisting, keyword targeting and external data audience purchase
  • Different levels of audience targeting. Is it best to buy mums, or mums with children 18 to 36 months, or mums with children 18 to 36 months who live in the home counties? We’ve tested it.

Quite often what we find is that there is a balancing act that needs to occur. All the really smart stuff is very effective at reducing the wastage of deployment. However, this comes at increased cost and diminished reach, so the trick is working out the sweet spot for your brand.

Are you a hyper-niche brand that needs to be perfectly targeted, therefore some of the more expensive tactics are worthwhile as broader ones will not pick up enough of your audience, or are you a broad church and therefore you should accept some wastage in exchange for cheaper reach?

Most often the truth is somewhere in the middle, some degree of targeting is good, but without the need to go super niche. So, to take an example, for a car insurance aggregator, working against a ‘car owners’ audience is sensible, but going down to the level of ‘car owners whose renewal is in the next month, who have previously used price comparison sites’, whilst removing pretty much all wastage reduces your reach dramatically (missing out on first time buyers, for example) whilst increasing cost. For more on the importance of reach, see our article, ‘Reach, Reach & Reach.’

So, there we have it. With a bit of thought, planning and co-ordination we can use our econometric models to help us understand the impact of our digital activity and get the two talking to each other.


Sam Watts

Sign Up