Increasingly more suppliers are providing automated market combine modelling (MMM) methods utilizing AI/machine studying as cookie-based attribution declines – right here, 17 specialists have revealed an open letter warning that many such methods should not nuanced sufficient to supply efficient measurement.

Pricey mates,

We analysis specialists care deeply about you getting the very best response out of your promoting.

We’re within the MMM/econometrics enterprise as a result of we love crunching the numbers and fixing a puzzle. However we’re additionally right here as a result of we wish to see change in the true world, within the type of development in your companies.

It’s on this spirit that we’re writing to you. To warn you. As a result of not all evaluation that appears stable from the surface is equally good.

The factor is, your companies are distinctive and multi-faceted organisations, working imperfectly in contexts no different agency has skilled earlier than. And good analysis that may genuinely untangle gross sales pushed by adverts, has to account for that complexity correctly.

Equally, and we hope you don’t thoughts us saying this, though you might be sensible at quite a lot of issues, you aren’t at all times that good with information. You generally report it in messy methods and also you don’t at all times know for positive what every quantity means or how nicely it’s measured.

All this implies there isn’t a single strategy to analysis that works in each circumstance. The ways in which your world might be totally different or your information might be mucky or lacking are too quite a few and curious and sophisticated to deal with with “if this then that” code.

Even AI can’t establish what isn’t within the dataset it’s . And it will probably’t speak to your workforce concerning the time you misprinted the barcodes after which innovate a manner to make use of your information to seize the impact. Or discover out that you simply ran a big radio marketing campaign independently and didn’t inform anybody.

Don’t get us mistaken. We love a little bit of code as a lot as the following particular person, and doubtless extra. All of us use it to automate information assortment and prep, and we automate producing customary outputs too.

However there are some bits that may’t be automated. Ensuring fashions actually mirror your enterprise, getting the nuances of what occurred up to now straight, and dealing along with your folks to get findings about massive expenditures acted on are all issues that require actual life human beings.

We’re sorry we didn’t flag this with you sufficient when last-click attribution was new. It wasn’t as a result of we didn’t care, it was that each time we tried to say, we had been accused of being dinosaurs or luddites. We’re gutted about how a lot you spent serving adverts to individuals who had been already on their approach to you.

In order that’s why we wished to jot down to you now. As a result of there’s hazard on the horizon once more.

With cookies disappearing, platform-based, automated variations of MMM are coming to market to resolve the issues with attribution.

We’ve appeared on the algorithms and we’ve got to inform you, they’re a lot less complicated than they should be. With MMM, each time you don’t embrace one thing that issues you get a mistaken quantity for the impact of promoting, and there are fashions on the market that don’t even embrace COVID-19 or worth.

We maintain getting folks that have invested in these platforms coming to us saying that the numbers make no sense, that no-one believes them, that the 9-12 months it took to get arrange had been wasted.

Now, we’re not that vocal a bunch normally, and we don’t usually provide recommendation until we’re requested for it. However if you wish to know what we recommend you do, it’s store for MMM such as you’re shopping for a brand new kitchen. Ask round for suggestions, get at the least 3 quotes, and don’t be too trusting.

Ask these questions and look to see if the particular person you’re speaking to squirms earlier than answering sure:

Does the mannequin embrace components like worth, financial system, seasonality, covid? Will you report on how this stuff have an effect on our enterprise?

Does the mannequin cowl at the least 2 years of information, ideally 3?

Do you measure how higher funnel adverts like TV & YouTube have an effect on outcomes in decrease funnel adverts like PPC?

Will you share promoting response curves with our media planners?

What would occur if outcomes got here again and we didn’t imagine them, say as a result of they didn’t line up with one thing else we knew?

Will you be capable to clarify the mannequin to our Finance folks? Will your numbers line up with theirs?

Might our analysts who perceive regression look below the bonnet on the mannequin and all of the exams and statistical due diligence?

Are you able to exhibit to them that your fashions are good at forecasting?

We’ll be rooting for you.


Les Binet (Adam & Eve DDB)

Grace Kite (Magic Numbers)

Louise Cook dinner (Holmes and Cook dinner)

Mike Cross (Measure Monks)

Andrew Deykin (D2D)

Matt Andrew (Ekimetrics)

Neil Charles (ITV)

Sarah Stallwood (Magic Numbers)

Sara Jones (Pearl Metrics)

Jamie Gascoigne (Measure Monks)

Pleasure Talbot (Magic Numbers)

Simeon Duckworth (WPP)

Sally Dickerson (Benchmarketing)

Stuart Heppenstall (D2D)

Dominic Charles (Wavemaker)

Steve Hilton (Measure Monks)

Tim Fisher (Measure Monks)