Interviews, insight & analysis on Ecommerce

Could machine learning fill the web cookie gap?

Mark Bakker, Regional Lead Benelux,

For twenty years, marketers have been building up more and more of a key resource: data—both online and off.

The problem is that very soon a lot of this will cease to be as useful to you as it once was… and you’re going to have some real challenges as a result.

First, let’s look backwards to understand all this. Since the start of the Web as a tool for the public, large amounts of data, such as weblogs, email clicks, shipping cart history, whether or not the user is accessing your site on their mobile or desktops, and then if it was your mobile or the desktop version of your site—you grabbed all that, and made money off it selling stuff better.

You also grabbed all the lovely contextual and customer behaviour, so his demographics, her expressed interests, that family’s interest (or not) in your customer rewards/loyalty programme. Seasonal data—the list goes on.

The end of the third-party cookie

But then GDPR came along. And for the last few years, you’ve probably been most grateful to all the amazing data Google and Facebook and SEO has been selling to you… but we’re about to move into a universe without third-party cookies; Chrome and Chromium-based browsers will stop supporting them in 2022, Safari has had an anti-tracking for some time, and the latter is also of course pretty much at war with the rest of the social web right now with its very tough stance on privacy all-told. At the same time, Google has changed its algorithm yet again to enforce authority and stop robots gaming the system with valueless content.

Yet perversely enough, just as the main way you’ve been trying to achieve personalisation and appropriate ads for punters has been hobbled… people still want personalisation. If you ask your friends in the pub, Hey, what does everyone think about online ads, everybody hates ‘em, and cookies are the worst thing ever. But when your same pals want to buy new climbing shoes, do a search and in 10 minutes they get served with the most beautiful shoes on the market because the system knows which shoes they like, then everybody’s still okay with that convenience.

And like I said, people want personalisation more than ever before. 81% of consumers want brands to know when and when not to approach them, and 63% of consumers go so far as to say they now expect personalisation as standard. So we seem to have a big headache coming up for the sector: how to somehow still access data, but still mobilise it effectively for sales and marketing purposes?

AI: your new best friend

I am going to suggest the answer is something the sector has always relied on perfectly happily, mathematics and statistics. Use of so-say General Linear Models have been used in marketing since 1972, especially in the area of marketing mix modelling, as they allow for an additive approach (sales = baseline + TV spend + radio spend + other channels). This, in combination with clear model specifications, has always provided analysts with the opportunity to identify incremental contributions of marketing channels to sales.

But now, mathematics is getting a lot more useful. That’s because of the introduction of advanced software techniques to make the most out of vast heaps of data and numbers in the shape of Machine Learning (ML). And my contention is that ML can and will help marketers provide better experiences for customers and improve performance and conversions, balancing out to some extent what you’re losing from web data.

At its most simple, that’s because ML models the world. But not all of it; we’re not talking ‘The Matrix’ or ‘Devs’ here. Instead, it models a representation of the world, only extremely well, and in detail much, much greater than you or I or even Albert Einstein could manage.

The opportunities that are emerging out of this deep machine focus on narrow models of the world (= customer behaviour) range from hyper-detailed segmentation of customer types, which would allow you to offer truly dynamic pricing, really, really specific and accurate ad targeting, tailored messaging and performance measurement. ML is already being used in the sector to predict customer needs and future trends, to derive deeper understanding of the customer journey via attribution modeling, better budget allocation to improve ROI and allowing campaigns to dynamically flex to customers’ next actions and experiment with personalised experiences (‘Next Best Actions’ work).

Pushing democratic AI

However, as it stands too much of this excellent work is done in larger companies with big teams of data scientists. There is a definite gap between the data and ML specialist and the frontline Marketeer right now, even if the latter is au fait with data, search engine optimisation and even Python.

Companies like mine are trying to close that gap. We’re doing it via what we call ‘Explainable AI’: systems set up to be so user-friendly that there is as little ‘black box’ mystery and lack of explanation as we can manage. This way, teams like yours can access amazing tools that can crack very large amounts of data and work with very tough statistical models with ease—as our client Allergan has been doing.

If you’re not familiar, Allergan is a big US pharma company and is in fact the maker of Botox. And for two years it’s been working with the team at my company,, to move beyond its older models for estimating the impact of promotion campaigns and instead start to optimise the marketing mix of literally dozens of products across a variety of marketing channels.

How we’ve been doing that: by applying highly accurate machine learning models with high explainability that are leading to accurate marketing mix attribution models. It is specifically doing this via a Machine Learning technique called gradient boosting that uses a collection of prediction models that typically show a higher accuracy in out of sample predictions than the General Linear Models I referenced above. What its Marketing team have found: better results, more conversions, a better marketing budget mix—quicker and cheaper than before.

This is what your team can and should be achieving too. AI is becoming more mature, explainable AI is starting to really happen—and by turning to ML now, you could head off all the wasted months ahead as you try to work out how to live without cookies and ad tracking.

At the very least, it seems you’d have nothing to lose by trying.


More posts from ->

Ecommerce Age

How do we solve the issues with lockers?

We’ve all seen the banks of parcel lockers that we’ve all seen outside supermarkets and train stations. They are clearly a sensible idea, and one that I’ve been more than willing to use, but I’m very rarely given the option to do so when ordering goods online. Part of the problem, according to Gary Winter, VP of global strategic initiatives for parcel lockers at Quadient, is that they are invariably linked to a single delivery firm – such as Amazon or InPost – and this limits traction.


Related articles