Building Top of Funnel in a Cookie-Free World: A Manifesto

The marketing world is well aware that there is a crisis in marketing measurement today.

As I write today, Google Analytics has been deemed to be illegal under GDPR by two EU nations – France and Belgium – with the rest of the EU expected to follow.

This has been anticipated since the Court of Justice of the EU’s acceptance of Schrems II in 2020.

To simplify the issue at hand: Alphabet (Google’s parent company) cannot guarantee there will not be PII information accessed by US law enforcement in their Google Analytics products as the US obliges US companies to comply with data access requests – including encrypted data – and they are unable to prevent data being passed through US servers, which host the bulk of their data processing activity.

Google addressed this illegality finding by admitting that their legacy analytics product – Google ‘Universal’ is indeed illegal under GDPR, and committing to make GA4 compliant in the future. The date they set for that change has been and gone – with no data regionality added to the product as promised.

As it stands then – Google Analytics, in all its forms, is illegal for GDPR.

Currently, Google holds somewhere between 56% and 86% of the global marketshare of analytics. Given the long tail of the internet is very long indeed, we can safely assume that for the majority of meaningful websites in the world, Google Analytics is closer to 86% than 56%.

What does Google Analytics being illegal under GDPR mean for me?

In short, if you are collecting data from any EU or UK individual using Google Analytics in any of its forms – including GA4 – then you would be deemed to be in breach of GDPR today, and therefore liable for a fine of up to €20 million EUR or 4% of your turnover, whichever is higher.

I make the distinction that this applies to UK individuals as well as European individuals because there are in fact two GDPRs – the UK GDPR is separate but essentially identical to the EU GDPR. So despite the UK leaving the EU, GDPR applies across both territories.

As anyone who has operated a website knows, you cannot control where your traffic comes from in any reasonable way.

Sure, you could look at Geo-blocking solutions, but the complexity and inevitable workarounds that would invalidate the effort in a court of law make that a Sisyphean effort at best. And lets not forget that for the majority of sites, there is a commercial reason to allow traffic from EU and UK markets – they are large, valuable markets. So, excluding traffic is not an optimal solution.

What options do you have?

Compliant, effective marketing measurement

I will start by making a disclaimer – I’ve been looking at marketing measurement professionally as part of my role at QueryClick for over a decade, and eight years ago signed off on the development of alternative approaches to measurement based on what I found when analysing multiple millions of pounds of client revenue and marketing activity data.

Ultimately, QueryClick invested around £5 million in the development and launch last year of a standalone SAAS analytics solution – Corvidae – that replaces existing approaches to analytics measurement with an AI-based approach.

I would have been delighted to find an off-the-shelf solution that would have worked for our clients here at QueryClick. The reason we could not find an effective marketing solution was not because of legal issues – as at the time, GDPR was on the horizon and not mature enough to cause businesses’ enough concern about being taken to court for significant fines.

No, the reason we were unable to find an analytics solution that offered marketing measurement accuracy was more simple and prosaic: the technology used by all analytics platforms is fundamentally not fit for purpose.

There are a great number of analytics products out there, many with specialist use cases – mobile app measurement was briefly seen as a separate, specialist measurement requirement for example, until cloud architecture matured sufficiently to make live event streaming trivial.

They all, however use a pixel and an ID to join together multiple pixel hits into ‘sessions’ and ultimately ‘visits’. These sessions and visits purport to represent an individual behind their device as they move along a conversion path towards – hopefully – some kind of tracked conversion event – a transaction, or perhaps a dwell time on a particular page, or an app install, or app usage times – and so on and so on.

Ultimately all analytics exist to allow some form of measurement towards a particular event or sequence of events that are meaningful and valuable to the business operating the website.

The ID in the pixel and ID combination is typically a 1st party cookie on the wider web, or a device ID in an app environment. Sometimes cross device data sets are used – illegally under GDPR – to try to stitch together multiple IDs to a single individual.

Or, if you are Google, or a similar operator of large scale web services (looking at you Apple and Amazon!) you might use those logged in states to build a ‘Signal’ that you can use to extend the ID beyond just a single device or application.

In our analysis of the accuracy of these approaches, we consistently found – from 8 years ago and up to today – that these deterministic approaches (trying to generate an ID and match to other IDs) fail to generate an accurate map of an actual individuals real conversion path around 80% of the time.

Meaning that when you look at the Visit data in your web analytics – whether it’s in Google, Adobe, LiveRamp or some other solution – only 20% of the time is that a true and accurate picture of the full conversion path.

Understanding accuracy online & the attribution challenge

If you are an experienced marketer used to dealing with multiple millions of pounds of revenue online, and you have ever looked at attribution, then you will not have been surprised by the finding I’ve just outlined – and the evidence is there in the practical difficulty and outcomes delivered by any and all attempts to get simple attribution functioning in any meaningful way for your marketing optimisation.

Who hasn’t tried to base day to day decisions on ‘Data Driven’ attribution, or even a simple rules based approach like ‘Time Decay Weighted’ and found that when reallocating budget using those approaches overall performance got worse?

And if you run digital and traditional – or operate bricks and mortar – then you will be very familiar with the phenomenon that what is reported digitally just doesn’t integrate with those data points consistently, even when applying sophisticated data science to try to join statistically over that gap.

The reason is not that the offline data is in silo. The reason is that the digital data is broken.

Attribution is seen as a dirty word in large parts of the digital (and, increasingly, traditional) marketing world, as we’ve all heard about how it is coming to save us from wasting the apocryphal 50% of our marketing spend so we can remove the cannibalistic media spend we know we have, and allow us to buy ads at the top of the conversion funnel at lower costs – up to 87% cheaper on a CPA basis in our published case studies – due to the lower volume and price of auctions at that point due to the world and its dog all being stuck with Last Click models that prioritise the very last touchpoints before conversion.

The underlying reason for this frustration with attribution, and the enduring persistence of Last Click – ascribing 100% of the value of a conversion to the last touchpoint visited – is that deterministic measurement – the use of an ID, stored in a cookie – is not fit for purpose in the multi device, multi app world in which we live today. And especially when 3rd party cookies which are already heavily removed from the browser marketshare will completely disappear by the end of the year.

Eight years ago, it became possible to deploy AI models and feed very large volumes of data to those models without breaking the bank. At QueryClick we began working with AI and our customer data to see if we could improve on 20% accuracy.

Three years ago, and several global patents later, we had the core of what has become the Corvidae AI ready for client testing. Last year, we released a fully Azure deployed SAAS product that takes 10 minutes of technical setup, requires 30 days of data, and which provides above 95% accuracy even when tracking across multiple data silos or into offline activity as well as online.

Sure, today we still need to help walk customers through onboarding, and we also know that accurate attribution is disruptive to businesses – for example with Tesco we found that the sources of their digital revenue were 75% different from what was being reported by Google Analytics. That’s a tough message that requires education, and proof – and of course plenty of value for the business to more towards as the new truth about what is actually effective in their marketing activity.

So, we wrap our deep experience in attribution and taking action on changing marketing focus to top of funnel to help our clients and their agency partners get he most out of Corvidae and deliver at least 10X ROI on their Corvidae deployment in the first year.

Test and prove – and the value of accuracy

So how do we know we are 95% accurate compared to just 20% or so in old approaches to analytics that use cookies? Simple.

Corvidae’s AI is a predictive Neural Network, meaning we can test how well it is building individual conversion paths by testing its ability to predict if those paths will convert or not.

What the conversion actually is doesn’t matter. It might be a transaction, or simply dwell time, or a particular event fired by the pixel.

Whatever event is selected, we wait until we have fed sufficient data to the AI for its training process – typically 30 days of data – and then take ten thousand paths that we know converted – because they contain the conversion event – and then select ten thousand paths that are as similar as possible to those paths, but do not convert, and then we strip the conversions from the first set and ask the AI to sort the paths into converting or non-converting.

Three years ago, our best accuracy was 85% or so, today we don’t release Corvidae’s AI from its initial training phase of onboarding until we are hitting 95%+ accuracy in sorting paths correctly.

This accuracy means we have a very granular accurate map of conversion paths – both converting and unconverting – that is collected compliantly – only a 1st party pixel is used to collect data – and to which we can probabilistically join to a wide variety of data sources – including traditional ‘walled garden’ data sources such as Facebook, TikTok and Google or Microsoft Paid or Organic sources – but also offline data sources, CRM data and more.

For each joined dataset, we validate the accuracy of the AI to ensure the enriched conversion paths featuring ad impression events and the like that have been derived from the siloed data sets continue to retain their accuracy.

Taking action & delivering value

For every marketer, once you have genuine accuracy, there are profound opportunities to deliver value for your business.

Accurate analytics give you fair scoring of marketing events that come right at the very start of conversion journeys. Meaning we can understand prospecting and brand activity’s contribution to revenue accurately – and there is an interesting insight about the required lookback window to establish that contribution and what the AI can tell us about buying habits.

Most importantly we can spend at the top of the funnel and test and learn what ad creative is most effective for later conversion touchpoints.

We can remove cannibalised spend, as those events will not be attributed value by the AI.

And we can extend those ‘pre-conversion paths into our CRM systems to better understand the source of our best and highest lifetime value customers and meaningful cohorts to outreach to.

We can spend on digital TV slots at half the CPM cost of traditional TV in the same household and measure the placement’s influence over conversion and how we can adjust our media plan in future to improve conversion volumes.

In short: marketers can finally stop talking about technology and obscure measurement KPIs and instead focus down on thinking about their customers more deeply and innovating marketing messaging and being great marketers – not digital, not traditional – just marketers that deliver for the bottom line for their businesses.

It’s been a long time coming, but digital has finally matured – welcome to the future of marketing.