Does your remarketing really drive incremental revenue?

There seems to be a common belief among digital marketeers that remarketing is a no-brainer because it delivers conversions at a low cost. Being our critical selves we question to what extent remarketing really delivers incremental revenue.

Client case study

We regularly get confronted with the remarketing ROI question when we perform digital marketing audits for our customers. They typically show us reports that their agencies send to them or which they get from Google Analytics, such as the one below where Display refers entirely to remarketing activity via the Criteo advertising platform.

In this example we notice first of all that remarketing seems to generate 320K revenue. The cost for generating the traffic that generated this revenue was just over 25K. I can hear you think “that’s a nice return on investment”. Did you also notice that remarketing actually generated the lowest conversion rate of all channels?

Going beyond last-click attribution

Last-click attribution is besides the truth as you may know. It actually overemphasises the value of all digital marketing channels at the expense of your direct traffic. When we take a different perspective on this by confronting two attribution models in Google Analytics with each other we come to an interesting find:

  • Last non-direct click, the default attribution model in Google Analytics, reports remarketing revenue with a value of 273K.*
  • Last interaction, which attributes conversions to the last interaction even when that is direct, reports remarketing revenue with a value of 18,5K.

Remarketing now only generated 18,5K revenue. The advertising spend on this channel of 25K actually exceeds its revenue!

Conversion path analysis reveals insights

I hear you think, “let’s kill it”. So did we. But our inquisitive nature forced us not to jump to conclusions. Our analysis continued by taking an in-depth look at all the different paths in which remarketing featured using the Top conversion paths report in GA. There were 2.084 paths that led to conversion in which remarketing featured. These 2.084 paths can be broken down as follows:

  • Paths where Criteo remarketing was the first step in the path: 349 paths **
  • Paths where Criteo remarketing was the last step in the path: 231 paths
  • Paths where Criteo remarketing featured in the middle of the path: 1.544 paths

Clearly the vast majority of paths that generated remarketing revenue feature remarketing traffic in the middle of the path. The screenshots below are samples taken from these 1.544 paths.

What stands out is that many of these revenue generating paths start with direct traffic or email interactions. Which begs the following question: Visitors who come to my site directly and even repeatedly do that, should I really be remarketing to them? These actually are my loyal customers who easily find their way to my site. They may even find these remarketing ads disturbing.

The same for my email visitors. The only reason they come via email is because I have their email address as a result of a previous purchase or visit. They also know us so should we really be remarketing to them?

The reason this even happens is because Criteo tracking tags do not make any distinction between your visitors. Criteo tracking tags are implemented on your homepage, your product overview pages, your product pages, your basket pages and your conversion page. Because this Criteo tracking is so omni-present on your site it is almost impossible not to be retargeted!

Action based on data-driven analysis

Let’s cut to the chase. Based on the above data-driven findings the assumption grew that an awful amount of remarketing effort was directed at very brand loyal customers who did not quite need this. They would come to the site anyway and buy.

And with the fact that in a last interaction model the cost for this activity (25K) was higher than its revenue (18,5K) it was decided to discontinue Criteo remarketing altogether. Did this actually lead to a 320K revenue drop thereafter? By no means!

The remarketing activity was discontinued in week 11 after which remarketing revenue is gradually reduced to zero. Total revenue remains completely stable.

Moral of the story

Because of the machine-gun remarketing approach that Criteo uses a lot of remarketing activity is directed at loyal customers who regularly revisit the site. Turning the remarketing activity off does not have an impact on revenue but saved 25K in marketing costs.

Are you using Criteo and do you want to find out how much you can save? Go to, enter your Google Analytics account number and your saving will be calculated.


* The two Google Analytics reports respectively quote 320K and 273K in remarketing revenue, albeit using the same attribution model. The difference is the attribution window. In the default model where 320K revenue is attributed to remarketing the standard 6 months window of GA is applied, where the second report was set to the maximum 90 days attribution window GA offers.

** Remarketing can in theory never be the first interaction in a path. After all it is RE-marketing, meaning there had to be a previous interaction via another channel. They exist in GA for a number of reasons of which device fragmentation and the limitation of the lookback window are the most important ones.

Performance Marketing Meets Alley Oops and Three Pointers

Eye for performance, also outside of our professional digital marketing lives.

As of this season Engage Digital is the main sponsor of Jets Basket. This basketball team, based just outside of Brussels, Belgium, achieved a promotion to a higher division by winning ALL of their competition games last season. How about that for performance?

Remco van der Beek, founder of Engage Digital: “My kids play basket in this club and we love watching the other teams. The men’s team had such a memorable run last year. Their clean sheet inspired us to give something back to the club.”

“We hope our support will relieve some of the struggle non-professional clubs are facing to make ends meet. There are so many passionate volunteers involved in keeping the club going and making sure all the players receive training and can enjoy competitions. This was a no brainer.”

Sven Ceuppens, President of Jets Basket: “We currently have 13 teams playing in competition. Being a relatively small club we completely rely on our members, volunteers and sponsors for the day-to-day running of the club. Every bit of help, big and small, is very welcome. We are therefore extremely pleased with Engage Digital as the new main sponsor of our first men’s team and we’re looking forward to a lasting relationship.”

How much do your display advertising campaigns really contribute to conversion?

Establishing the real value of display through placebo-based attribution

The progress in ad tech in recent years goes hand in hand with increasing concerns from advertisers around viewability, fraud, measurement, brand reputation and industry malpractices. In October 2017 P&G, the world’s biggest advertiser, cut back its digital advertising budget by 200 million dollar1. In February Unilever, the world’s second largest advertiser, announced it may also be cutting digital ad spend2. The writing has been on the wall for a while for display advertising:

  • Average display advertising click-through rate in Europe is at 0,3% or lower 3
  • Display advertising viewability in most European countries is around 50%-60% 4
  • 20% of internet users in Western Europe use an ad blocker 5

With the above statistics in mind, how can you be sure as an advertiser that your investment in display advertising actually makes a difference to your bottom line?

Awareness or conversion?

Many advertisers have no sound model in place to show the real correlation between display advertising and conversions. Conversion revenue is attributed on a last-click basis and because of poor CTR’s the revenue usually is far below the ad spend (read on for post-view conversion). For this reason marketers often turn to awareness as an objective to justify digital ad spend. But how do you measure that your advertising campaign has generated awareness?

The most accurate way to measure awareness is through upper-funnel metrics such as brand awareness, ad recall, brand favorability etc. which are measured through post campaign testing that requires interviewing human beings. Since this is very expensive most advertising campaigns, even in large companies, are not post tested. Instead many marketers and their media agencies justify ad spend through fuzzy metrics such as impressions and post-view metrics.

Impressions are not views

An impression means an ad was served. It doesn’t mean the ad was actually viewed. The IAB standard for ad viewability stipulates that a valid ad view is generated when 50% of the pixels of the ad were in a viewable position for at least one second. Is it really possible to understand what an ad is about after seeing half of it for 1 second ? I beg to differ. Even so, with this highly lax definition viewability rates are around 55% for many advertisers. Meaning that as an advertiser you may reasonably assume that almost half your budget goes down the drain before you even started.

“Half the money I spend on advertising is wasted; the trouble is, I don’t know which half.” – John Wanamaker (1838-1922)

Building on that assumption: if the other half of your ads are served in a viewable position, this still does not mean they were actually seen. Heatmap studies have confirmed that website users tend to ignore display ads.

What about post-view metrics?

Post-view conversion refers to website users who did not click on display ads but who were exposed to them, then visited the advertiser’s website through another channel than display ads, and then converted (buy something, fill out a form etc.). So the outcome is attributed to the ad impression on the assumption that viewing the ad made a difference. And this is where post-view attribution breaks down. Because, once again, ad impressions don’t mean ads were viewed. In fact, about 50% of those ads weren’t viewed.

How then can we measure the true impact of display ad campaigns in terms of conversion attribution? Enter placebo testing.

“Viewthroughs are the backstop of every Marketer when they know there is no provable value.” – Avinash Kaushik

Placebo testing

The concept of placebo testing is rather simple: you ramp up two identical display advertising campaigns with the same targeting and landing page. The only is difference is the ad creative. One ad communicates your brand message, the other ad, also called the placebo ad, communicates something that is totally alien to your brand. If you are in fashion this could be dog food, travel insurance, sanitary towels or whatever really. You split run the campaign and then you compare the post-view conversion stats. If your dog food ad generates a significant amount of fashion purchases (here’s a little secret: they always do) then you now know that your post-view conversions are to be taken with a margin of error. How big a margin of error? The calculation as seen in the example below is rather simple:

placebo ad testing

Creative A: brand message => 100 post-view conversions

Creative B: placebo ad => 50 post-view conversions

Margin of error = 50/100 x 100% = 50% => only 50 of the 100 post-viewed conversions that were attributed to creative A are genuine.

Nice theory but…

There’s no but. This is the way to approach it. The outcome for one of our clients was that post-view conversions that for years had been attributed to their display advertising campaigns had to be discounted by almost 40%. There was a deep-rooted belief with this client that display ads were indirectly responsible for the bulk of their online revenue, based on years of media agencies reporting shiny post-view metrics.

Their marketing budgets were spent accordingly with a heavy emphasis on display advertising. Yet when all display advertising was paused over a brand safety concern their website traffic wasn’t impacted at all. People barely click on display ads remember. More importantly their online revenue was not impacted either in the many months that followed. Food for thought for next year’s marketing budget allocation.

What about programmatic?

Programmatic advertising is still display advertising just done in a different way. Placebo testing on programmatic campaigns is not possible though. Since you buy impressions one at a time it is impossible to do split run testing. But you can identify on which sites your programmatically bought ads frequently appear and then do a media buy on one of those top sites to run your placebo test. Chances are you’ll be very surprised by the outcome. This doesn’t mimic programmatic buying, but it’s pretty indicative.

What about remarketing?

Time and again I hear clients stating how well their remarketing campaigns work. Of course do people who already visited your website convert at a higher rate. But do you really believe that showing that one additional display ad pushed people over the edge towards that conversion? I encourage you to challenge this common belief using objective data from placebo testing.

Most of the remarketing I come across is quite poor. If I have to single out one thing that particularly disturbs me then it is remarketing ads that are shown to me after I have already converted. This regularly happens to me when I buy travel or make purchases on e-commerce websites. It skews remarketing effectiveness data and is a sheer waste of your marketing money.

So think critically about your display advertising. Don’t let yourself be led by agencies who earn their money from media buying. Ask the difficult questions and use objective neutral data to answer them. If you’re curious about placebo testing or in need of help with your attribution questions, don’t hesitate to give us a call.





Why you have far less unique visitors than your web analytics tool tells you

Web analytics tools heavily overstate the number of unique visitors your website has. The degree of error ranges anywhere between a few dozen percent to a few hundred percent. Here we discuss the variety of causes for this erroneous unique visitor count.

Javascript disabled

The smallest issue for distorting unique visitor count is visitors who have Javascript disabled. For many years the percentage of users who have Javascript disabled has evolved around 2%. Often these are visitors who use Firefox NoScript add-on. When users who have Javascript disabled visit your site none of their activity will be measured. This can be compensated somewhat by adding a tag to your normal analytics tag.

In the Google Analytics example below this will register a hit via the Measurement Protocol even when Javascript is disabled. This will enable you to register a pageview but most of the other information that a tag would register won’t be captured.



The use of ad-blockers, particularly by younger users, significantly increased in recent years. In many European countries ad-block usage is above 20%. Many of these ad-blockers have the ability to block web analytics trackers. Some ad-blockers block analytics tracking by default, others make it very easy to enable it.


Image source:


The margin of error this creates varies according to your user base. One experiment established that Google Analytics script was blocked in 11% of cases.

In-private browsing

Nearly one out of five internet users use in-private browsing. With regards to analytics tracking this means only temporary session cookies and no permanent cookies to identify unique visitors will be placed. When visitors come back to your site they will be considered a new visitor thus inflating your unique visitor count. However, their session information and hits will be registered correctly.

The above applies to nearly all browsers, except Firefox. In-private browsing in Firefox means the NoScript add-on will block any form of Javascript tracking. As such no measurement will be made at all.

Cookie deletion

It is hard to find statistics on cookie deletion behaviour. Yet when I ask in the many analytics courses that I teach whether participants delete cookies I find that without exception several of them do at varying frequencies (anywhere between 6 weeks to several months). If I were to stick a number on it I’d probably pick 20%. Visitors who delete cookies obviously heavily distort your unique visitor count since they will be counted as a new visitor every time they come back to your site after having deleted their cookies.

Multi-device & multi-browser usagemulti-device-usage

Users who use more than one device to visit your site will have a tracking cookie on each
device and will therefore be counted as multiple unique visitors. The question is, how
many of your visitors do use multiple devices. The only way to prove this is by making visitors login to your site every time they visit. You will then be able to record a unique user id instead of a cookie id.

In the example on the right taken from a site where users are always logged in we see that about 30% uses more than one device to visit the site. If no unique user id would have been captured this would mean the unique visitor count would be inflated by 30%. On your site the same thing happens although maybe to a lesser degree (the example was taken from a site to which visitors return very frequently).

When users use different browsers on the same device to visit your site the same principle applies. They will have a cookie in each browser will leads to inflation of the unique visitor count.

In conclusionuser-id-vs-cookie-id-3

There are different reasons why the unique visitor count of your analytics tools is overstated. Some issues cause that unique visitors are underreported (javascript disabled, ad-blocking), other issues will inflate the unique visitor count on your site (in-private
browsing, cookie deletion, multi-device usage, multi-browser usage).

Clearly inflation is far more likely to happen than under-reporting. Yet, it is virtually impossible to quote a generic inflation factor.

In the example on the right from a site where users a always logged in we confronted the number of unique user id’s with the number of unique cookies in Google Analytics. The result: 93.681 unique user id’s matched to 312.577 users or unique cookies.

The real number of unique visitors was overstated by more than a factor 3 ! With some visitors equaling as many as dozens to hundreds of unique cookies !!

So…be warned…unique visitor does not equal unique human being…far from !