Stop Wasting Money On Retargeting

andrewshebbeare"Data Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.

Today’s column is written by Andrew Shebbeare, founding partner and global chief strategist at Essence.

I’ve warned before that digital marketers spend too much time chasing conversions that were likely to happen on their own, at the expense of generating truly new business.

Now let’s put a spotlight on the most extreme example: retargeting. It is one of our industry's most overvalued and under-optimized tactics. I’ve measured anywhere from  20% to 75% of retargeting spend to be going to waste.

At its best, first-party segmentation, of which retargeting is one branch, can transform digital display, turning advertising into conversation. This is the beginning of a new level of personal relevance and contextualization in media. At its worst, however, the same technology creates a zero-sum game where audiences are hammered with repetitive ads desperately trying to intercept a purchase and earn some credit for something that was going to happen anyway.

Don’t fall into this trap. Here are seven ways to ensure your retargeting is delivering the good stuff:

1. Understand What’s Really Incremental

I recommend using marketing experiments to verify that retargeting is genuinely adding conversions that wouldn’t have happened otherwise, or is at least accelerating conversions that would have taken longer to materialize. Some insight may come from a sophisticated attribution model, but the best read will come from a dedicated experiment.

First create an unbiased holdout group that is deliberately not exposed to your retargeting activity, ideally by randomly selecting a cross section of cookies and excluding them from your retargeting pool. You could use a DMP to split your audience in two, hack together JavaScript to prevent retargeting recruitment for a slice of your audience or use third-party adserver audience segmentation to show some of your retargeting audience a placebo ad. Bear in mind that this last option is more wasteful and may be unpopular with any retargeting partners you reward for performance.

Make sure your experiment is persistent and delivers sufficient samples to get the read you want with a statistical power and significance you are prepared to accept. It should also be media addressable and deliver granular reads that can be sliced and diced for different types and combinations of exposure.

Once this is done, you can measure the conversion rate of the control and exposed populations and attribute the difference to your retargeting. Compare this aggregate result to the total conversions you currently credit to retargeting for an interesting starting point and overall feel for the true value of your retargeting campaigns.

2. Timebox And Know When To Quit

To get from interesting to actionable, cut the results from the analysis above by recency. The value of retargeting will vary depending on the amount of time elapsed between the action used to retarget and time the retargeting ad is served.

For most online retailers, conversion likelihood decreases over time after an initial indication of intent. The more recently someone came to your store, the more likely they will still want what you sell. But the most recent aren’t always the best candidates for retargeting; you should optimize retargeting based on incremental conversion propensity, not absolute purchase likelihood.

Some brands are wise to hold cookies out for a few hours before setting their ads on them. For others, the time window is shorter and you need to get right in there or miss the boat. For any product, there comes a point where it is too late. When conversion likelihoods taper off to the same level for both groups, it’s time to stop.

3. Cut It The Other Way

Recency is just the beginning. Your incrementality experiment will tell you which retargeting events and audiences are worth pursuing and which aren’t. Lapsed customers, periodic renewers, browsers, toe-dippers, window-lickers, brand loyalists and comparison shoppers might all have different optimal treatments. Slice and dice the results to find the important axes and plan campaigns and investment accordingly. Since conversion rates among these populations are high, you should get more granular reads than you’d expect from incrementality experiments you might run for prospecting efforts.

4. Be Interesting

Since retargeting looks so good on the surface, the creative for these campaigns often gets neglected on the assumption that it is doing a great job already. In reality, this should be one of the areas you work hardest to deliver messaging that tips the balance. These audiences know who you are; it’s time to tell them something they don’t know. Talk about a product feature they didn’t read about last time they visited your site, give them a special offer, tell them a joke, whatever. Be interesting. Content is more important than concept here. That cool brand message has played its part already; you are moving into CRM mode.

5. Choose One Platform, Carefully

You don’t want to bombard your customers from all sides or compete with yourself in RTB auctions. So pick one platform for retargeting for one set of frequency caps and one set of auction bids.

Make sure the platform has huge reach and capping that is appropriate and under your control. It should access all important inventory sources, support the right brand protection tools and you should trust it/them with your data. Make sure you have the right level of creative control, whether natively or through third party adserving.

Don’t unwittingly retarget through other media buys. You probably have some third-party tags in your site, so be aware that unscrupulous ad networks could exploit these to retarget without your knowledge, perhaps through black box “lookalike” media. Make sure you know exactly where your data is going. Lock in contractual protection for your most precious asset.

6. Beware “Pay For Performance”

I am skeptical about performance-based retargeting. Paying a third party on a CPA basis is tantamount to giving them your customer data and seeing what they can squeeze out of it. The margin a retargeter makes using your data is margin you could have held onto.

There’s also the impossibility of creating a model that aligns incentives. Attribution models are too easy to game. We know that brand exposure can play a big part in the effectiveness of retargeting, so a click-based deal misses a chunk of the point. Yet include any measure of post-impression credit and it is in the retargeter’s interest to cookie-bomb the world. If they buy below the fold and set the frequency cap at one per user, they’ve just taxed every sale you ever make online.

If you must work on a CPA basis, define strict rules around all the areas mentioned above. Don’t assume that a CPA buy will look after itself. If anything, it needs more supervision.  Test periodically for incrementality to make sure the extra sales are real.

7. Do It Yourself

It’s much easier to get control of all of the above if you take matters into your own hands, or those of a trusted agency partner, by directly leveraging a DSP or self-serve ad network.  This will make it much easier to deliver on all these safeguards and really understand what’s under the hood. You might need to layer in a good dynamic adserver that supports product feeds, decision trees, etc. If your audience segmentation needs are more advanced, add a good DMP that delivers flexible rules-based segmentation.

Specialist retargeting companies can add value for clients with really complex needs; they still have some of the most advanced and flexible audience management and dynamic creative capabilities, but getting the most value out of these arrangements requires real effort.

I usually steer clear of retargeting through standard ad network offerings because the value-add just isn’t there. I’ve seen many instances of retargeting acting to cross-subsidize underperforming broad network buys. This just muddies the water.

Ironically retargeting is often under-optimized because it looks so damn good on reports. It’s easy to max out retargeting, and for most advertisers it represents a small enough chunk of advertising spend that it gets lumped into “no-brainer strategies.” Don’t assume retargeting will look after itself, and don’t neglect it just because it looks good.

Follow Essence (@essencedigital) and AdExchanger (@adexchanger) on Twitter.


  1. Loving your stuff as always Andrew.

    We're running a number of these 'incrementality experiments' (I'm going to adopt your phrase from now on if that's alright). They aren't easy though are they? Even with a Cambridge Mathematician doing the set up and number crunching we find we're still learning how to set up the basic experiments themselves. The environment in which the experiments are running is complex and we often find variables for which we haven't originally controlled. Then cutting for all the variables like recency requires a lot of data which is not always available unless the advertiser is of sufficient size.

    We run placebo ads on our experiments using the AppNexus platform. These are 'wasteful' in terms of taking media budget but this cost compared to wastage of overall media budget or using survey methods seems small. I'd love to talk more about the other methods you mention; if you don't use placebo ads how can you understand the behaviour of similar, unexposed users?

  2. Yes, yes, and yes. Marketers are being ripped off, particularly by companies offering CPA and CPC deals for site retargeting. Why anyone would pay over $5.00 CPM for site retargeting in this day and age is absolutely beyond me. Had a great case last week at a conference where a marketer had given permission to include FBX on their plans for site retargeting - we estimated for them that about 40% of their budget had moved (to right rail, not newsfeed), yet the vendor had INCREASED their CPM, not decreased it.

  3. Hi Guys,

    First off, thanks!

    @John - yes, it's definitely not trivial, and it definitely costs money. The cost is a lot more bearable for high-incidence outcomes and for large campaigns - we find that 1-2% holdout is generally sufficient for large, national US initiatives with a reasonably high conversion rate.. but you have to factor in baseline, minimum detectable lift, confidence, significance, power, reach, overlap, contamination.. make sure you look after your buddy from Cambridge 🙂

    The cost-saving option I mention really only works in the confines of retargeting / 1st party segmentation where you can split the entire audience into control/exposed upfront and thereby hold out the control audience from the campaign. When you're prospecting this isn't possible; the only practical way to avoid bias is to hold out at the point of ad delivery. One thing we do find helps is where you might be able to work with 2 brands that can act as mutual controls - some of our clients are starting to work together in this way, and it enables much higher control proportions, which greatly increases the granularity of reads.

    For the most part, we've managed to avoid experiment contamination (or are able to correct for it) but there are always a few trouble spots!

    @Dax - glad to hear I'm not the only one with horror stories! This stuff gives our industry a bad name.



    • Thanks Andrew. We'll keep at it! Now if someone would just devise an algorithm which could build in the incremenatlity experiment and optimise to uplift rather than last cookie correlation we'd solve a lot of the industry's issues in one go. Perhaps a DSP could serve 1-2% charity ads on all campaigns and provide this feature to all its clients while writing off the charitable donation against tax.

  4. Stefan Vos

    The issues with retargeting stem from a fundamental problem with properly tracking value for online advertising campaigns. Any sort of last touch or fractional attribution will make retargeting look amazing simply because it is farthest down the funnel and those users tend to receive a higher frequency of exposures compared to prospecting.

    Unfortunately taking all retargeting in-house and A/B testing do not address the issue of value tracking and attribution. Less advanced advertisers look at attributed sales and don't question incrementality, or even if they do they say, "Well yeah, but we still need our eCPA to back out to xxx goal based on the attribution model." This means that any prospecting line item will show "poor" performance because they will be attributing a large number of sales to their in-house retargeting regardless of whether or not that user was first brought to the site through a prospecting campaign.

    Most agencies have no incentive to solve this issue for their clients, either, because they are not paid on performance or on true added value. An advertiser would probably need either a well-staffed team of statisticians & data analysts to tease out what is driving value, or a truly impartial ad platform that is capable of doing path-to-conversion analysis and automated A/B incrementality tests (or use regression for virtual A/B testing) on every part of the online media buy to develop a better attribution model. Once advertisers are properly tracking where value comes from, it becomes much easier to intelligently allocate online ad spending.

  5. All good points Andrew and some other great analysis in the comments here from everyone.

    One thing that seems to be missed right now (industry wide) is the role of behavioural economics / behavioural game theory - understanding consumer's intent to buy a certain brand, category, vertical will enable us to better understand the role of media in actually influencing and causing behavioural change. Behavioural economics provides a foundation for us to quantitatively evaluate human economic decisions.

    You can hire as many stats guys as you want, use as many laws (Bayseian, Rubin) as you want, but we need to start focusing on user intent modelling alongside standard ways of determining causality. Lets be honest, we solve that, then we solve the over-dependence on retargeting.

    It's something we're cooking up.

  6. The most interesting man in retargeting

    You can't get inside the mind of a consumer. Which means you can't take an incrementality test and assume that everyone who was exposed, bought because of that ad.

    Sure, things can be made better. CRM data and segmentation are a great start. But incrementality testing in the manner you describe is very problematic. You need more than a large enough sample and you need a lot of budget and relaxed cpa targets to make any measurable impact. And lets say you prove that retargeting was X% more incremental in one month...then what? You can't consistently test without losing the opportunity to sell, so what does one do with the data from one month to another? Does data from June apply in November?

    You're not going to be able to consistently say that retargeting is XX% incremental with any confidence, so why bother? The answer is attribution modeling and distribution of revenues that come into the cash register. Make assumptions, sort it out, live with the decision and look at the big picture unfold. I've been in the business for long enough to see retailer after retailer try to prove things with an incrementality test and come out on the other end scratching their heads. There's just too much that can go wrong.

    I think you've forgotten that Advertising is often about being there, instead of your competition. (BTW, have you tested the increment of branded terms? Didn't think so.) Fact is, its a war and advertising is your front line. Take it away because you can't prove that its "incremental" enough and you weaken your front line.

  7. Great article. Regarding Point #6 in the article, what is the recommendation? Strive to pay retargeting companies on a CPA, CPC, or CPM?

  8. I've been learning about retargeting recently and was about try it as a way to generate more conversions but after reading this article I'll have to re think things a bit. Thanks for the article.

  9. Hi Andrew, This is a good subject to discuss and an excellent article. A lot of website owners dive into retargeting based on the simple concept that if you follow these website visitors around they will eventually come back to you website and buy your product or service. When you weight the cost factor and the overall performance of the retargeting, it is not the greatest avenue to take. Many people believe it is a sure fire way to get more effective advertising but when you actually try it out for a while you will see it is not. I believe it is a fad that will eventually start to burn out once people start trying it.


Add a comment

XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>