Read our new manifesto

Now available on-demand

Get inspired. Find solutions. Harness the power of digital marketing.

Featuring Speakers from

Agencies 4 Growth Festival Logo
Agencies 4 Growth Festival Logo
Agencies 4 Growth Festival Logo

When data bites back

Hardly a day goes by without another piece of evidence that most businesses are woefully ill-equipped to manage the abundance of data that they now find resting uneasily in their care. A recent report by Dentsu Aegis Network found that, among CMOs and senior marketers, a data breach or misuse is the number one strategic risk they face today.

The fallout of having your customer data hacked is obvious and have been faced by everyone from Facebook to Equifax to TJ Maxx.

But the next frontier of data risk is ‘well-intentioned misuse.’

A striking example of this is the now-legendary story of Target’s data mining operation being able to determine which of its female shoppers are pregnant based on their purchase patterns. In one case the resultant direct mail pieces were intercepted by the father of a teenage girl. Affronted by Target’s insult to his daughter’s virtue, he phoned Target’s customer care team to rant about the situation – only to call back days later to apologize as she was indeed pregnant.

A similar case played out recently with Netflix. Famed for its expertise in personalization, Netflix has been embroiled in accusations of racism – a painful episode for a business whose progressive credentials are well-known.

The accusations stem from a number of African-American viewers angered because their recommended titles feature black actors on the carousal artwork, but those actors have a minor role at best. At first glance, this seems like the same kind of cynical bait-and-switch that movie houses have been known to make in the past – such as when posters for ’12 Years A Slave’ seemed to present Brad Pitt in the starring role.

But the truth is a lot dumber than that. An algorithm, totting up the cast members of all the films that someone has watched, goes through the list of films that the viewer hasn’t yet watched to see where those actors crop up again. If a viewer’s favorite actors score a hit in an unwatched film, then, presto, a title card with that title and actor is compiled and served into the recommendations carousel.

The problem is not that the algorithm is a bad idea – countless of these algorithms run harmlessly in almost every digital experience we undergo. It’s a matter of poor execution and a lack of application of three fundamentals of data usage.


All data points should be viewed to have different weightings of importance. When working in e-commerce fashion, for example, “this item is out of stock for 4 weeks” is a data point that should be weighted with much higher salience than, “the user probably likes the color of this garment.” One is a purchase-killer and the other has a slim chance of driving preference, at best. These variables should be used to make case-by-case decisions programmatically on which recommendations to make.

Netflix should place a very low salience rating on “actors who are sixth or lower on the bill.” It’s not just their African American audience who are having a poor experience of bait-and-switch promotions, it’s likely everyone.


Many data outcomes that we, as strategists, might see as harmless are not perceived as such by their recipients. Imagine being repeatedly targeted for weight loss or debt relief products – a brand that is trying to be helpful and relevant might quickly irritate customers, particularly if the data read is inaccurate. We should also bear in mind that whatever personalization we deliver is likely to be seen by an entire household, not just by the intended recipient, as Target discovered.

As seen in Netflix’s case, you build an array of scenarios and edge cases to forecast risk. Some meta-data (even uncoded meta-data such as an actor’s ethnicity) can only reveal its risk by working through some real use cases.


All hypotheses that you deliver to your audience about what their preferences might be, must be done with a degree of humility – we can never programmatically predict human nature, as any man still being retargeted with lingerie ads months after his wife’s birthday will attest. New personalization campaigns should be micro-managed with daily reporting – not just from site metrics, but also from the broader miasma of social media chatter, call center inbounds, and the company email inbox.

In Netflix’s case, this problem should have been detected earlier to prevent it from ascending to the pages of the broadsheet newspapers.

My advice is, as an obvious first priority, make sure your data is secure to avoid being tomorrow’s catastrophe headline. But, alongside that, apply some of these nuances to ensure that your data, no matter how lovingly tended, doesn’t bite back.

Tim Dunn is vice president, head of strategy, at Isobar US

Join us, it's free.

Become a member to get access to:

  • Exclusive Content
  • Daily and specialised newsletters
  • Research and analysis