IAG: Marketers must end the blind faith in martech and adtech

Marketers with blind faith in adtech and martech are at risk of putting advertising in front of people who were going to buy anyway and waste time and money on useless personalisation activities.

This was the message from IAG’s director of media and technology, Dr Willem Paling, and one-to-one marketing director, Jason Ridge, who took to the stage at CeBIT 2018 to discuss where and when to use AI-enabled martech, and why it’s vital to set the right goals around it.

Ridge said all consumers expect personalised experiences, and marketers can throw data into AI platforms to ensure the content coming out of their tech engines is personalised. But IAG is starting to ask if it really needs the technology to achieve relevance.

“No one will say delivering customers’ personalisation isn’t a good thing. However, Amazon has hundreds of millions of products, and Facebook has millions of advertisers, and given the opportunity to put a product or content in front of customers, they are left with a dilemma of which product or content to put in front of which customer. So of course it makes sense to use AI or ML to do this,” Ridge said.

The problem is not how marketers are doing personalisation, it’s about when it makes sense to do it, Ridge said.

“For us, we’re only dealing with three products. As much as we’d like to think people are engaged with insurance, the reality is they’re not. So we had to ask, what is personalisation when you only have three products? When should we be using AI, ML and personalisation?”

IAG recommends a three-stage approach before implementing martech:

1. Spend more time defining the problem
2. Spend more time defining the details of the goal
3. End blind faith in technology, and learn a little tech.  

1.Spend more time defining the problem

“We can sell one product with a load of different content; we could capture loads of information on customers and design different websites and pages we could then dynamically serve to our customers,” Ridge explained. “But in the case of things like A/B testing, it’s very hard to get a clear idea of what’s influencing whether customers will buy or not.

“When you have a single product, should we be spending the time and effort designing additional content to sell the products in different way? Unless we can show a real statistical difference in doing that, we’re not going to.”

Paling added it’s very common in business to see hype around AI and ML. But way too often, it doesn’t add up.

 2.Spend time defining the goal of martech

To know in which camp you sit, it’s vital to spend time defining the goal, Paling said. “In general, marketing’s goal is to get more sales. But having a goal of sales using martech that deploys ML or AI doesn’t necessarily mean adding more sales.

“An important component is setting the goal to extract value of the AI system. As marketers, we’re in charge of AI systems often without understanding we’re setting the goal.

“For example, if a car AI system's goal is to minimise accidents, then the best thing for it to do is remain in the garage. So we need to think hard what those goals should be. The goal of marketing is not just sales, it’s to maximise sales generated by marketing, or incremental sales that wouldn’t have happened without marketing.”

Just because someone buys a product, doesn’t mean it happened because of an ad. Even if clicking on an ad leads to a purchase, it’s possible the consumer would have bought that product anyway, Paling said.

“A lot of marketing systems that automate marketing and bidding systems are naïve to the issue of causality,” he said. “If it’s properly designed, AI will find and target compliers – those who wouldn’t have bought without marketing. This is how AI should be deployed.”

However, this is not typically what happens, Paling claimed. What is actually happening is marketers work with vendors to select a martech platform with an AI component. They then work to identify a goal, and leave the platform to run, believing they are extracting value.

“In an ideal scenario, the marketer will identify the goal, they will provide the domain knowledge to identify the relevant input variables, and then work with data scientists to make sure it’s delivering a causal outcome,” Paling said.

“If we set our AI goal as just being sales, then we want to have our ads in front of as many as possible. Therefore, we are just getting our ads in front of people who were going to buy anyway. This is the industry standard. What we really want is to leave those people alone and put ads in front of those whose behaviour can be changed to cause them to buy, to get additional sales to the business. We want to get those compliers who buy subject to marketing.”

To do this, Paling said, it’s important to end the blind faith in assessing adtech and martech, and learn a little bit about technology. Marketers need to be learning more so they can be technologically and statistically literate in order to have a conversation with data scientists to add a solution that actually adds value.

3.End the blind faith in martech

“We all use DSPs [demand-side platforms] to optimise the purchase of digital display ads. Our job is to set the parameters and decide how much we’re willing to bid for an ad placement, and a sales conversion event is used as an outcome variable,” Paling explained.

“Then the system optimises by building a model assessing the likelihood of conversion, taking all features into account and adjusting according to the likelihood to convert. But it’s not adjusting the bids according to the likelihood to convert subject to advertising. There’s no distinction between compliers, always takers, never takers and defiers.

“It’s not necessarily adding in any incremental sales because it’s just putting ads in front of people most likely to buy anyway.”

Related: Exclusive: IAG, Columbus debate digital advertising measurement pitfalls and perils

Paling went on to say while this is not terrible, it is naïve to intent characteristics of a buyer when systems don’t account for the impact of adverting or if it’s causing any change in sales, or adding any incremental value.

“Causality is so important to setting a goal. Browsing intent can easily be correlated with purchases whether advertising is present or not. So with the wrong goal, the AI system can optimise with any ads. We could also apply this to personalised content,” he said.

IAG ran a recent test on clicks or ad impressions to see if there was any value in this approach to measurement. When users came to the website, half were targeted with advertising, and the other half didn’t receive any. The standard approach to attribution saw 1800 sales attributed to marketing.

Looking at a newer algorithmic approach to attribution, 1700 sales were attributed to marketing. The big difference, however, was in the group that received no advertising at all. There, IAG saw 26 more sales.

“We turned off that advertising and saved $500,000 a year, and threw out that method of attributing value. This shows how out the goal can be if we don’t think properly about setting the right one,” Paling said.

“In summary, spend time defining the problem, and if AI is the solution, it needs to be solving a problem, rather than the other way around. Spend time defining and understanding the detail of the goal.”

Follow CMO on Twitter: @CMOAustralia, take part in the CMO Australia conversation on LinkedIn: CMO Australia, or join us on Facebook: https://www.facebook.com/CMOAustralia, or check us out on Google+: google.com/+CmoAu