Source: cmo.com.au
For a provider of loyalty programs across some of the biggest brands in the world, it’s pretty important Aimia gets personalisation and customer engagement right.
Which is why the leader in loyalty marketing, services, technology, and data platforms recently turned to artificial intelligence (AI), to better utilise its vast amount of data.
Director of insights and analytics at Aimia, Scott Pete, told CMO one of the challenges it faces is it’s dealing with a lot of different types of industries including CPG, retail, travel and hospitality. The customer base, the way they interact, and purchasing behaviours vary tremendously, he said.
“And the data sets aren’t the same. They vary from hundreds of thousands of customers to smaller programs, tens of millions in the larger ones. So, what we needed to do is to improve our ability to understand the engagement levels in those audiences. But with the volumes of data that we’re dealing with, it can get a little daunting if you use traditional query tools and that type of thing,” Pete explained.
“We got into machine learning [ML] over the last six or so years, and we’ve been moving more and more into the ML side. And then we came across Driverless AI about a year ago and decided to see if that might help us improve our time to market and accuracy of the models, which is which is how we started.”
As Aimia was already using H2O elsewhere in the business, it decided to run a trial of Driverless AI, which uses automation to accomplish key machine learning tasks quickly by offering automatic feature engineering, model validation, model tuning, model selection and deployment and machine learning interpretation.
“Our development environment is Python, and the H2O open source project product was much, much faster. It allowed us to fit our models in about 25-30 per cent of the time it would take us using other methods,” Pete said.
“When we tried Driverless, we were actually initially a little sceptical with the result because it seemed maybe a little too good, too easy and too fast. But we found Driverless was actually even more capable than the open source we were previously using. And we did evaluate other products before we made our purchase, but H2O came out on top.”
Since the trials, Aimia has developed multiple data models, including predictive churn, smart journeys, and fraud detection, with plans for more into the future.
“We’ve done a few predictive churn optimisations and either starting from scratch with a brand new model from H2O or taking existing models, and basically looking for ways to optimise miles and so we have had really good result with that,” he said. “The other one that really got our attention was fraud detection. And in some cases, some of the clients we deal with do have some incidents of fraud, and it’s difficult to detect because it’s really a very small percentage of the audience. And we found AI was a perfect tool for looking for a tiny signal in the data.
“The other application we are really excited about smart journey diagnostics. This allows us to look at the engagement levels in the customer base, and better understand and classify where each member is in their engagement, and also to predict where they’re likely to go next. We can actually quantify for our clients where the real opportunity is and where the risks are. We had really good results so far.”
In terms of measurement, Aimia is looking to accuracy to assess the results it is getting from AI, as well as speed and costs savings for clients.
“Accuracy is one of the top two, and our time to market. Time to insights is really an even better way to say it, because one of the one of the things marketers struggle with is trying to get insights out of their data, so that they can better prioritise marketing initiatives and marketing strategies,” Pete said. “And that’s something that we’re finding a lot of success with.
“Ultimately, once we get those insights, the accuracy in being able to predict a certain percentage of the audience likely to churn, showing a retention problem for example, is invaluable.”
As an example, one case study Aimia did recently broke the audience down into 20 different groups. The team was able to score all the members and then organise them into 20 different groups and identify the top six to eight groups where it could get the low hanging fruit and most impact from retention efforts.
“This saved the client about 700 per cent in campaign costs, because they weren’t spending money on offers in campaign communications unnecessarily,” Pete said.
Now the AI is working well across certain models, Aimia hopes to take it further over the next 12 months with a model it calls ‘next best experience’.
“Some of the research we’re doing right now we’re actually working on a model called ‘next best experience’. Marketers are familiar with next best action or recommendation engines, this model is going to take elements of those types of models and it’s going to take it to another level,” Pete said.
“So if you if you think about our ability to better understand the audience and what engagement level they’re in, then do some behavioural segmentation, and some value segmentation so we can understand the potential value of those customers. Then ,we are better able to target our one-to-one communications and really achieve real personalisation.
“That is something we’re really excited about. Because what will happen as we achieve this is we will deliver a better user experience for the end customer. And we believe that will help deepen the relationship and improve the loyalty as well.
“I think the expectation of what we can do with technology is rapidly changing, as companies start to see the time is right to implement AI.”