In recent times, Gemalto mostly made the news with Atos’s buying attempt and with Thales’s succesful one. However, the Gemenos-born company remains the SIM card world leader, in charge of this key element of the gigantic data collector that has become today’s mobiles.
It was therefore natural for Gemalto to test machine learning, which it applies to mobile marketing. In this regard, I talked to Frédéric Martinent, head of the company’s mobile marketing business line.
Phocea Tech: Before talking about machine learning, what is the Smart Message?
Frédéric Martinent: It is about using SIM cards as a new mobile marketing channel. Our service provides a much better click rate than a SMS because it can handle a mini-marketing scenario. It can contain a real call to action, screens, send back to a mobile website, use pre-formatted SMS, … We have been offering it for 8 years and currently carry out thousands of campaigns per year.
Phocea Tech: You would have an example to mention?
Frédéric Martinent: A classic use case is that of an operator detecting that a consumer has a new mobile. With our service, it will be able to send him a message offering an insurance. In 2-3 clicks, the consumer will be able to adhere and will be invoiced on his operator invoice.
The service can also manage consent. With a Brazilian bank, which in this case was our client, we managed the transition to the dematerialized bank statement.
Phocea Tech: Why were you interested in a machine learning approach?
Frédéric Martinent: We tested it in Africa to refine targeting. For one of our customers, we had the right to send to its users only one message per person per week. The challenge was therefore to know which service should benefit from a promotion, which message was the most relevant for both the user and the operator.
Testing Two Algorithms
We relied on several months of campaign history to refine the targeting, with two types of algorithms. Either we look at the similarities between people, considering that two people with similar profiles buy roughly the same services, or we look at the similarities between offers.
At the end of the experiment, we chose to use a mix of these two algorithms, which resulted in a scoring about which service a customer could buy. The benefit was very clear since we got an acceptance rate up 50%. We made sure there was no bias, for example by sending the offers exactly at the same times.
Phocea Tech: As this first test has been successful, what is the next step?
Frédéric Martinent: We are now in a generalization phase. It will be necessary to refine, among other things using more data on consumers. For example, we could determine if some people are particularly price-sensitive, and give them priority offers that meet this criterion.
[Version française: Gemalto met le marketing mobile à l’ère du machine learning]