“It’s Hard to Convert Loyalty into Euros and Cents”
Interview with Marco Benninghaus, Connected Touch Point Suite, Tribe Lead
Marco Benninghaus insists he isn’t a hardcore business person, but rather a “heart-core” kind of guy. For him, it’s a labor of love to make sure that all touch points where a customer engages with Deutsche Telekom are well connected. In order to do that, he’s working with a collection of marketing tools called Connected Touch Point Suite (CTPS). They gather, analyze and visualize insights into customer use of touch points. Going forward, these insights will be leveraged even better with the aid of artificial intelligence.
I recently booked a ski trip, and now I’m being inundated with more travel offers, but what I really need is a warm sweater, gloves or moonboots. What’s wrong with this picture?
It’s a classic failure mode. Amazon, for instance, often keeps recommending items we already bought. That might be OK and even useful for consumables such as ink cartridges since it saves me time searching for products. But sometimes I receive recommendations for for things you certainly won’t buy again anytime soon, such as a trip, luxury items or tech gadgets. So-called targeting simply doesn’t work perfectly every time.
We want to prevent that by using AI. Every customer shares information about her several thousand times a day. We can remember many of those details and can use them when we communicate with the customer, but neither can we nor are we allowed to store every last detail for future interactions.
Deutsche Telekom, though, has committed to get better at this thanks to its Connected Touch Point Suite...
We offer many digital touch points where a customer can engage with us. We don’t want to lose the information about a customer’s interests, his state of mind or habits in particular. Rather, we want to remember them and make them available at any other touch points, assuming the customer has agreed to it. Touch points come in many shapes, from apps, websites and social media to products such as Entertain TV. Right now, the information we collect is siloed and we can only use such data at that particular touch point. The Connected Touch Point Suite will help us bridge those gaps.
Sounds like a lot of technology and a massive database.
That’s a fair description. The key is for us to build those technical bridges and treat data with respect because sometimes a customer shares details about a specific question they don’t want to see reused. There’s a good reason we speak of the so-called “creepiness factor” in targeting. At some point, it would become really eerie if we used one piece of information to creatively infer something completely different. You need to strike the right balance and show respect. In the name of treating the customer fairly, there are certain pieces of information that you may be legally allowed to use but won’t.
What’s the goal here for Telekom?
It’s not about collecting data, but to have a better dialog. We want to speak with our customers based on what we know about their behavior and we want to analyze usage to improve the user experience. We accomplish all that without knowing who you are, what your name is, where you live, how old you are -- we only look at anonymized user behavior.
So what you’re after is building a sustainable relationship with your customers?
Customers will only stay with us if we treat them with respect over the long run and only approach them on those few occasions when it makes sense. It’s hard to convert loyalty into euros and cents, but experience shows that providers who do a good job, don’t annoy me and treat me fairly give customers few reasons to leave. Long-term relationships maximize earnings much more than selling a customer five extra subscriptions over the course of the first two years with a provider.
It sounds like this accumulated knowledge helps all product managers at Telekom to have a better customer dialog. Don’t you also have to create a common language or define shared standards?
We make it as easy as possible for each product manager to connect with the databases where we collect user behavior. That’s why it’s so important to create standards. We can’t have 20 Telekom apps, each with its own way of handling and defining data. For instance, what constitutes launching an app, and how do I define a session? If it’s comparable, we should all use the same tools. We wouldn’t be able to map the customer journey if those apps identified their users with different tools and stored user data with yet another tool. For that reason alone, we had to use the same tools, systems and backend service providers across the board. We as Deutsche Telekom want a holistic view into the customer and think it makes sense to define standards for that. Now we’re able to speak with a single voice instead of touch point by touch point. Before, we’d see four customers if the same person contacted us through four of our apps. Now we can treat him or her as one and the same individual and make sure she doesn’t receive more than a handful of push notifications instead of being bombarded by each of her apps.
How do you verify the data is interpreted correctly?
We can’t verify it yet, that’s what we’re aiming for participating in the eLIZA project. Before we joined, we could look at user behavior, store and interpret it and draw some conclusions. If you have enough user data you can gain quite a bit of insights, but they’re all hypotheses. AI will give us the opportunity to do what we were lacking so far, that is to permanently verify if our hypotheses were correct. It gets harder and harder the more customer signals we receive and process since they’re not always logical or make sense. At one point we have to bring in AI to recognize patterns humans can’t see.
Where exactly do you want AI to help out?
We want the AI to predict upcoming customer needs to really offer targeted services and products. It’s hard to do that with rule-based algorithms. For certain decisions, a rule-based system with the help of a seasoned human editor works, even without AI. But we are approaching a level of complexity that calls for AI.
As recently as last year, we didn’t have the need for it in our CTPS, it would have been complete overkill. So you have to take a good look if it’s worth it, and that’s what we are testing this year. It doesn’t always pan out, quite the opposite. Take an AI that recommends a customer ditch their mobile phone because he only sends SMS and doesn’t really need it. It also wouldn’t help us if the AI only made absurd suggestions that annoy the customer. It begs the question how much autopilot we need -- perhaps we do need a human plausibility check after all. So we have lots of things to explore this year.