© 2022 Black Swan Telecom Journal | • | protecting and growing a robust communications business | • a service of |
Email a colleague |
May 2013
The software dating game is now in full swing in telecom analytics.
A few dozen software vendors have entered the biz in recent years. And in that time, the telecom analytics market has matured, brought on the innovation, and treated most telecom operator clients to a nice ROI.
So now begins that age-old courting ritual where software vendors show their plumage, flap their wings, decorate their nests — you name it — to attract one of those affluent fellow-fowl — an Oracle, Amdocs, Netcracker, Tektronix, Ericsson or IBM.
Unfortunately, only a handful of those analytics vendors will marry into big money for the number of eagles who soar over TelecomLand has declined in recent years. Yet, as a consolation prize, there are still a few circling hawks (Redknee, Hansen Tech, Neustar) who are willing to invest if the price is right.
But what if these birds of prey could pre-empt the dating game altogether by buying into a software market early? Wouldn‘t that save them a bunch of time and money?
Well, that’s actually Comptel’s strategy. Long known for its mediation and fulfillment claws, the Finnish B/OSS company has attracted a large flock of customers too — 290 operators across 86 countries. And over the years, Comptel has smartly anticipated its customers‘ needs, and either invested early or organically built almost every hot B/OSS application that telecom has seen in the past decade or so — from network inventory to policy management.
In the analytics area too, Comptel invested early, announcing the acquisition of Xtract, another Finnish company, in January 2012. Here to explain how Comptel is leveraging that strategic investment is Matti Aksela, vice president of analytics.
Dan Baker: Matti, tell us a little bit more about the origins of Xtract and the synergies Comptel hopes to get out of buying the company. |
Matti Aksela: Sure, Dan. Before Comptel acquired the company, Xtract had been around for ten years serving customers in insurance, retail, and the media businesses. When I joined Xtract about seven years ago, prior to the acquisition, that’s when we started to really invest in our product development to build a global analytics presence in telecom.
Our mission was looking at the full breadth of operating problems, with churn being the most obvious, but also issues such as optimizing mobile top-ups and stimulating recharges.
Now when you take the DNA of Xtract, which is information science and machine learning, and you combine that with Comptel’s ability to process billions of CDRs per day for hundred million subscriber networks, I think we can derive a lot of value for our telco customers.
Looking back on Comptel’s history, in one way or another you’ve leveraged mediation in almost all your products, I guess. Analytics is another example. |
Yes, for example, with CDRs, you can predict demographics by driving a kind of iterative algorithm across network usage, determining values from only a small percent of subscribers, making predictions based on what neighbors you have in your social circle, as well as the data we have on them. If most of the people you contact are females in their twenties, you can reliably make predictions on the callers‘ profiles. But you don’t need to rely on only two data points.
The modeling we do sometimes has us maintain about one thousand unique features from the CDRs alone: time of call, length of call, location of the call, etc. Then you add to that mix any intelligence we can get on the subscriber, such as the device they are using. In some cases you have a CRM system or a subscriber information system to draw on telling you how long the phone number has been active, what kind of service plan they are using, and maybe even a customer address.
CDRs are a rich source of information, especially in markets like prepaid, where you don‘t have much information on subscribers because they haven’t registered with you. Here, the only intelligence you often have comes from watching how the end-user uses their mobile phone, how often they top up, etc.
The usefulness of the CDR comes together when you do social network analytics on the call interactions and utilize that as a part of the predictive analytics, which is at the core of the value that we add. Our algorithms generate the full social network based on historical interaction among subscribers. We can note how many neighbors you have and how active you are in your social circle. The information you have on each person in the social circle influences the mathematical profile of how you characterize the social circle itself, in addition to others in it. Then you apply machine learning to get your results.
The data that surrounds analytics decision-making can be visualized in four dimensions.
So now let me walk you through a quick use case around recharging the phone to show you how we leverage those dimensions.
Twice a month a customer of ours was making recharge offers to their customers. They differentiated their offers based on usage, such as the low, medium, and high usage bands of the subscribers. Through trial and error, they iterated these numbers and achieved fairly good take ups in all those usage bands.
Then they asked us: can you achieve better results?
The difference with the predictive approach is, rather than rely on simple rules and three broad customer segments, you instead create a model based on individual subscribers‘ information that forecasts the probability of the subscriber accepting this offer given that incentive.
Then we calculate probabilities for subscribers, accepting an offer of, let’s say, $10.00 in exchange for $2.00 of free credit as a reward. And this offer was automatically trained into our model and run into a test campaign. We then evaluated the results over all possible combinations to train our models to come up with a maximum expected value of the offers minus the incentives.
In the end, our trained predictive model delivered a significant increase over the customer’s own campaign. That was worth hundreds of thousands of Euros in revenue improvement, and frankly, our customer was surprised at the improvement we achieved.
So the point here is that most of us are comfortable balancing in our head the simple three dimensions. But when you add the fourth dimension of predicting based on 10, 20, 100 or even 1,000 variables, you step beyond the reach of human understanding and this is where the machine algorithms do the work and deliver better results than any simple rules of thumb could produce.
How important is having a real-time capability in analytics? What can do with real-time that you can‘t do at a more relaxed campaign pace? |
Well, real-time decision-making adds a lot of power, but to achieve it requires a things, namely: a good architecture, an understanding of the data, some efficient big data processing, and an ability to put predictive models in the middle of it all.
At Comptel, that’s what our Event — Analysis — Action framework is about. Using data traffic analysis as an example, we could drive smart real-time throttling of data and enforce policy to relieve network congestion. So, combining all of these elements to make it really happen end-to-end — that’s what excites the operators.
Ensuring VIPs get a high quality of service is another good use case. If the customer is getting poor quality of service at the wrong times, it’s very irritating to them. So if you see the subscriber is experiencing dropped calls, you can take action. So our predictive models are trained to recognize the effect of QoS dropping, factoring in things like increase in churn risk to assess the true impact. By integrating this capability with our mediation strength, we can drive actions in real-time.
So now the system is able to say: “Hey, a VIP subscriber has a QoS problem,” and we initiate a direct action. That action could be something simple like sending an apology message with free credits. Or it could be taking action on the network, perhaps to increase the user’s priority on the network. But the key is to have the predictive analytics tell you what you should do and when in order to truly be able to bring the full value out of your data.
Matti, thanks for this briefing. I learned a lot. Where do you see analytics going next? |
I think the analytics field has a long way to go in terms of driving a telecom’s business. Today, much of the focus is on what can be done to solve individual subscriber issues or treat individual cases with, for example, churn prevention steps.
In the future, I foresee we’ll get better and better at exploiting the data, bringing to life the whole Event — Analysis — Action framework and combining real-time information with Big Data-based profiles to drive actions on a more holistic level. It is crucial to bring in the subscriber data, technical data on the network, and external data sources to utilize all the information on the subscribers. This helps improve the customer experience by driving actions based on the context, and utilizing analytics to truly drive business benefits for the operators.
Copyright 2013 Black Swan Telecom Journal