Don’t Rely on Witchcraft: Question the Status Quo of Customer Analytics

Late 9^th century, somewhere in Europe, a fictitious story:

Life is hard these days. The attacks of the Vikings have always been brutal, but this time it was worse. Your body is peppered with wounds from their battle axes and spears. Your body is aching and the wound fever is rising and draining energy out of your cells.

But there is help - you can count on the village healer’s support. The wise old woman follows the secret old healing traditions passed on through generations of her family. She places a crow feather on your forehead. The air is filled with heavy scents of burning sandalwood and thyme. Rhythmic witch spells – in what sounds to you like a foreign language – create a mystic & slightly frightening atmosphere. Despite the unspecific feeling of anxiety, you put yourself in her hands and let things happen.

The above short scene could well be part of the plot of a typical medieval movie. Certainly (and very thankfully) real life has changed tremendously for the better compared to medieval days. The descendants of the Vikings are very peace-loving European neighbors, the average everyday life never experiences anything like the above brutality, and modern medicine relies on solid science and proven treatments for the root causes of the disease.

Nevertheless, every now and then - especially when thinking about telco customer retention - some strange and diffuse feeling of uneasiness reminds me of the healer. An illustration …

Early 21^st century, somewhere in Europe, a fictitious story:

It is hard to imagine how life without mobile technology would look like. Mobile services became essential to you. You have managed to live with the white spots of mobile reception, and drop calls almost became part of the normal service. But lately your mobile became just “unusable”. Your calls get interrupted several times during your typical route of travel, your battery dies at lunch time, and your friend pays only half the price for what seems to be a very similar service.

But there is help - you can count on your operator’s support line. The customer service agent offers a wide range of tariffs which have amazingly innovative names - although they seem to comprise well-known services. Right away in the call the agent offers you a 6-month free trial period for an additional virus scanner “to show our gratitude for your service” and to apologize for the “inconvenience you might have experienced”. Every now and then - typically when you asked a question - the procedure gets interrupted by repetitive music and friendly computer voices asking you to stay on the line as you are “important to us as a customer”. It is hard to follow the discussion as it is filled with acronyms many of which you do not really understand. An atmosphere of unspecific suspicion arises, but to the end of the call, you agree to sign up for a new 24-month contract.

Although the above analogy surely is of a provocative nature, I nevertheless ask myself why other industries and disciplines (like medicine, science or sometimes even government) put huge efforts into identifying the right (big) data and exploit it through innovative analytics. Just to name a few, examples come from a wide variety of disciplines from DNA sequencing and analyzing to optimizing complex cities through smart IoT analytics and real-time streaming analytics of dozens of petabytes of data from the Large Hadron Collider at Cern.

But what is current practice in telco? Even leading telcos very often still rely on nearly the same kind of data sets and the same traditional analytics they already applied ten or more years ago. Like the witch in the above short story lots of telco data scientists (have to) rely on “inherited” models and analytics. They (have to) feed their models primarily with data like sociodemographic profiles, calling and top-up behavior, extrapolated NPS scores or “needs-based” segment information. Yes, the more innovative data scientists experiment with adding data points like social media or web. I do not intend to sound cynical, but using the above analogy, does it really make a difference if the witch adds a spider leg to the treatment?

You disagree? Congratulations, you are perhaps one of the very few who managed to overcome the current as is. But considering the almost countless discussions I had with telco customer management executives, retention managers and data scientists, I had to acknowledge they more or less all openly share my observations.

Do telcos really understand the majority of the individuals of their customer base well enough to design real customer-individual activity plans? The answer to the questions – with some very few exceptions – is a very clear NO! At the best, most telcos have the right analytics in place to get along ok with perhaps the “best” 10-15% of their customer base. And even for those, they perhaps “know” what to do, but not exactly why - not to speak about any insight of how the situation can be improved other than gradually.

Obviously, there are reasons for why the situation is as it is. The reasons range from technological challenges, to organizational barriers, budget restrictions, legal boundaries, missing tools, resources and knowledge and many more. But it is certainly worth it to overcome those. In mature telco markets like in Western Europe, an average telco spends roughly 20% of its overall operational expenditures (opex) on customer management. Without going into any kind of financial details and leaving all side effects apart - e.g. substantial reduction of network roll-out costs (Arias2015) - this number alone indicates the huge potential of a more fundamental readjustment of telco customer analytics.

From witchcraft to modern treatment

How would a doctor nowadays act differently compared to the medieval witch? First of all, she would carry out a very thorough medical examination. For example, she would examine the wounds carefully, perhaps test all vital functions and examine the patient for hidden injuries. She would obviously rely on her deep medical knowledge and combine all the different “data points” to come to a very individual, well-balanced, cohesive and fact-based diagnosis. She would obviously also ask the patient (sometimes very specific) question, which are relevant to the matter, and incorporate the answers into her diagnosis. Nevertheless, she would consider his feedback carefully as patients sometimes make quite misleading statements. For example, patients in the mid stages of hypothermia become very happy and incredibly energetic, as the body instinctively makes one final effort to warm itself up, shortly afterwards they become unconscious and eventually die.  The doctor would also try to eliminate certain diseases to improve her diagnoses.

Only then would she define a specialized treatment including the best combination of medication – all based on medical evidences. She would take possible side effects for this individual patient into account and evaluate how different medications influence each other. Then she would apply individual treatments: She would clean and nurse the wounds using - if needed - antiseptic ointment, prepare a saline infusion with the right individual mix of drugs stabilizing the patience and treat him with the right dose of pain killer to ease the actual discomfort and minimize side effects.

How does this all relate to customer management and customer analytics?

If our aim is to take customer management to a complete new level of effectiveness, we have to follow a similar, much more structured approach, act in a more scientific style. Obviously, it is beyond the scope of this blog post to detail down the desired approach in all features, but let me outline the high-level structure.

1. Individual examination
We start with the customer, neither with the data we have nor the analytics we can apply. This does not sound like a new statement, but more or less common sense. We are bombarded with marketing slogans like “customer first” or “customer centric” or many more. Unfortunately, when it comes to customer management and analytics, very few telcos really follow this common sense. For me the currently existing churn models are perhaps the clearest proof: individual service experience and value for money (in the absence of real product differentiation: price) have been for years by far the most important reason for churn for a very significant portion of customers (Dass-Jain2011). Nevertheless, most telco churn models are not able to provide a real comprehensive measure for the customer’s service/network experience as they do not even incorporate service/network experience data. No, I am not talking about simple drop call counters derived from call detail records (CDRs) or about the (often categorized and extrapolated) feedback from customer experience surveys. These pieces of information are simply not detailed enough to understand anything like the individual service experience. We can prove through analytics that for some customers a certain set of specific network failures are the primary reason to churn, whereas other customers tolerate the same set. We know for example that the sequence of failures, the type of usage at the time of failure and the failure density are important (and not so much the isolated number of failures). In order to analyze this kind of churn drivers, you need to source very detailed network data (e.g. SS7), integrate it (e.g. across different bearers), but keep all information with indicates any kind of service deterioration.

On a side remark, talking about direct customer response: Such feedback is a two-edged sword anyway. First of all, you can almost never obtain it in any kind of relevant detail for a significant portion of your customer base. This makes it per se tricky to use. Secondly, customer feedback is hard to compare. Even an “8” or a “9” on a simple and standardized NPS scale will mean different things to different customers.  Such feedback – and even more so free-text feedback - has to be filtered and sometimes even ignored – just remember the hypothermia example above.

Individual customer price pressure is another example for a weak area in customer analytics. There are very few telcos today which understand the price pressure for all their customers individually. They are not able to answer basic questions like “Which tariff from my portfolio/in the market is pricewise the best for a particular customer with his individual usage?”, “What maximum price premium is this particular customer prepared to pay?” or “Which competitor tariffs would allow the customer to save more than the above maximum price premium?” But how can a telco ever understand a customer’s price-driven churn potential if they do not know these basic facts? All of the above data and analytics to answer that kind of questions are available and industry-proven. There is very little reason not to apply them and generate actionable insight on a totally new level.

2. Cure the root cause not the symptoms (alone)
Today still many customer managers think (only) of campaigns and offers when it comes to retaining customers. Campaigns and offers are not necessarily a bad thing, but honestly, we all know that giving some goodie to the customer or talking him into a new contract will not solve the churn challenge fundamentally. There is much more a customer manager has to incorporate into his thinking than campaigns. Let us for – just for illustration purposes – revert to the initial example again. The customer is experiencing severe network problems. The free trial period for a virus scanner will at the best make him recognize some appreciation, but not in any way change his propensity to churn. It simply does not cure the root cause. Even worse, if the additional virus scanner is not providing any additional value, the effect might well be the opposite. Sticking a plaster on the wound makes it disappear for a while, but only thoroughly cleaning the wound and applying antiseptic treatment eliminates the root cause.

It is also fundamental that the diagnoses as described above eliminates churn reasons. Treating non-existing churn reasons is at best a waste of resources/money, at worst it annoys the customer and potentially even drives churn. In reality, I have seldom seen a churn management process that specifically looked at eliminating churn reasons.

3. Develop an individual treatment plan
Looking at our “real lives” we intrinsically know that an isolated action will seldom create a lasting relationship. The sense of trust and appreciation comes from the sum of experiences. If on average these experiences are relevant and positive for the customer, he will even forgive single negative experiences. Astoundingly enough, research shows that after experiencing a couple of positive events a well-handled negative event will even reduce churn by 67% (Kolsky2015). This train of thoughts has a significant impact on how we manage customer relationships and how we build analytics model (e.g. churn models).

Consequently, a customer management strategy which purely builds on a next best activity (NBA) or, even worse, on a next best offer (NBO) approach is too shortsighted. We need to plan the best sequence of activities for each individual customer. I even prefer to use the term “event” in exchange of the term “activity” as the latter might suggest restricting our thinking to company-initiated events.  On the opposite, we also need to consider how to optimize events which are initiated by the customer. Let’s say you are a high-value customer, lately changed your bank account and therefore unfortunately missed to pay the last monthly bills. Just image the negative loyalty effect your operator would avoid by individualizing the collection and dunning process. We have to plan the optimal, individual customer journey for all our customers. For obvious reasons, this planning has to happen on a rolling basis which creates some complexity. But also in this context industry-proven methodologies and technologies do exist. They “just” need to be applied (Teradata2017).

The above also has implications on analytics. Currently most (retention) analytics have far reaching limitations as they are static in a way that they look at a “status” at a certain point of time - now or in the future. But they do not analyze any sequence of events and their interdependencies. They provide no direct explanations for why events occur and do not incorporate customer interaction in its entirety (for more info: Hassouna2015). But as described above, this is not how (customer) relationships work. Analyzing customer-related event paths and predicting future events provide a totally different view on how each customer relationship develops. Furthermore, this view creates highly actionable insight as almost each event can be influenced. You can plan and - based on prescriptive analytics - change the future event path for each customer relationship. For me, this is the essence of customer relationship management.

Do not get me wrong. From burning sandal wood and witch spell to modern treatment, medicine has gone a long way – and similar is obviously true for customer management. But as Oscar De La Hoya rightly said: “There is always space for improvement, no matter how long you've been in the business”. And in order to innovate, you have to question the status quo. I hope the above blog provoked some fresh thinking and can contribute to develop telco customer management a noteworthy step further.

In case you disagree with some (or all) the above or want to discuss, please reach out to me on LinkedIn or twitter.