As communications service providers (CSPs) rise to the challenge of generating value from their big data analytics, Philipp Gröne, a senior delivery expert at Unitymedia, explains how the German service provider has uncovered real cash from its deployment of Guavus big data analytics – an estimated seven-digit euro sum annually in opex savings and a nearly one point increase in overall net promoter score (NPS).
At Unitymedia, we started off by looking at our software and assessing how we can have an impact on our books. We wanted to derive actionable value for our company and, of course, our customers.
We started by considering what customers think when they think of service providers, what is the thing they’re most worried about? If you ask the customers, very often, the first thing they say is they can’t live without their internet. So to be able to differentiate on the customer experience side is very important for us. If we look at the issues we have as an ISP, we need to manage our opex. Therefore we want to make sure we have our technical service cost at a low level. To achieve this we want to avoid calls, truck rolls and tickets.
Customer experience is a big one for us. In the telecoms market, it’s quite difficult to develop a unique selling proposition (USP) beyond quality and price. To protect your top line you need to differentiate by providing a very good customer experience. What we wanted was to have real-time actionable insights so what we see and what we have on the data side drives impact in our business. We wanted to translate that directly so we avoid calls, tickets and truck rolls and reduce our opex, we also want to improve our NPS.
To achieve this we looked at how we can do this and what kind of platform we can use. It’s a make or buy question, but eventually we decided to work with Guavus, a Thales company, which offers the capability to have impact and deliver the value we were looking for. We targeted two use cases to assess and said, we have platform that we need to feed with all kinds of customer care interactions. That’s everything that you have got in your IVR system or your ticket system or for truck rolls.
We wanted to understand what the cable modem telemetry is working like, and how the customer premise equipment (CPE) error system and the network alarm system are performing. We want insight into the Internet Protocol Detail Records (IPDRs), network topology and how the network is structured. All kinds of custom information, such as what kind of CPE, what kinds of products, software, firmware and so on need to be considered and included. This information was then fed into the platform and delivered in real-time to use cases which were relevant to the business owner, which was the technical service site. The focus was to detect anomalies and predict outages and of course you always get the timeframe pressure, in this case five months to deliver.
We chose an approach where we said: We know we have a real-time data platform so let’s go back in time and take the data from all the sources I mentioned and take data from the last six months and feed that into the system and simulate over the entire period what the platform would have delivered. Afterwards, we asked Guavus to give us a list of anything they would claim their platform would deliver. We went back in time and actually were able to confirm what we would have been able to see, what real value would have been created.
In the call centre if you have 100 calls per hour, our internal goal, our key performance indicator (KPI), is to make sure 80% of our customers have a maximum waiting time of 20 seconds. So, what if a problem occurs? Let’s assume in our network we have an issue where CPE type one has a problem with node two and all the customers in that segment are being impacted so people are calling in. There is a normal fluctuation in the volume of calls coming so the issue is not really recognised by our staff or colleagues – at least initially. People call in and you have 105 calls or 110 calls per hour and agents feel this is a little bit out of the norm but it’s not until you’re hitting 120 calls per hour that people know something is wrong.
Once you’re aware of a problem you need to identify what kind of problem it is. Some agents might correctly identify that we have a CPE type one issue but others might say we have a node two issue, which is also correct. Further agents might falsely recognise a different type of error by saying maybe it’s something to do with the product or something internal, let’s send a truck to the house to sort it out. Again opex is wasted.
The mean time to repair (MTTR) from the problem occurring to the solution is therefore taking some time. This causes more tickets, more calls and more truck rolls. I estimate calls cost in the range of three to five euros, a truck roll costs somewhere in the range of 50 to 120 euros and then we have tickets and all of that is, of course, impacting our customer base’s experience which is more and more a key differentiator in the telcoms market.
With Guavus in the background monitoring the calls coming in, we know earlier what CPE the customers have, where they are in the network topology and agents will be able to identify that customers with CPE type one on node two are calling in on a much higher basis than normal. We know something is wrong much earlier and agents know to watch out because there is an awareness of something being wrong.
This was a reactive use case in which we reacted to customers giving us information but there’s another part which is predictive. Historically, if you go into the network data we have there are sometimes patterns that occur prior to a network incident. You can see different types of upstream or downstream numbers which result in a specific kind of issue. What the platform says here is: If you see that pattern again, which we learned in the past, we can predict with a high level of certainty where the issue may occur before the customer has any kind of impact. These are the use cases that help us solve the problem before it occurs, thereby reducing our time to respond and making sure issues can be managed more effectively rather than reacting to something.
Real cash
We estimate that we have made savings of 4-7% on calls, a similar saving on truck rolls and a reduction in ticket costs of 5-8%. If you think about what that means for a provider with 40 million customers, this is real cash, this is a lot of money. On the proactive, predictive maintenance use case, we estimate annual savings could total a seven-digit euro sum.
In addition to these traditional opex-related savings, NPS also increased by slightly under one point overall. If you’re able to increase your NPS by one point this has a financial impact which sometimes outweighs the official opex reduction because you have so many other benefits behind it.
