How many tourists swap to a local SIM while travelling in New Zealand?
Recently Qrious was asked whether we could use our aggregated and anonymised mobile location data set to help determine approximately how many international tourists purchased a prepay New Zealand SIM card rather than using the international roaming services provided through their overseas mobile provider.
Why was our client interested in this? Because, although international tourists that use a prepay New Zealand SIM tend to spend more than the average kiwi on their mobile phone, they are also only active for a much shorter [time] than the average prepay mobile customer (hardly surprising given that international tourists are only in NZ for a short time!). The dilemma this created for our client was that one of their key performance indicators – (which measures the volume of customers that stop using mobile services each month) – would occasionally spike higher for no easily explainable reason. Our client had a theory (more a gut feeling really) that these spikes in churn coincided with peak tourist seasons and therefore that the higher churn was probably due to an increasing number of tourists purchasing a NZ SIM card for use whilst they travelled the country. If their hypothesis was correct, then they would know not to over-react if the churn KPI spike was due to an influx in tourists. But how could you validate this theory?
Enter Qrious. Every day Qrious processes approximately 1 billion rows of anonymised mobile event data. Interpreted correctly, the data can be used to understand the movement and behaviour of different types of customer groups.
We approached our client’s problem by first filtering out any customers that we knew were inconsistent with the agreed definition of a typical short stay international tourist. So we immediately ruled out any customers that had been active on the network for longer than 2 months. We also ruled out any customers that had “ported in” their mobile number.
Because we had defined a typical short stay international tourist as someone that was likely to have been actively sight-seeing around New Zealand we also needed a filter to only include those customers that had moved around the country more than the average customer, travelled further than the average customer, and who didn’t stay at a single location for very long. However, that’s easier said than done. Mobile event data can be quite messy. We needed to smooth out the “noise” originating from cell tower handovers and bouncing effects by grouping events based on their location. Clustering billions of events is a challenging technical exercise and to achieve this Qrious utilised its state-of-the-art distributed computing platform to run the necessary machine-learning algorithms at scale. Once the events were clustered we enriched them with features such as the number of consecutive days a cell tower cluster was contacted by a particular mobile device, the hour of the day, day of the week and the duration of events. These features helped us to identify devices showing typical tourist behaviour.
Profiling this subset of remaining devices provided some very interesting insights. This group of “likely tourists” were found to be six times more likely to have visited Queenstown airport than the average customer. This group of customers also made and received more than three times as many international calls than average. In fact, all of the usage and behaviour attributes that we profiled were consistent with our definition of a typical short stay international tourist. So at this point we were pretty confident that by using these usage and behaviour filters, we could identify what portion of customer churn was attributable to short stay international tourists. And for our client, this was exactly the information they were after.