The CRM department at Ticketmaster International faced with the problem that their existing predictive analytics stack without significant investment was not able to handle the increased load caused by the new BIG data sources integrated recently. They realized that the integration of new BIG data sources and the analytics capabilities on the top was relatively painless due to the availability of cloud infrastructure and BIG data tools, however the low latency large scale predictive utilization of this information in their marketing automation suite is blocked by their predictive analytics stack that was purchased from a big vendor a few years earlier for a significant investment.
read more
It was likely that a similar magnitude of investment would solve their problems only permanently and it would increase significantly their dependency on the vendor they have chosen earlier.
The business executives at Ticketmaster International aimed to find a better solution that achieved maximum scores evaluated by the following criteria:
Due to the strict specification first TicketMaster International considered to hire additional staff to develop the required solution in-house, but they realized that this approach will be costly, time-consuming and project risk associated would be also significant. Finally they managed to find Predictron Labs and participated in the closed alpha test of the framework.
The results can be summed up in three words - faster, better, cheaper. Using the predictive models provided by Predictron Labs Ticketmaster International managed to achieve significant uplift in prediction performance compared to their existing models. The new solution also helped to streamline their scoring process and lower the scoring time meanwhile it provided an environment to monitor scoring processes and prediction performance. Due to the easy use of the framework interested data engineers with good sense of understanding of business problems managed to develop models that proved to be useful in various marketing scenarios. The turn-around time between model development to ready to deploy state shortened from weeks to days.
Our client, the subsidiary of one of the biggest mobile telecom provider in the EU, was aware that its churn models have suboptimal performance which tended to overstate the churn rate and the resulting success rate for acquisition campaigns were equally fantastical.
These churn models, it turned out, were trained on events that were not real churn but on rotational churn (sometimes called spinning). These events are typically triggered when an existing customer gets a new subscription from the same provider usually to take advantage of promotional offers targeted new customers.
read more
As an operator in a very competitive and commoditised market with so many different customer segments, each having very different motivations for churning, our client needed a better model to identify the real churners or segments of churners and focus retention strategies on them.
The aim of the project was to identify and exclude rotational churn events based on similarity of call usage and location based attributes between churned and newly created contracts. Using predictive modelling techniques we were able to identify (with 95% probability) the events of rotational churn, which were then filtered out from the traditional churn database.
By retraining these models using real churn events, the performance and accuracy of churn prediction was significantly increased, resulting in a targeted profile of churners. This also meant our client was better informed on the reasons for churn whilst improving ROI of future retention campaigns.
This was a follow-up project for a leading telecom operator, now pushing to increase product usage and penetration across its network footprint.The operator already offers triple play services (broadband, landline and television) to its customers but needed to sell additional high value services like on-demand video service, domestic security monitoring and device protection insurance. By selling these additional services, it is thought that customers will use more of the operator’s products — reducing churn in the longer term by creating higher switching costs for such customers.
read more
To gauge customers’ interest and drive forward this new proposition, a market research was commissioned to provide insights, amongst other things, into: market readiness, price and value points, and customers’ likely consideration for similar proposition from competitors. Given the findings from the market research, the operator needed to 'zero-in' on the key factors which most closely correlate to particular purchasing behaviours, and so better predict customers' likely responses to the new service proposition.
To solve this challenge, we needed to join-up the dots by combining market research insights with transactional data attributes such as product usage, lifetime value, churn probabilities, loyalty status, and other behavioural data in the CRM database. By adopting a data fusion approach, using probabilistic data association algorithms, we were able to identify from the CRM database which customer groups were sampled in the market research and using a number of predictive models we predicted their attitude toward the biggest competitrors and their affinity toward additional products.
The CRM behavioural data further provided the explanatory variables needed to predict the probable response for each customer, compared to the relatively small response gathered from the market research. Crucially, this fusion approach is more indicative of actual customer response as it helps to overcome response bias associated with attitudinal opinions gathered from the market research.
Based on the success of this project, we later applied the same method on a countrywide household dataset to determine regions with the highest interest for such high value services, which relies on a high-speed fibre optic network. The outcome of the analysis was used to inform key decisions on the fibre network development plan as it identified areas/regions with the best ROI from a fibre rollout programme.