Implementing Personalised Pricing – The Practicalities
Testing and Control Groups
In the context of personalised pricing, a culture of testing – creating the conditions to accurately quantify the effectiveness of any initiative – is critical. Most organisations are aware that control groups are an essential part of this, but many treat it as an afterthought or are not prepared to undertake the necessary investment to ensure success.
There is no one-size-fits-all approach to control groups. The appropriate control group size depends on the nature of benefit you are trying to realise, and with insufficient numbers in the group even the best models will struggle to accurately gauge your activity’s success.
Paul van Loon, Forecast’s Head of Analytics, and Neil Macdonald, Director of Forecast UK, discuss this issue in the video below.
The journey towards succeeding with personalised pricing starts with the recognition that experimentation is not only desirable but unavoidable. Although critical in discovering your optimal state, experimentation will take time and comes at a cost. In order to measure the effectiveness of any activity, we need to compare it against the status quo, ie. utilising a control group. If that activity is successful, congratulations, but by withholding it from the control group, a cost is effectively incurred. Further experimentation, although necessary, comes with additional costs.
When considering the components of predictive modelling, what are the true drivers of success? Modern modelling techniques, as used regularly by Forecast, are well established, and there are an array of powerful tools and methodologies available. But a truly successful personalised pricing initiative hinges primarily on the data used to feed the model. Not all data has predictive power, so harnessing the right data has the potential to make a material difference. So, is it as simple as having the right model and the right data?
A successful data scientist relies not only on available tools but spends considerable effort in understanding the broader context of the challenge. Accurately defining the problem, optimally identifying and securing the most effective dataset, and determining the right experiment are critical in generating the best recommendations, as is identifying the optimal means of evaluating the results.
Clearly, with regard to personalised pricing, some of the methodologies underpinning data science are highly complex but have an impressive ability to achieve meaningful results for organisations. But in the process of coming to the resulting recommendations, does the business have a clear idea about how it got there? Is this even necessary? What obligations, if any, do we have towards our customers in explaining how and why they received the offer they did?
Paul and Neil discuss these themes in more detail in the video below.