For sales professionals, the promise of predictive analytics holds obvious appeal: knowing when your customers are ready to take action could offer a serious competitive advantage.
Could is, of course, the operative word, because the ‘predictive’ part really amounts to simple numbers: they don’t illuminate ‘why’. They’re built on big data, but contrary to conventional wisdom, bigger is only sometimes better. Without the proper systems, procedures, and tools, predictive analytics is a very good means of identifying the likelihood of a customer behaving in a certain way, but it’s not necessarily more than that. If you’re a furniture supplier, knowing that customers buy more desks in March is useful, but it’s knowing why that’s the really useful part.
Propensity modelling allows you to use predictive analytics to simultaneously know these things and understand them. In essence, it uses context, relevance, and segmentation to discern both the actions of individual customers and their meaning.
Theoretically, propensity modelling makes predictive analytics more effective: it empowers sales professionals to take an informed, trend-based approach to data gathering and interpretation.
The cornerstones of propensity modelling
There are three foundational principles of effective propensity modelling.
The first is behaviour. This aims to put each individual action within a wider pattern or trend: it identifies correlations, but it lends them contextual weight. Propensity modelling allows you to separate and segment customers according to their preferences and purchase history – giving you an idea of what they might respond to, and allowing you to target them with bespoke sales and marketing communications.
The second is forecasting. Analysing behaviour allows you to sell to customers today and tomorrow. With a steady stream of actionable information, you can pre-emptively up-sell and cross-sell to customers based on the timing of their purchases.
This ties neatly into the third and final cornerstone of propensity modelling: evolution. Customers don’t stay still, and nor should your analytics: over time their preferences change, their financial circumstances fluctuate; their purchasing behaviour demonstrates differences both subtle and major. Propensity modelling is a long-term endeavour, and it needs to adjust to these changes, fluctuations, and differences – so your marketing and sales material can do the same.
Dos and don’ts
Translating this into reality requires some strategic thought, though the benefits are obvious. We have helped a leading car manufacturer use propensity modelling to predict the next model of car someone who had previously bought with them would buy next, with 80% accuracy. Over time, clear 'dos' and 'don’ts' have been established.
Firstly, don’t underestimate the complexity of your customer base. Your data sets must be finely balanced: if they’re too small, they’ll be unrepresentative of your wider audience; if they’re too large, they’ll be impossible to meaningfully separate.
Do keep all information up to date. A customer who behaves a certain way today may behave an entirely different way in a year; their data changes, and your models should change to accommodate this.
This data should be available to everyone who needs it, so don’t restrict it to the sales function. “Right hand/left hand” situations are anathema to customer experience, and undermine the success of your models: the employees on your service desk need to know how customers behave as surely as your business development and marketing teams.
Finally, do think about how you can turn these insights into value for your customers. This goes beyond mere marketing: you can use it internally to promote superior user experience. In financial services, for example, you can address your customers’ needs throughout the entire payment journey. Predictive analytics can indicate the challenges each individual faces as they navigate your systems and processes – allowing you to provide the support needed to smooth other any potential difficulties.
Indeed, ‘support’ should be your watchword: one issue with predictive analytics is when different predictive models are assigned to different products. To continue the example of financial services: a bank may have separate predictive models for loans, credit cards, and mortgages, but sometimes the same customer is at the end of each model – causing internal arguments about which team talks to them about which product in a given month. The company should instead examine the customer’s overall lending need, instead of attempting to sell them on individual services.
By gathering and interpreting data effectively, brands can use propensity modelling to lay the foundations for better campaigns and stronger customer relationships – but these are just the foundations. The data must be used in conjunction with accurate customer information to be truly effective.