Pricing Best Kept Secret: Optimization
Something is cooking right now in Big Data… After years of test-bed initiatives, companies are now launching more projects in production (mainly around data) and starting to look at how to extract value with advanced analytics.
As an avid reader of the great contributions of Jean-François Puget (Chief Architect Analytics Solutions at IBM), I have tried to ground ideas around analytics maturity and optimization into the pricing field.
First, the vast majority of companies are still at early levels in pricing maturity. Large efforts have been made in the past to solve the data issue in pricing – mainly around transactional data. Players like Demandtec (now IBM Omnichannel), Pros, Vendavo and Zilliant have made a great contributions to the field by evangelizing IT and Business players, and demonstrating the profitability that could be achieved through improved processes and tools. Better analytics and tools are getting an important traction, and cover a large part of Pricing initiatives’ budgets.
Is this important? Obviously yes. A recent survey by Aberdeen Group (2014) showed, among other consistent results, that 90% of leaders (sample top 30%) where able to “maintain consistent and competitive pricing, regardless of channel” than followers (at 20% only).
Is this enough? No. Mainly because in the vast majority of cases data sets are purely transactional (ERP based), and analytics stay at a descriptive level (waterfall, dispersions, etc.). Most of the value, and particularly in pricing, comes from blending different data sources together (CRM, Research, Panels, GIS, Open Data, etc.), and from more advanced analytics (predictive and prescriptive).
Elasticity is a start, not an ending
Where to start? Elasticity? No. The first step should be segmentation. Data driven segmentation is often lacking in pricing analysis. Cluster analysis, as one of several segmentation methods, should be systematically used on transactional data in early stages. And this first step should irrigate any further analysis. Combined with other sources, you get a powerful structure to reveal actual and contextualized pricing behaviors, such micro-segments being added as extra variables to your pricing database.
The same happens with elasticity. Elasticity and cross-elasticity are often presented as the Holy Grail in pricing. An unachievable quest. Running elasticity measures is pretty straightforward (multivariate regressions). Using them to impact pricing decisions is more complicated. Why? Because they tend to be considered as a solid metric when elasticity is highly volatile, non-linear and context dependent. Going in back to data variety…
Rules exist to be changed
So, we have a variety of data souces, blended in one single flow, extra variables (segments, elasticity, etc.), it is now time to get serious. Predictive analytics have been widely used in other disciplines (CRM, Credit scoring, fraud detection, etc.) but rather limited in Pricing. What can we expect from this type of analysis? First, association rules are one of the major applications. Market basket analysis, promotion performance analysis (decision trees) can be leveraged in pricing for instance. The second field is a robust description current and future behaviors: purchase likelihood, loyalty and churn, purchasing patterns at different product x price combinations, etc. The output of these two types of predictive analytics are rules (“if…then…”) that can be included into current systems (CPQ for ex.) to automate adjustments to an always moving environment. Going back to data, velocity is key here to influence real-time decision.
Optimization is the next big thing
The last big thing is optimization. Not because it is new, of course, as sectors like airlines or hospitality have been using it for decades. The first reason, as reflected in the model above, is that optimization and prescriptive analytics allow to reduce the distance between data and decisions as optimization is by nature oriented toward finding the optimal results from multiple alternatives. Pricing is complex and optimization enhance human capabilities to make a decision by simplifying the underlying complexity of the decision to be made. The second reason is that pricing decisions are also by nature surrounded by uncertainty. Stochastic optimization methods allow to tackle this uncertainty and allow to reduce the risk when using measures such as elasticity. The last reason I think, is the direct result from optimization projects. Descriptive or predictive analytics work row by row, at a granular level while the decision to be optimized are more global: total portfolio value, total ticket per POS, etc. When running an optimization analysis on pricing data, winners and losers emerge, increases and decreases for ex., allowing an optimal combination of multiple and complex inputs. Reducing this complexity, getting closer to the final decision to be made, is a great support for pricing decision making and a driver of superior results.