Gartner recently coined the term “Algorithmic Business” to state that data volume or variety do not really matter, but what companies do with that data – how they turn it into proprietary algorithms – is a powerful competitive advantage and therefore the cornerstone of business growth.
Most of Pricing professionals are lagging behind in the digital transformation, and despite an important investment in software and data capture, most of the companies are facing a black box when it comes to understanding how pricing decision is decided and executed.
Mastering the Algorithmic Era in Pricing means creating in glass box situation within the organization, by getting back control over pricing as a key competitive advantage.
Where to start?
To decide on the main algorithms to use, a Pricing Director must understand first how Pricing Optimization works.
Our framework, based on the seminal work by Robert Phillips (2005), describes the main layers to take into consideration:
Data Sources: must have data (transactions, inventory, demand) and enrichment data (contextual, strategic);
Foundations: both Segmentation (see Clustering) and Response Functions (see Regression, Decision Trees) are key to any Pricing Optimization initiative;
Target & Constraints refer to the expression of the tactical decisions we want to optimize: Pricing Optimization goes beyond price points, and spans from dynamic pricing, interaction effects between price points and products (mix), Discounts and Markdown logics, Promotions efficiency, to pure Optimization vs Targets and Constraints.
Figure 1: The Pricing Optimization framework
So, here goes our Top 5 Algorithms for Pricing:
Clustering: Clustering, K-Means are central mainly because segments (both customer and products) are at the very heart of pricing performance and need for granularity.
Regression: OLS, Multivariate, Logistic will allow you to identify predictors with or without interaction. Must have for elasticity and cross-elasticity measurement, and finding the demand curves behind your customers’ behaviours.
Association Rules: even if the two first are obvious, this one is often missing. A priori rules are at the centre of market basket analysis for example. Very useful when you have a large product portfolio and need to identify product sharing consumption patterns. It can allow you to extrapolate accurately value related insights from market research to your transactional data.
Decision Tree: Pricing is all about rules. Decision Trees (CART, CHAID, etc.) allow a representation of pricing decisions relating antecedents and outcomes, as well as a definition of the rules behind them (if / then). The backbone of pricing execution and efficient promotions and discounting for example.
Optimization: coefficients and predictions are great but Pricing Prescriptions command to add an optimization layer to the modelling approach in order to maximize (minimize, etc.) a function (gross margin) under constraints (units, etc.). Mathematical Programming (Linear, Dynamic, etc.), Stochastic Optimization, among others can be good candidates, that will need to be completed with advanced testing protocols (A/B, Pre/Post/, Bandits, etc.).
Algorithm and mathematics are at the very heart of the scientific process. And Pricing needs to be treated as a science. If you want to be prepared for the algorithmic era, you’d better start now!
To find out more about Algorithmic Pricing, please send an email to email@example.com