Bayesian Inference for Guided Selling: SAP Predictive Summit Talk
Guided Selling is the answer to the following paradigm: “I want, as a customer, to go in that store because the salesperson understands exactly what I need, and I’ll get the best directions for my purchase”. It addresses the problem of reducing complexity and increasing decision confidence when it comes with complicated products. Why should that be important? I’m one that loves to be treated with fairness and to get useful advice, and I guess you belong to this club, too. Think of last time you bought a camera. You probably were overwhelmed by a myriad of options, but the salesperson pointed you to the most suitable product once s/he knew your motivation. Is it for shooting wildlife in the Savanna? Or for recording memories of your children? Look here, that’s the camera you need.
The salesperson’s work is to get the customer to pay for that item. After all, the seller is trained with a long track record of deals on different customer archetypes. We can hardly imagine how the selling process works backstage and how we can make a program that helps salespeople. Is it possible to automate it?
The Context Profiler is a humble and limited attempt to resolve the information gathering task. It is a service based on Bayesian Inference and it is running in the proof of concept Zefiro.
- It is extensible because it facilitates the development of pluggable user interactions.
- It interfaces with the merchant catalog and it is intended to make sense of marketing and selling analytics that might say something on user behaviour, expectations and needs.
- Ultimately it provides to the selling agent some precious indicators that have a heavy influence on how to deal with customer.
The Power of Networks
When we are looking into any process that we are willing to emulate and automate, we will clash soon on how to represent the domain concepts and how to make them runnable in a software program. Nothing more complicated than the human mind with its incredible capability to tangle beliefs, desires and intentions!
On pursuing that objective, one of the many studies for modelling human beliefs and desires as a causal relation caught my attention, and for this project, the Bayesian networks lay the foundations for the context profiler engine. What does it mean in practice? Our ideas, opinions and needs are deeply affecting our everyday choices, including purchase selections. Concepts that appear to be unrelated, they might actually have an impact on our decision to buy a product, or instead, not to buy.
While the main goal of the Context Profile is to get all possible information, that does not mean we should prompt to the user yet another 10mins boring survey nobody wants to complete. Guided selling walks the customer through a series of interactions for grabbing anything might be useful for deducting how the profile of the customer is. I want to diversify the user experience in order to spark surprise and make the user happy to tell us everything. This is why the system allows the diversification of user interactions in an easy way.
The Trivial Pursuit of Selling
On keeping the user experience flexible, the core of the system is driven by one single goal, to know more about any single customer. Game theory has been a source of inspiration for enhancing that functionality of the system. Game theory is a branch of economics dedicated to selecting the best strategy according to the opponent’s best strategy. It helps to predict what others do and how we should act accordingly. But in this case, is there really an opponent? The application is constantly fighting against a fierce adversary, ignorance. We have to maximize the amount of information and at the same time minimize the chances of ignorance to prevent it. Want to discover more about that? Here you can find a video that explains the principle behind that
Giancarlo Frison is Technology Strategist at SAP Customer Experience Labs