Sign up for our newsletter and get the latest big data news and analysis.

Using AI to Optimize Pricing

According to a recent KPMG study, nearly all of the executives they interviewed expect AI to create new winners and losers, but only 17% had deployed AI at scale. Another study in 2019 indicated that 91.7% of executives are accelerating their investments in AI, fueled primarily by fear of disruption, and many do not expect return for years. Research has shown that companies are willing to pay more for AI and machine learning enabled software. This can lead to hype from software vendors and when you peel back the onion you find it’s either not very sophisticated, practical or valuable in the real world.

What we want to do in this article is to remove the mystery around AI based price optimization and explain the value, challenges and techniques and how it is being used today to drive better prices and make better decisions and improve company revenue and profitability.

The Value of Price Optimization

Numerous studies have shown that investing in the people, process and systems to optimize prices can yield between .5% to over 4% return on sales.  By using AI, you can do “more with less” and make faster and better quality decisions using the “Enterprise Big Data” that you have around customers, products, purchase behavior, buyer journeys. These gains are achievable now using proven AI technology that has been deployed across many industries and companies and the use of AI enables a higher ROI than solutions deployed without them.

The Challenges with AI Driven Price Optimization

There are numerous challenges that must be overcome to optimize pricing effectively. Data is often cited, but usually not the biggest challenge. The larger issue is digital transformation required to embed optimized AI pricing guidance into everyday business processes. Then delivering this guidance in a way that makes sense to the sales organization and customers, otherwise change management can be a big obstacle to success.

Keys to Success in AI Driven Price Optimization

Controls over the algorithms and transparency of decisions being made are two of the key components to implementing successful AI based technology, as is change management in getting the buy-in from the organization on how to leverage the AI most effectively.

Different Techniques for Price Optimization

1. Regression Modeling

This is the standard approach to identified trends and correlation between data and to create predictive models. Most of ML approaches are de facto regression models (Regression Trees, NN etc.). The goal of such models is to predict a value from a set of other parameters (numerical or not). From a price optimization perspective, it is very useful to predict different values such as volume variation or competitors’ prices in reaction to your own price modification. This helps positioning your prices at the most profitable point.

However, Machine Learning has a principle drawback because it requires qualitative and quantitative data to provide reliable predictions. Providing this, creating valuable predictive models, specifically in a B2B pricing context is often far from easy if not impossible, due to the lack of data on specific segments. But it can be relevant to give insights at a global level, for key customer segments or runner product categories.

2. Elasticity Approach

The ability to model price elasticity is the holy grail of price optimization. With such a model it becomes obvious to provide for each product and customers segments optimal prices recommendations either to maximize revenue or profitability (or something in-between). Elasticity approach relies on regression models it still has the same limitations and is quite difficult to obtain especially when it comes to B2B due to the complexity of business relationships.

3. Machine Learning

Machine Learning can also be used for classification or clustering. It is a powerful means to automate product and customers segmentation. It allows to define and maintain homogeneous segments regarding the behavior of customers based on transactions and master data. Thanks to this segmentation one can define optimal price corridors (floor, targets and stretch) that can be easily explained and shared in the sales organization. ML models can also be trained to provide products recommendations to improve up and cross selling revenues.

4. AMAS – A Novel Approach that Mimics the way Organizations Work

AMAS stands for Adaptive Multi-Agents System, it comes from an academic research field called Distributed Artificial Intelligence, it is a bio-inspired paradigm devoted to solve complex optimization problems. The main principle on which it relies is the emergence properties of self-organizing systems (e.g. like ants’ colonies, birds flocking, etc.). In other terms, the whole organization provides solutions to problems that are way more difficult than what a single of its element can.

While other approaches to price optimization focuses on the “what” (which price or margin to propose for this segment), AMAS focuses on the “how” (what are the best sales conditions to reach my objectives). This is often a neglected and far more complex task to implement a final price recommendation into an organization that possesses thousands of list prices, hundreds of special discounts and agreements as well as intricated business rules to respect. Finding the optimal configuration in such a context is simply impossible with standard pricing software, but it is where AMAS excels.

It models the pricing infrastructure (Prices, discounts, rebates, costs, etc.) as small pieces of code (agents) that interact and cooperate to solve the problem they are facing — reaching a strategic objective while respecting a whole lot of business rules.

By solving this gigantic Rubik’s cube, it makes recommendations to human business experts in the form of prices, discounts, or rebate values to apply. However, they always have the final choice to keep, modify or reject these recommendations.

One other interesting property is that it is highly configurable: objectives and constraints that drives the optimization can be easily specified by the final user. Thus, it helps him to fine-tune its strategy by providing results and business impacts for each of his settings.

AI systems are often viewed as black boxes providing recommendations without helping final users to understand why. As it is a model of how the pricing works within a specific organization, AMAS recommendations can easily be interpreted by business users. Moreover, AMAS also provides insights about the optimization process itself allowing to understand why an objective is attained or not and if not, what is preventing it to doing so. In other words, this is much more of a white box.

Conclusion

There are numerous AI and Machine Learning techniques available for segmentation and optimization. The right choice depends on the situation: data available, industry and routes to market. We are excited about the application of AMAS as a newly emerging yet proven technique to optimize prices and margins and increase the profitability which works the way organization do.  These techniques have been applied successfully at hundreds of companies and driven significant gains to the top and bottom line.

About the Authors

Sylvain Rougemaille Ph.D. is Price Optimizer Product Manager at Pricefx. He was co-founder and Chief Product Officer at Brennus Analytics. Pricefx is an industry-leading price optimization, management and realization software provider.

Gabriel Smith is the Chief Evangelist and Vice President of Innovation at Pricefx and has 19 years of experience in Quote to Cash, CPQ, Pricing, Promotions, Consulting, Product Management, Sales and General Management.

Sign up for the free insideBIGDATA newsletter.

Leave a Comment

*

Resource Links: