By Cesare Rotundo, VP of Product Management, High Tech
AI for B2B Pricing: The Last Decade’s Approach
A tough problem, but the good data scientists that approached this issue in 2010 found a compromise: what if we could flatten some of the dimensions, and focus only on those that make a difference based on the data? Let AI crunch for us what those dimension groupings are. Let’s create groups, or to be more precise, segments that are created from the data: data is what drives segmentation. If the data tells us that all deals belonging to Segment 108 (EMEA, Italy, Pharma, Growth, Product Family X, Customer Size = B) behave similarly in terms of price sensitivity, and if we can determine the optimal price for that segment, then we can apply to the next deal for Segment 108!
The approach has some merits, and it looks promising. We are giving up on traversing the tree of every possible combination of dimensions, instead only focusing on some of the leaves in that tree, of which 108 is one example. As a new deal comes in, it will be placed in the proximity of one of these leaves and will inherit the same behavior, including the optimized price. Does this “gray box” work? Many times, it doesn’t. Segmentation attempts to resolve the data scarcity issue, but its success is partial and temporary. It may have worked for past data, but new data may fall into new segments. Also, some of the data-generated segments may just be “wrong”. We can then manually adjust those segments, drop some, rearrange others, refresh the tree. This improves the model and increases accuracy, but it’s not getting us to a place where we can really trust the box outcome. Ten years later we’re still trying to use this hammer to nail the right price. But is that really a nail?
The 2021 ML Revolution
Meanwhile, AI/ML has undergone an incredible evolution. Not only the power but also the variety of tools, entire new platforms, now accessible via Cloud, make AI/ML one of the most competitive markets where juggernauts like AWS and Google Cloud innovate at a blistering pace.
The issue with segmentation is that it’s trying to simplify the complex problem of multi-dimensionality a priori, therefore losing the ability to dynamically finding the best data to predict the outcome at decision time. Also, it’s reducing the question to a single number rather than to a range with an associated probability. We live in an uncertain world, and while we can’t measure everything, we can make decisions that will increase the likelihood of success using the AI/ML tools that we have in 2021.
It’s not all about the newfangled AI/ML 2021 tools. It’s about targeting the right business issues and ultimately building AI/ML models that can solve those business issues based on the available data. Even today, if you put garbage in you get garbage out. The richness of B2B business transactions is what today’s AI/ML tools can harvest better than the 2010 approaches.
The Model N Objective
Defining a successful model requires the cooperation of business and technology experts and, last but not least, access to data. Model N is in the enviable position of having access to lots of data, not only in quantity but also in scope. Targeting what we call Revenue Management means that Model N customers transact data from list price thru pricing business rules, contracts, quotes, rebates, orders, opportunities and registrations, POS. Our customers can relate POS back to contracts and quotes within the same system. Model N doesn’t need to reach out for most data, because it lives within Model N. This offers Model N customers an enormous advantage because they can cross-functional silos when making margin decisions, and this makes Model N the perfect breeding ground for AI/ML pricing optimization. The same dimensions are used throughout, and data is naturally correlated. The data preparation activity, a big portion of an AI/ML project, is greatly simplified.
We targeted business issues that were most valuable for our customers: not only the sought after “optimal price”, but also the Win probability. Is it worth spending time on a quote approval workflow when the win probability is 7%? Should the resell price and/or the distributor margin be increased, or should you focus your attention somewhere else? Another important B2B measure is the expected compliance: what percent of the quantity the distributor is promising will they ultimately ship?
In our third and final post, we’ll take a look at tomorrow’s recipe for AI/ML success and how it helps companies make better and faster decisions. If you’d like to learn more about Model N’s price optimization approach, go here.
