A.I. Policies in Price Setting Model Design (Upfront Work)

Our price optimization engines are built using advanced statistical models that leverage A.I. (this has been true since our firm’s 2014 foundation). Our A.I. policies governing the underlying model designs are anchored in the concepts of transparency, explainability, and human oversight:

We only use explainable A.I. models to optimize price points. For example, we do not use neural networks, where the pricing logic is a “black box” to all involved, including analysts.  This means that price point calculations can be understood and explained by analysts.

Our clients have access to all details about their A.I. models. In fact, we often invest time in knowledge transfer, to ensure that our clients understand how the models operate, and how pricing recommendations are generated. This enables clients to conduct better informed reviews of the pricing recommendations before implementation. This approach helps ensure that resulting recommendations make sense for the business, that stakeholder input can be considered, and that any necessary model adjustments can be made in a responsive manner.

Clients with adequate capabilities can take ownership of the A.I. models. While initially designing the pricing algorithms tends to be a complex undertaking that requires significant sophistication and more specialized tools, once the optimization models are built, existing or low-cost solutions frequently suffice to keep them updated (some incremental technology investments could be required).

Note about Dynamic Pricing and Machine Learning: Some clients desire a dynamic pricing engine. We define dynamic pricing as a pricing system where the pricing model itself (underlying segmentation scheme, pricing factors such as markups or discounts) are continuously updated with relatively high frequency (often on a weekly basis), resulting in price changes that may not be directly related to core data such as cost fluctuations or changes in competitive price points. This is in contrast with periodically updated pricing logic where absent changes in core inputs (such as changes in cost, customer, or competitive pricing data), pricing recommendations remain relatively stable between model updates, as segmentation schemes and pricing factors are updated with less frequency (perhaps a few times a year, when significant new data becomes available, market conditions change, or business strategies shift). 

The main benefit of a dynamic pricing system is the quickness and agility in reacting to market changes. Some drawbacks include increased model complexity and resource requirements, along with less pricing stability in general. For example, dynamic pricing engines can drive customer-specific prices to increase, despite no changes in costs or other core pricing inputs (and such price changes may be best implemented with due care and monitoring).

To manage dynamic pricing systems, Machine Learning algorithms may be leveraged to continually keep pricing models up-to-date. If M.L is used, clients are advised to ensure that a human remains in the loop. M.L. generated model logic updates (such as changes to segmentation schemes, for example) should be reviewed by a human prior to being moved to a production environment. The human should have knowledge about the business, as well as about the model’s operation. This approach helps ensure that model updates are understood, align with business strategies, and A.I. “mistakes” do not disrupt operations.

A.I. Policies Related to Pricing Model Updates (Ongoing Pricing Model Operation)

Once pricing is live, prices need to be updated regularly. Price changes may be due to fresh inputs (cost data, customer data, transactional data, competitive/market data, etc.), or refinements in the model logic (updates to segmentation schemes or pricing factors such as markups/margins or discounts).  

Model input data updates are typically managed by clients. To ensure that best available current data is leveraged, data update processes may be supported by intelligent automation, including A.I. agents. Such processes are best designed by clients who are knowledgeable about their source data and systems. We are available to support efforts to create such processes if clients seek such assistance, and we recommend that clients follow the core A.I. policies outlined above related to transparency, explainability, and human oversight.

Model logic updates are also typically managed by clients. Upon request, we remain available to perform model updates, or to train new client personnel on how to perform them. This approach helps substantially minimize long term costs to our clients (many other pricing providers charge hefty recurring subscription fees for model update services). As noted above, M.L. tools may be leveraged to continually refine model logic in instances where dynamic pricing is desired. Updating more stable (less dynamic) pricing models typically requires less advanced sophistication levels in terms of statistics and A.I. use.

A.I. Policies Related to Supporting Pricing Processes

In addition to model sophistication and fit, a pricing engine’s effectiveness in driving performance is also impacted by the quality of inputs, and the processes aimed to ensure that the pricing guidance is utilized in the context of real life transactions. Clients are encouraged to leverage A.I. tools in both of these areas. We are available to support relevant efforts upon request, and we recommend that clients follow the A.I. policies outlined above.

For example, in the area of producing quality inputs, A.I. tools can be leveraged to collect competitive data, or to clean transactional data.  In the execution area, A.I. tools can be used to help streamline and guide approval processes for discounting requests, or to suggest value communication strategies that fit customer-specific selling situations involving particular types of products/offerings. Given the significant variation in the dynamics across different pricing environments, and the quick pace of change in relevant A.I. technologies, the number and nature of possible application areas is extensive and increasing.