This section proposes an idea that can eventually help to get a better grasp of a problem. Currently, it is possible to describe the whole problem regardless of these three parts.
The computation graph of a pricing optimization problem can be seen as threefold:
- "Pre-market" is the pricing structure, which includes all the kinds of prices/discounts/misc. that lead to the computation of the price actually paid by customers. In particular, it contains the values we want to optimize and, potentially, business constraints such as price alignments.
- "Market" transforms these invoiced prices into volumes. Depending on the available data it can be volumes straight from the history, or trained ML models with elasticity and WTPs, or anything in between.
- "Post-market" are computations and variables whose whole purpose is to support the various criteria on margins, margin rates, turnover/revenue, or other similar metrics.
For now, this is only an intellectual consideration. Currently, you need to describe the whole problem regardless of these three parts. Moving forward on a "market model toolbox" and automatically inferring the computation graph for the criteria are two paths that could significantly improve the efficiency of OE deployment.
The figure below attempts to synthesize these three aspects of a typical pricing optimization problem.
