Negotiation with Price-dependent Probability Models

Antonio Bella, César Ferri, José Hernández-Orallo, María José Ramírez-QuintanaWAT 2009 · OpenReview

Abstract Traditional Customer Relationship Management (CRM) systems typically operate on a "take-it-or-leave-it" basis, ranking customers by their likelihood to purchase at a fixed price. This paper proposes a shift toward active negotiation by modeling purchase probability as a continuous function of price. By deriving Expected Profit Curves, the authors identify optimal bidding strategies for negotiable goods. The framework introduces the Best Local Expected Profit (BLEP) strategy and a method for iterative model updating based on customer feedback, enabling automated bargaining in complex multi-party and multi-product environments.

Introduction

In many high-value markets—such as real estate, automotive sales, or specialized professional services—prices are rarely fixed. Instead, they are the result of a negotiation process. Traditional CRM systems struggle in these environments because they treat the "buy/no-buy" decision as a binary outcome at a single price point.

Traditional fixed-price CRM model where probability drops to zero at a specific threshold.

The core challenge addressed here is that a customer’s willingness to pay is not a single value but a probability distribution. Asking for too high a price risks losing the sale entirely, while asking for too little results in missed revenue. This paper introduces a systematic way to automate this bargaining process by treating price as a continuous variable in the profit equation.

Methodology: Profit Probability Curves

The authors propose modeling the probability of a sale as a distribution—typically normal or skew-normal—centered around an estimated "fair price" for a specific customer. This allows for the calculation of an Expected Profit Curve.

The expected profit is defined as the product of the price and the probability of acceptance at that price: $E[Profit] = Price \times P(buy|price)$. This calculation identifies a "sweet spot" where the seller can maximize their return while maintaining a realistic chance of closing the deal.

Negotiable price model showing a smooth probability distribution and the corresponding expected profit curve with a clear maximum.

The BLEP Strategy

To navigate the negotiation, the authors introduce the Best Local Expected Profit (BLEP) strategy. This is an iterative approach where the seller offers the price that maximizes the current expected profit.

Model Updating and Normalization

A critical component of the framework is how it handles rejection. If a customer rejects an offer at price $X$, it provides valuable information: their maximum reservation price is certainly below $X$. The algorithm then "clips" the probability distribution at $X$ and re-normalizes it. This iterative refinement allows the seller to narrow down the buyer's price range in subsequent rounds, effectively learning from every "no."

Theoretical Framework and Insights

The shift from fixed to negotiable pricing changes the fundamental nature of customer leads. In traditional CRM, rankings are static. In a negotiable framework, rankings are dynamic and price-dependent. A customer who is a poor lead at a high price might become the most profitable prospect if the price is adjusted, as their probability of acceptance may scale more favorably than others.

Furthermore, when dealing with multiple potential buyers for a single product, the seller can visualize an "envelope curve"—the upper boundary of all individual expected profit curves combined. This envelope guides the seller not only on what price to bid but also on which customer to prioritize in the next round of negotiation.

Results and Discussion

The research demonstrates that a simple BLEP strategy can sometimes be overly conservative. By pre-calculating potential bid points and ordering them from highest to lowest, sellers can often achieve higher final transaction prices. The paper successfully extends this logic to complex scenarios, such as one product with many buyers (Case 6) or one buyer with many products (Case 7).

Limitations and Future Directions

While the framework provides a robust foundation, the authors note several areas for improvement:

  • The "Final Boss" Scenario: Managing multiple negotiable products across multiple customers simultaneously (Case 8) remains a high-complexity problem that may require evolutionary computation or advanced simulation rather than simple analytical curves.
  • Model Sensitivity: The strategy's success is highly dependent on the accuracy of the initial machine learning model used to estimate the "fair price." If the baseline estimate is flawed, the negotiation strategy will likely fail.
  • Distribution Assumptions: Real-world behavior, such as "bargain hunting," may not follow a symmetric normal distribution. Future work will explore non-symmetric distributions to better capture these psychological nuances.

Summary updated

BibTeX
@inproceedings{bella2009negotiation,
  author = {Antonio Bella and César Ferri and José Hernández-Orallo and María José Ramírez-Quintana},
  title = {{Negotiation with Price-dependent Probability Models}},
  booktitle = {WAT},
  year = {2009},
  url = {https://openreview.net/forum?id=oRoJHO8EbX},
}
Figure