P2Predict is a parametric price prediction tool designed to turn procurement history into a conversational price model. It was developed by Ahmed K. Hafsi, based on his procurement negotiation experience at Infineon, Dyson, and TWS Partners. At its core, the tool learns the factors that drive price in a user's historical purchasing data (such as supplier, material, dimensions, specifications, region) and provides a defensible target price for any new or proposed part.
Key capabilities include:
- Learning from actual paid prices to predict what similar parts should cost.
- Attributing the predicted price to each specification and supplier, showing the factors behind price changes.
- Providing a calibrated likely range for each estimate and flagging areas where the data is too sparse to be reliable.
- Improving as more procurement history is entered.
How to use it: P2Predict interacts with any AI agent (including Claude, GPT, or local models) through standard connectors (such as MCP). Users ask questions in natural language through their existing AI assistant, and the tool runs the analysis on the local machine. Installation is done with a single pip command; specific steps are in INSTALL.md, and technical details are in TECHNICAL.md. It has no dashboard or standalone application; the interface is the agent the user already has.
Use cases:
- Procurement and engineering teams in industries such as automotive, semiconductors, electronics, industrial, pharmaceuticals, and chemicals, for price benchmarking and negotiation preparation.
- Specific applications include: assessing whether a quote is fair, breaking down price components, simulating the impact of supplier switching, pricing specification changes during design reviews, identifying unnecessary design cost in cost reduction workshops, and RFQ triage (quickly identifying the few items that need negotiation).
- Public case studies include battery management ICs, used cars, and aerospace fasteners, each showing results, confidence, and data sources.
Limitations:
- P2Predict is not a should-cost tool; it cannot build up part costs from raw materials, labor, and machine time, nor can it tell you a supplier's true costs or profits.
- It predicts only based on what is shown in the user's data. For parts that are not similar to historical data, it will widen the range or recommend obtaining quotes rather than guessing.
- The specification breakdown shows factors in the data that are related to price, not causal or engineering models.
- It does not fabricate data: without relevant history, there is no model.
Other facts:
- The tool runs entirely on the user's machine, data is not uploaded, and there is no third-party training.
- Every estimate comes with a confidence interval; in public case studies, about half of predictions fall within 16% of the actual price, and ninety percent fall within 73%.
- Free to use, including internal use by for-profit companies; a commercial license is only required for serving third-party clients or embedding it in a paid product.
- Open source, with readable and adjustable source code, making it easy to integrate into a company's ecosystem.
- P2CLPFD is a companion tool for deciding who gets the volume, based on awarding to the lowest total cost of ownership, taking into account capacity, minimum order quantities, share caps, dual sourcing, and volume discount rules.
