Price Distribution Elicitation from an LLM

Author

Pavel Logačev

Published

June 20, 2025

Summary

Demonstrates a two-stage pipeline that uses large language models (LLMs) to elicit reasonable price distributions for consumer products when competitor pricing data is insufficient or historical sales data is sparse. It extracts structured product attributes from free-text titles and descriptions (via GPT-3.5-turbo), then uses GPT-4o to elicit a retail price range (minimum, typical, maximum) based on those attributes and regional context. The elicited price ranges are used to reverse-engineer log-normal distribution parameters, which are validated against actual prices scraped from Mercado Libre. The result is a calibrated, probabilistic estimate of product pricing grounded in LLM-elicited knowledge.

1. Implementation with ChatGPT

---
title: "Price Distribution Elicitation from an LLM"
author:
  name: "Pavel Logačev"
  email: "pavel@example.com"
date: today
format:
  html:
    toc: true
    code-fold: true
    code-tools: true
    df-print: paged
engine: knitr
editor: visual
---

## Summary

Demonstrates a two-stage pipeline that uses large language models (LLMs) to elicit reasonable price distributions for consumer products when competitor pricing data is insufficient or historical sales data is sparse. It extracts structured product attributes from free-text titles and descriptions (via GPT-3.5-turbo), then uses GPT-4o to elicit a retail price range (minimum, typical, maximum) based on those attributes and regional context. The elicited price ranges are used to reverse-engineer log-normal distribution parameters, which are validated against actual prices scraped from Mercado Libre. The result is a calibrated, probabilistic estimate of product pricing grounded in LLM-elicited knowledge.

## 1. Implementation with ChatGPT

\[[Notebook](./price_distribution_elicitation.html)\]