Skip to content
RevSeekr

AI & Optimization Decisioning layer By Nicolás Sierra Rojas

The Data Nobody Has, and the One That Blocks Every Pricing ML Project

A model cannot learn from lost quotes if the organization never recorded them.

Read the original on LinkedIn


Pricing machine-learning projects often begin with algorithms, features and data pipelines. The more basic dependency is quote-outcome history: what was offered, whether the customer accepted it, and why the deal was lost when they did not.

Transaction history shows only the wins

ERP systems preserve orders and invoices because completed transactions create records. Lost quotes frequently disappear into email, free text or a salesperson’s memory. A model trained only on completed business therefore sees accepted prices but cannot learn where price contributed to a loss or where the company could have captured more value.

Instrument the outcome before training the model

A useful quote record connects the offered price with a controlled outcome such as won, lost or no decision. Capturing that consistently is a commercial-process change, not simply an integration task. Sales behavior, workflow design and ownership must be settled before a credible ML delivery date can be promised.

Create value while the evidence accumulates

Teams do not need to wait for a model. Pricing floors, targets, stretch guidance and approval guardrails can improve decisions immediately. When the same workflow records every quote and its result, today’s rules also generate the evidence needed for tomorrow’s recommendation model.

The roadmap starts with one question: how many recent quotes have a structured outcome, and where does that information live?

Sources

This on-site edition is a concise summary of the original article published on LinkedIn by Nicolás Sierra Rojas on March 7, 2026.

How we talk about results →

Written by Nicolás Sierra Rojas. Nicolás is a data scientist and economist at RevSeekr working at the intersection of commercial data, pricing systems and machine learning. If you are working through any of this, get in touch.

From argument to evidence

If this essay describes your program, find out how far it goes.

The essays are the argument. The diagnostics are where it meets your data: twelve minutes, no login, a scored report in your inbox — or thirty minutes with the person who would run Discover.