BTW EP 27: Data or Delusion? Procurement's Future Runs on Truth

Mar 18, 2026

Jason Busch, founder of Spend Matters and builder of AI "co-workers" for procurement, explains why partial data is dangerous as procurement adopts autonomous AI. He contrasts copilots, agents, and multi-agent co-workers. He stresses broader data needs—financial health, inventory, tariffs, market signals—and warns supplier-shaped data and cultural bias can institutionalize harmful decisioning.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

AI Co-workers Are Whole-System Agents

AI co-workers are multi-agent systems that combine retrieval, model-context protocols, and governance to execute complex procurement work autonomously.
Jason Busch contrasts copilots (guidance) and agents (basic tasks) with co-workers that find work, orchestrate data, and act like digital employees.

INSIGHT

Context Beats Price For AI Decisioning

Contextual data matters far beyond negotiated price: supplier financial health, inventory, tariffs, competitor pricing, and contract performance must feed AI decisioning.
Busch says model-context protocols and RAG plumbing let systems assess source trustworthiness and clean data on the fly.

ADVICE

Don't Expect AI To Auto-Clean ERP Messes

Don’t assume RAG or AI will magically fix dirty ERPs; pre-define which sources are 'truth' and embed that into MCPs and reasoning models.
Busch cautions organizations must over-invest upfront in data prep and context rather than hope AI will clean everything later.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

Procurement doesn't have a data problem. It has a data delusion.

For 25 years, the function has told itself the same story: if we can just clean up our spend, we'll finally be in control. And yet here we are… swimming in the same dashboards, drowning in fields, and still struggling to answer a simple question: what do we spend?

In this episode of Buy: The Way…To Purposeful Procurement, Jason Busch, founder of Spend Matters and now a self-described builder of AI "co-workers," returns to the podcast to pressure-test BuyLaw #5: "prioritize comprehensive, high-quality data."

If procurement wants to operate in a world of AI employees, continuous validation, and P&L accountability, their data cannot remain partial, fragmented, or shaped by suppliers.

Jason draws a sharp distinction between the roles or entities that manage procurement data: copilots, agents, and what he calls digital co-workers (multi-agent infrastructures capable of executing complex work autonomously). But all that capability comes with a catch. When the marginal cost of activity drops toward zero, the absolute risk of bad data increases exponentially.

Humans have the battle scars and the intuition to know when something isn't quite right with the data. AI doesn't, unless we explicitly teach it what 'right' looks like. That's where procurement's comfort with incomplete data becomes dangerous.

For decades, the function has relied on narrow slices of information: negotiated price, historical spend, maybe a market index or two, but in an AI-enabled world, that's insufficient. Jason explains why context means everything – supplier financial health, commodity forecasts, tariffs, inventory signals, competitive pricing, risk data, contract performance signals, governance structures, and the cultural guardrails that determine how decisions are made.

If procurement feeds incomplete, biased, or poorly governed data into increasingly autonomous systems, those systems won't just make mistakes faster; they'll actually end up institutionalizing them and making procurement's data problem unnecessarily worse.

Jason's advice for procurement is pragmatic and urgent: set up a data governance committee tomorrow. Not to tidy historical spend, but to define what data matters, which sources are trustworthy, what tolerances exist for error, and at what point autonomous systems are allowed to act on that data.

In a world of digital co-workers, incomplete data isn't a nuisance. It's a real, human liability.

Links: