Picture an AI agent with a simple instruction: find the best supplier for this component and start the conversation.
It is fast. It is tireless. It reads every record you have.
Including the three versions of the same supplier, set up by three different plants. The price list that expired in March. The part number that means a bracket in one system and a gasket in another. The bill of materials from two engineering changes ago.
The agent does not know any of that is wrong. It simply acts on the data it has at speed and at scale.
AI does not clean up your procurement data. It amplifies it.
In plain terms: imagine giving a brilliant new assistant a filing cabinet where half the folders are mislabeled. They will work twice as fast as anyone before them and file everything in the wrong place twice as fast too.
In 2024, a Canadian tribunal had to decide who was responsible when a company’s own chatbot gave a customer the wrong answer. The company’s defense did not go well. We will get to that. First, why is this suddenly urgent?
Agentic AI is moving from slide decks into software roadmaps. In August 2025, Gartner predicted that 40% of enterprise applications would be integrated with task-specific AI agents by the end of 2026, up from less than 5% at the time.
Procurement is not exempt. But Gartner has also been direct about readiness. In August 2026, Gartner analyst Lynne Phelan wrote that Gartner predicts that through 2027, only 20% of procurement organizations will have sufficient data and process maturity to use multiagent systems.
Separately, Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027, due to escalating costs, unclear business value, or inadequate risk controls.
Put those together and a pattern emerges. The ambition is arriving faster than foundations.
So what actually goes wrong when an agent acts on data that is not ready?
Garbage In, Decisions Out
When a report runs on bad data, you get the wrong chart. Someone notices, questions it, and fixes the source. When an agent runs on bad data, you get a wrong action. Here is where procurement data usually trips it up:
- Duplicate suppliers. The same company under three names splits your spend, hides your leverage, and makes one dependency look like three.
- Stale prices. Last year’s quote becomes this year’s benchmark, and the agent negotiates against the wrong number.
- Inconsistent parts and categories. If the same item is classified three ways, the agent cannot see total demand, or worse, compares items that are not alike.
- Obsolete BoM versions. Sourcing a part that engineering removed two revisions ago is fast, efficient, and completely pointless.
For an AI agent, data quality is not only about whether a record is correct. It is whether the record is current, consistent, and authoritative.
None of these are AI problems. They are data problems that AI makes faster and harder to spot.
And when something does go wrong, every organization needs an answer to one question: who is responsible?
When the Bot Gets It Wrong, You Own It
In November 2022, a customer asked Air Canada’s website chatbot about bereavement fares. The chatbot told him he could submit a purchased ticket for the reduced bereavement rate within 90 days of the ticket being issued. Air Canada’s actual policy, published elsewhere on its own website, required the request to be made before travel.
When the airline refused the refund, the dispute went to British Columbia’s Civil Resolution Tribunal. Air Canada argued, in effect, that the chatbot was responsible for its own actions. The tribunal rejected that argument and held the airline responsible for the misleading information provided by its chatbot. In February 2024, it ordered Air Canada to pay the customer CA$812.02 in damages, reflecting the difference between the fare paid and the bereavement fare.
The amount was small. The principle was not. Two sources of truth inside one company disagreed, and the automated one was the one talking to the customer.
Now move that gap into procurement. An agent negotiates using terms from a superseded contract or recommends a sourcing commitment based on a price that expired months ago. The procurement lesson is harder to escape; when an agent acts on your data, the organization still owns the consequences.
So where should procurement start?
Fix the Foundation Before You Hire the Agent
- Consolidate supplier identities. One supplier, one record, with every site and alias linked to it.
- Normalize and classify spend. Consistent categories let any system, human or AI, see true demand and true leverage.
- Connect sourcing to engineering data. Pull BoM and part data from ERP and PLM, so decisions run on the current version, not a copy.
- Decide what the agent may do, and who checks. Autonomy should grow with data quality, not ahead of it.
Now, the test. Pick one of your top ten suppliers. How many supplier records does it have across your systems? If the answer is more than one, the next section is for you.
Where MeRLIN Fits
MeRLIN’s spend analysis collects, cleanses, and enriches procurement data, using integrated AI and fuzzy logic to categorize and normalize spend so reports reflect reliable insight. Supplier 360 consolidates financial, performance, risk, and compliance data into a single supplier profile, helping teams avoid duplication and rationalize the supplier base.
MeRLIN also integrates bill of materials data at the part and component level with supplier quotes, contracts, and category spend, and its direct sourcing supports PLM and ERP integration, so sourcing can work from the same engineering and enterprise data as the rest of the business.
That is not an AI agent. It is the foundation every AI agent will need.
Speed is only an advantage when you are pointed in the right direction. Clean data is what points you there.
