Published on Procurement Insights | Jon Hansen
For The Busy Executive
What is the difference between “clean” data and “blind” data? If you don’t know, then read on before you start (or finish) your next ProcureTech initiative.
Every data governance initiative starts with the same assumption: that the data being cleaned, tracked, and managed is the right data to begin with.
That assumption is almost never examined.
And in my experience, it is the assumption that most often explains why governance programs deliver technically clean data that still produces the wrong outcomes.
Two cases from the archive illustrate this more clearly than any framework I’ve seen.
Case One: The Department of National Defence
When I came in, the MRO procurement platform was delivering 51% next-day performance against a 90% contract requirement. The client’s instinct was immediate: automate the system. Clean up the process. Get the data moving faster.
Before touching anything, I asked one question: what time of day do orders come in?
The answer was 4pm.
That was not a data problem. That was a behavioral signal. The service technicians were incentivized to complete as many service calls per day as possible. Ordering parts after each call — as policy required — was cumbersome and cut into their daily numbers. So they sandbagged. They held all their orders until end of day, submitted them in a batch, and the next-day window closed before the procurement team could respond.
The data being tracked — order volume, delivery rates, parts availability — was perfectly accurate. Every number was real. The governance was sound. But it was capturing the output of an incentive misalignment that no one had identified as the root cause, because no one had asked whether the data reflected actual procurement behavior or the downstream artifact of a structural problem.
Automating that system would have produced faster, cleaner, more efficiently governed data about the wrong thing.
The fix was not data governance. It was an incentive realignment, an agent-based model that correctly identified the technicians as impacting agents, and a supplier, UPS, and customs integration that extended the next-day window. Delivery performance went from 51% to 97.3% in three months. It held for seven years.
Clean data never would have gotten there.
Case Two: The PC Retailer
Shortly after, I was approached by one of the largest PC retailers in the United States. They had executed a vendor rationalization strategy — compressing hundreds of suppliers down to 100, with sound logic: better volume leverage, lower administrative overhead, more manageable contracts.
Two years in, their data showed savings. Year one, strong. Year two, still positive but declining. They wanted an assessment.
I looked at it and said: you do realize that when you compressed your supplier base down to 100, that became your source of truth?
They had lost sight of the rest of the market. Their data governance was excellent — clean contracts, tracked savings, and auditable vendor performance. But the dataset they were governing had quietly become a closed system. The 100 suppliers they were measuring against each other were no longer being measured against a market they had stopped looking at.
When I ran the assessment, they were paying 21% over market price.
Not because the data was dirty. Because the data was perfectly clean — and perfectly blind. Every governance process they had built was optimizing performance within a reference frame that had silently become disconnected from competitive reality.
The question that no data governance program asks
Both cases share the same structural failure. And it is not a failure of data quality.
It is a failure of data scope.
In the DND case, the data accurately measured what the system was producing. It did not measure whether the system’s inputs were shaped by behaviors that contradicted its objectives.
In the PC retailer case, the data accurately measured performance within the rationalized supplier base. It did not measure whether that base still reflected the market it was supposed to represent.
Data governance assumes the data is measuring the right thing. It invests heavily in ensuring that measurement is accurate, consistent, and auditable. But accuracy within a mis-scoped dataset does not produce better decisions. It produces more rigorous confirmation of the wrong conclusion.
The question that should precede every data governance initiative is not: how clean is our data?
It is: does this data actually measure what we think it measures — and are the behaviors and structures generating it still aligned with our real objectives?
That is a readiness question, not a technology question. And it is the question that most data governance programs — and the vendors selling them — have no structural incentive to ask.
What this means for AI
The stakes of this failure have never been higher than they are right now.
Every AI implementation in procurement — every model trained on historical purchasing data, every recommendation engine built on supplier performance records, every spend analytics platform optimizing against category benchmarks — is only as reliable as the behavioral and structural integrity of the data it is trained on.
If the data reflects sandbagging, it will optimize sandbagging.
If the data reflects a closed supplier market, it will optimize within that closed market.
Governing that data more rigorously does not solve the problem. It accelerates it.
Phase 0 readiness — the diagnostic discipline that asks whether an organization’s structures, incentives, and data sources are aligned before technology is selected — is not an add-on to data governance. It is the prerequisite that data governance has been skipping for thirty years.
The 80% implementation failure rate is not a data quality problem.
It is a data scope problem that cleaner data cannot fix.
Current industry coverage is available at procureinsights.com. Hansen Models™ and the Hansen Fit Score™ framework: hansenprocurement.com
Jon Hansen — Procurement Insights | Hansen Models™ | Independent. Unsponsored. Archive-based. | procureinsights.com | hansenprocurement.com
-30-
Procurement Insights
Is Data Governance Really About Data?
Posted on March 9, 2026
0
Published on Procurement Insights | Jon Hansen
For The Busy Executive
What is the difference between “clean” data and “blind” data? If you don’t know, then read on before you start (or finish) your next ProcureTech initiative.
Every data governance initiative starts with the same assumption: that the data being cleaned, tracked, and managed is the right data to begin with.
That assumption is almost never examined.
And in my experience, it is the assumption that most often explains why governance programs deliver technically clean data that still produces the wrong outcomes.
Two cases from the archive illustrate this more clearly than any framework I’ve seen.
Case One: The Department of National Defence
When I came in, the MRO procurement platform was delivering 51% next-day performance against a 90% contract requirement. The client’s instinct was immediate: automate the system. Clean up the process. Get the data moving faster.
Before touching anything, I asked one question: what time of day do orders come in?
The answer was 4pm.
That was not a data problem. That was a behavioral signal. The service technicians were incentivized to complete as many service calls per day as possible. Ordering parts after each call — as policy required — was cumbersome and cut into their daily numbers. So they sandbagged. They held all their orders until end of day, submitted them in a batch, and the next-day window closed before the procurement team could respond.
The data being tracked — order volume, delivery rates, parts availability — was perfectly accurate. Every number was real. The governance was sound. But it was capturing the output of an incentive misalignment that no one had identified as the root cause, because no one had asked whether the data reflected actual procurement behavior or the downstream artifact of a structural problem.
Automating that system would have produced faster, cleaner, more efficiently governed data about the wrong thing.
The fix was not data governance. It was an incentive realignment, an agent-based model that correctly identified the technicians as impacting agents, and a supplier, UPS, and customs integration that extended the next-day window. Delivery performance went from 51% to 97.3% in three months. It held for seven years.
Clean data never would have gotten there.
Case Two: The PC Retailer
Shortly after, I was approached by one of the largest PC retailers in the United States. They had executed a vendor rationalization strategy — compressing hundreds of suppliers down to 100, with sound logic: better volume leverage, lower administrative overhead, more manageable contracts.
Two years in, their data showed savings. Year one, strong. Year two, still positive but declining. They wanted an assessment.
I looked at it and said: you do realize that when you compressed your supplier base down to 100, that became your source of truth?
They had lost sight of the rest of the market. Their data governance was excellent — clean contracts, tracked savings, and auditable vendor performance. But the dataset they were governing had quietly become a closed system. The 100 suppliers they were measuring against each other were no longer being measured against a market they had stopped looking at.
When I ran the assessment, they were paying 21% over market price.
Not because the data was dirty. Because the data was perfectly clean — and perfectly blind. Every governance process they had built was optimizing performance within a reference frame that had silently become disconnected from competitive reality.
The question that no data governance program asks
Both cases share the same structural failure. And it is not a failure of data quality.
It is a failure of data scope.
In the DND case, the data accurately measured what the system was producing. It did not measure whether the system’s inputs were shaped by behaviors that contradicted its objectives.
In the PC retailer case, the data accurately measured performance within the rationalized supplier base. It did not measure whether that base still reflected the market it was supposed to represent.
Data governance assumes the data is measuring the right thing. It invests heavily in ensuring that measurement is accurate, consistent, and auditable. But accuracy within a mis-scoped dataset does not produce better decisions. It produces more rigorous confirmation of the wrong conclusion.
The question that should precede every data governance initiative is not: how clean is our data?
It is: does this data actually measure what we think it measures — and are the behaviors and structures generating it still aligned with our real objectives?
That is a readiness question, not a technology question. And it is the question that most data governance programs — and the vendors selling them — have no structural incentive to ask.
What this means for AI
The stakes of this failure have never been higher than they are right now.
Every AI implementation in procurement — every model trained on historical purchasing data, every recommendation engine built on supplier performance records, every spend analytics platform optimizing against category benchmarks — is only as reliable as the behavioral and structural integrity of the data it is trained on.
If the data reflects sandbagging, it will optimize sandbagging.
If the data reflects a closed supplier market, it will optimize within that closed market.
Governing that data more rigorously does not solve the problem. It accelerates it.
Phase 0 readiness — the diagnostic discipline that asks whether an organization’s structures, incentives, and data sources are aligned before technology is selected — is not an add-on to data governance. It is the prerequisite that data governance has been skipping for thirty years.
The 80% implementation failure rate is not a data quality problem.
It is a data scope problem that cleaner data cannot fix.
Current industry coverage is available at procureinsights.com. Hansen Models™ and the Hansen Fit Score™ framework: hansenprocurement.com
Jon Hansen — Procurement Insights | Hansen Models™ | Independent. Unsponsored. Archive-based. | procureinsights.com | hansenprocurement.com
-30-
Share this:
Related