For two decades we tried to eliminate it. The AI era is about to show us what it was telling us all along.
For more than twenty years, procurement has treated maverick spend as something to stamp out. The logic was clean: if people followed the approved process, outcomes would improve; if they stopped buying outside contracted channels, savings would rise; if everyone complied, procurement would finally have control.
It is worth asking whether we have been reading the wrong thing the whole time. Not whether maverick spend is good or bad — but whether the deviation itself is a compliance problem, or an information signal we have been trained to suppress.
The question nobody was asking
In 1998, during a Department of National Defence engagement, MRO parts were arriving next-day only 51% of the time against a 90% contractual requirement. The expected move was to upgrade the technology — automate harder, throw more capability at the problem.
The breakthrough came from a question that sounded irrelevant: What time of day do orders come in?
The answer — most of them at 4 PM — unraveled a chain nobody had connected. Service technicians were batching their parts orders to the end of the day because their incentive was calls completed, not parts delivered. Late orders hit dynamic-flux pricing windows. Small suppliers without customs-formatting experience were held at the border. Manual courier dispatch compounded the delay. None of this was hidden. The data existed, the systems existed, the documentation existed. What was missing was the context connecting them — what I have come to call context debt. Within three months, next-day delivery moved from 51% to 97.3%, not because anyone added data, but because someone finally read what the data had been saying.
Here is the part worth being precise about. The technicians were not exercising some superior wisdom the system should have adopted. Their behavior was genuinely dysfunctional, and it was corrected. But the pattern of that behavior — the 4 PM cluster — was a signal sitting in plain sight, and the organization had been treating it as noise. That is the distinction that matters. The deviation was not the answer. It was the question the organization did not know to ask.
The same signal, a different verdict
Sometimes the deviation runs the other way. For certain commodity types — tail spend, indirect MRO, dynamic-flux items — the buyer who goes off-contract is not gaming a metric. They are right. They have learned that the centrally negotiated arrangement produces a worse outcome than what they can achieve themselves, and the official process simply has not caught up. Same surface behavior — spend outside the approved channel — opposite verdict on the behavior itself.
This is why “eliminate maverick spend” is the wrong reflex. It treats two different conditions as one problem and erases the signal in both. In the first case, the deviation reveals a misalignment that needs fixing. In the second, it reveals an official process that is wrong. The behavior is not the constant. The information is.
A practitioner arrives at the same place
Recently, in a discussion on technology debt, procurement strategist Paul Martyn connected his own work to that DND question without being prompted to. He described a sourcing optimization where the savings came out so large the purchasing manager asked for them to be reduced and spread across the contract term — because the result was big enough to create problems for him internally. The math was not wrong. The savings were real.
As Paul put it: “The math wasn’t the issue. The organization around the math was.”
Then he made the link explicit: the DND example was the same pattern he had been seeing in optimization work — operational debt accumulating quietly in the gap between the official process and what people actually had to do to make the system function.
That convergence is worth pausing on. A practitioner from a different corner of the field, working different engagements, recognized his own experience in a question first asked at DND in 1998. Two unconnected careers, the same finding: the deviation was never the problem. It was the organization’s most honest record of where its official model and its operating reality had quietly come apart.
Why AI changes the timeline
Earlier technology eras let organizations carry that gap for years. An ERP could run for a decade while users quietly built the spreadsheet workarounds that actually carried the load. A transformation program could declare victory while the unofficial process did the real work underneath. Organizations adapted — which is not the same as being healthy.
AI removes the slack. As decision cycles compress, the consequences of an unexamined assumption compress with them. What used to take years to surface can surface in weeks or days. Point an autonomous agent at an environment whose real operating conditions were never mapped, and it will not discover the workarounds — it will execute straight through them, faster and at greater cost than any system before it. The gap is no longer inherited slowly. It is stress-tested at the speed of thought.
The real question
The organizations that do well in the AI era will not necessarily be the ones with the best models. They will be the ones that understood the reality their models were entering. The deviations they spent twenty years trying to erase are the closest thing they have to a map of that reality.
Which leaves a different question than the one we have been asking. Not how do we eliminate maverick spend — but what has it been telling us all along?
The answer may decide whether AI retires organizational debt, or simply automates it.
-30-
Related
Maverick Spend Was Never the Problem
Posted on June 3, 2026
0
For two decades we tried to eliminate it. The AI era is about to show us what it was telling us all along.
For more than twenty years, procurement has treated maverick spend as something to stamp out. The logic was clean: if people followed the approved process, outcomes would improve; if they stopped buying outside contracted channels, savings would rise; if everyone complied, procurement would finally have control.
It is worth asking whether we have been reading the wrong thing the whole time. Not whether maverick spend is good or bad — but whether the deviation itself is a compliance problem, or an information signal we have been trained to suppress.
The question nobody was asking
In 1998, during a Department of National Defence engagement, MRO parts were arriving next-day only 51% of the time against a 90% contractual requirement. The expected move was to upgrade the technology — automate harder, throw more capability at the problem.
The breakthrough came from a question that sounded irrelevant: What time of day do orders come in?
The answer — most of them at 4 PM — unraveled a chain nobody had connected. Service technicians were batching their parts orders to the end of the day because their incentive was calls completed, not parts delivered. Late orders hit dynamic-flux pricing windows. Small suppliers without customs-formatting experience were held at the border. Manual courier dispatch compounded the delay. None of this was hidden. The data existed, the systems existed, the documentation existed. What was missing was the context connecting them — what I have come to call context debt. Within three months, next-day delivery moved from 51% to 97.3%, not because anyone added data, but because someone finally read what the data had been saying.
Here is the part worth being precise about. The technicians were not exercising some superior wisdom the system should have adopted. Their behavior was genuinely dysfunctional, and it was corrected. But the pattern of that behavior — the 4 PM cluster — was a signal sitting in plain sight, and the organization had been treating it as noise. That is the distinction that matters. The deviation was not the answer. It was the question the organization did not know to ask.
The same signal, a different verdict
Sometimes the deviation runs the other way. For certain commodity types — tail spend, indirect MRO, dynamic-flux items — the buyer who goes off-contract is not gaming a metric. They are right. They have learned that the centrally negotiated arrangement produces a worse outcome than what they can achieve themselves, and the official process simply has not caught up. Same surface behavior — spend outside the approved channel — opposite verdict on the behavior itself.
This is why “eliminate maverick spend” is the wrong reflex. It treats two different conditions as one problem and erases the signal in both. In the first case, the deviation reveals a misalignment that needs fixing. In the second, it reveals an official process that is wrong. The behavior is not the constant. The information is.
A practitioner arrives at the same place
Recently, in a discussion on technology debt, procurement strategist Paul Martyn connected his own work to that DND question without being prompted to. He described a sourcing optimization where the savings came out so large the purchasing manager asked for them to be reduced and spread across the contract term — because the result was big enough to create problems for him internally. The math was not wrong. The savings were real.
As Paul put it: “The math wasn’t the issue. The organization around the math was.”
Then he made the link explicit: the DND example was the same pattern he had been seeing in optimization work — operational debt accumulating quietly in the gap between the official process and what people actually had to do to make the system function.
That convergence is worth pausing on. A practitioner from a different corner of the field, working different engagements, recognized his own experience in a question first asked at DND in 1998. Two unconnected careers, the same finding: the deviation was never the problem. It was the organization’s most honest record of where its official model and its operating reality had quietly come apart.
Why AI changes the timeline
Earlier technology eras let organizations carry that gap for years. An ERP could run for a decade while users quietly built the spreadsheet workarounds that actually carried the load. A transformation program could declare victory while the unofficial process did the real work underneath. Organizations adapted — which is not the same as being healthy.
AI removes the slack. As decision cycles compress, the consequences of an unexamined assumption compress with them. What used to take years to surface can surface in weeks or days. Point an autonomous agent at an environment whose real operating conditions were never mapped, and it will not discover the workarounds — it will execute straight through them, faster and at greater cost than any system before it. The gap is no longer inherited slowly. It is stress-tested at the speed of thought.
The real question
The organizations that do well in the AI era will not necessarily be the ones with the best models. They will be the ones that understood the reality their models were entering. The deviations they spent twenty years trying to erase are the closest thing they have to a map of that reality.
Which leaves a different question than the one we have been asking. Not how do we eliminate maverick spend — but what has it been telling us all along?
The answer may decide whether AI retires organizational debt, or simply automates it.
-30-
Share this:
Like this:
Related