Marijn Overvest’s 121-leader survey landed on May 8th. The headline numbers — 47% of procurement professionals using AI every working day, 8% in organizations that have formally embedded it, 83% operating without an enforced AI policy — are striking. They are also not new. They are wave four.
I want to start with the data point and then step back from it, because the data point is not the story. The story is that this is the fourth time this pattern has played out in the last thirty-five years, and the discipline is naming it as if it were the first.
Marijn’s survey found that procurement teams have already estimated they could delegate 10.6 hours per week to AI — more than a full workday. The teams are not waiting for governance. They are not waiting for policy. They are not waiting for the ProcureTech vendor’s roadmap or the consultancy’s recommended framework. They are using consumer AI tools right now, today, on contract data and supplier data and pricing analyses, with no enforced guardrails sitting between the prompt and the output.
That is the AI wave. It is the most recent wave. It is not the first.
Wave One: Spreadsheets (1990–present)
The first article in the Procurement Insights archive about spreadsheets is a March 2008 guest post by Glenda Leatherman, a change management consultant working on a three-way telecom merger. The methodology she described for integrating three procurement organizations under regulatory pressure on Day 1 of a merger was not built around the merging companies’ ERP systems. It was built around spreadsheets. Each company would populate an agreed schema — names, titles, User IDs, manager hierarchies, supplier contracts, expiration dates, spend categories. On Day 1 the spreadsheets would be exchanged through third-party attorneys, and the integration would proceed.
Read that paragraph again. In 2008, at the height of the SAP/Oracle/PeopleSoft era, when three large telecoms were merging in a regulated industry, the practical operational tool that ran the integration was Excel. Not because the consultants didn’t know about ERP. Because the spreadsheet was the only tool fast enough, flexible enough, and accessible enough to actually move the work forward in the time available.
Seventeen years later, the Procurement Insights archive on spreadsheet adoption shows estimated organizational use moving from roughly 98% in 2005 to 85–90% in 2025. The decline is real but mild. Two decades of ProcureTech investment, hundreds of platforms launched, billions of dollars deployed — and the spreadsheet is still in roughly nine out of ten procurement workflows.
The error data tells the deeper part of the story. Spreadsheets carry an estimated 88% error rate at the file level. ProcureTech implementations carry a 70–80% failure rate at the initiative level. The spreadsheet wins the error battle by a wide margin — not because it is more accurate per cell, but because the human at the keyboard is in the loop, exercising judgment, catching mistakes, adjusting course. The technology-led system at the enterprise level fails to ship at all, or ships and is not adopted, or ships and is reverted.
That is wave one. The bottom-up tool was good enough, individual enough, and adaptable enough to outrun every sanctioned alternative for thirty-five years. It is still doing it.
Wave Two: BYOD (2007–present)
I have a personal anchor for this one. In the early 1980s, I was banging on the keyboard of my old Kaypro running CP/M before the DOS-based PC had landed in the corporate workplace. The IT department viewed the personal computer as a passing phase — something the workforce would get over once the central mainframe people stopped indulging it. Forty-three years later, the personal computer has been everything except a passing phase, and the IT department’s relationship to user-driven hardware has been a continuous rolling defeat.
The 2007 iPhone was the version of that story most people remember. By 2010 the corporate IT function was openly losing the perimeter — employees were carrying their own devices into work, accessing email from their own phones, syncing corporate calendars to their own iPads. The discipline named the resulting problem “BYOD” and built mobile device management platforms to recover the perimeter. The platforms helped. The pattern did not change. The perimeter never came back. The user kept the device.
Wave two looked operationally different from wave one — hardware instead of software, mobility instead of analysis — but the structural pattern was identical. Bottom-up adoption raced ahead. Enterprise sanction lagged behind. The gap stayed open. The vocabulary the discipline used to describe it (“BYOD policy,” “mobile device management,” “shadow IT perimeter”) was operationally legitimate and structurally insufficient.
Wave Three: SaaS App Sprawl (2012–present)
By the mid-2010s, SaaS had collapsed the cost and timeline of deploying enterprise software. A platform that used to require a six-figure license, a multi-quarter implementation, and procurement-committee sign-off now took a credit card and a thirty-minute signup. Departments stopped waiting for IT. Procurement stopped being consulted on departmental SaaS purchases — and over time, procurement teams started running their own shadow SaaS stacks too. The discipline meant to govern enterprise software acquisition became one of the functions undermining it. By December 2022, Forrester and Airtable found that the average large business was running 367 software applications, with workers spending 2.4 hours per day — nearly a third of the work week — sorting through silos to find the data they needed.
I was watching this in real time and writing about it in 2010, when SAP’s John Wookey took the stage at SAP Sapphire and described what he called the SaaS conundrum. Customers had bought SAP. SAP didn’t fit. Customers wanted upgrades. Upgrades were too expensive. Customers got frustrated and bought SaaS instead, departmentally, in parallel, without telling SAP. Wookey’s proposed answer was that SAP would now offer the on-demand functionality the customers had already gone elsewhere to get, and the customers would presumably consolidate back to SAP.
I responded at the time in a post called “SaaS Sprawl, One-Stop Shopping, and Free 8-Tracks To Boot”. The argument was that the snakes were already out of the playpen and SAP could not put them back. The functional silos that had given large ERP vendors their pricing power had already dissolved into enterprise-wide collaborative connectivity. The market had crossed a threshold, and the threshold did not run in the direction Wookey wanted to go. That post is fifteen years old. The CIODive 367-apps statistic from December 2022 is the empirical confirmation.
Wave three took roughly ten years to fully manifest. The discipline named it shadow IT, master data management, SaaS governance, application rationalization. None of those vocabularies asked the structural question. The structural question was whether the organization could absorb three hundred and sixty-seven independent decision-making applications without the implicit governance the original ERP architecture had provided — and the answer, demonstrably, was no.
Wave Four: AI (2022–present)
ChatGPT launched in November 2022, the same month CIODive published the 367-apps finding. Three and a half years later, Marijn’s survey arrives at 47% daily AI use against 8% formal embedding and 83% absence of enforced policy.
I wrote about exactly this gap last December in “Why Spreadsheets Keep Beating AI: The Case for Humans at the Wheel”. Spreadsheets succeed because the human is structurally in the loop — there is no way to use a spreadsheet without the human at the wheel. AI fails when humans are not ready to lead it, because the technology demands governance, clean data, and clear ownership the organization does not have, and the technology accelerates whatever it accelerates regardless of whether the underlying conditions are sound. The result is what I have been calling the velocity trap: the AI does not fix the dysfunction; it scales it.
The same point in slightly different framing: in May 2025 I asked, “If AI replaces spreadsheets, what will eventually replace AI for procurement?” The implicit answer is that AI is not the end state. AI is wave four. There will be a wave five. The question that matters is not which technology the next wave brings, but whether the discipline can recognize the recurring pattern in real time rather than documenting it after the fact.
The Pattern, Charted
Two lines, illustrative not measured, because no single dataset tracks bottom-up adoption against sanctioned adoption across thirty-five years. What we have is each wave’s individual evidence base — the spreadsheet usage data, the BYOD perimeter studies, the 367-apps statistic, Marijn’s 47%/8%/83%. The chart’s job is to surface the pattern, not the precision. The pattern is the persistent gap of roughly thirty to thirty-five percentage points between what people are actually doing with technology and what the organization has formally sanctioned them to do.
The gap does not close. It has not closed once in thirty-five years. Every wave has produced a fresh round of governance literature promising that this time the policy framework would catch up to the practice. None of them have. The reason is structural: bottom-up adoption requires only a credit card, a download, or a prompt. Top-down sanction requires committee, budget, alignment, audit, and absorption capacity. The adoption clock runs at minutes. The sanction clock runs at quarters. The clocks are not converging.
The Cadence Has Compressed
The wave-one bolt-on era took roughly fifteen years for its structural failures to surface in human-recognizable forms — Hershey’s order-fulfillment collapse, FoxMeyer’s bankruptcy, Cadbury’s chocolate-bar overproduction, the King County maelstrom. The wave-three SaaS app sprawl era surfaced its failures in three to five years. The wave-four AI era will not get those buffers, and I made the case for this directly in “Bolt-Ons. App Sprawl. Replicated Lakehouses. Same Open IKEA Box.”. AI workflows operate at machine cadence. They propagate through context layers that scale failure rather than contain it. Failures that took fifteen years to manifest in 1998 will manifest in fifteen weeks in 2026.
This is the part most current AI-readiness commentary misses. The structural risk of shadow adoption has been with us for thirty-five years. What changes in 2026 is not the existence of the risk. What changes is the speed at which the risk converts into consequences the organization can no longer absorb gracefully. The grace period is gone.
What Phase 0™ Is Built To Do
Policy, by itself, has never closed the gap. Four waves of governance literature should have settled that question by now. What closes the gap is structural readiness assessed before the technology lands — Phase 0™, in the Hansen Models™ vocabulary, but the deeper point is independent of any one framework. The deeper point is that the human strand, the incentive architecture, the workflow physics, and the absorption capacity of the organization have to be diagnosed before the technology gets a seat at the table, not after.
Spreadsheets succeeded because humans were structurally required to be at the wheel. AI fails when humans are not ready to be. Phase 0™ is the bridge between the two. It does not change what the technology can do. It changes whether the organization is structurally positioned to absorb what the technology does. Without that bridge, AI becomes the spreadsheet’s velocity-amplified shadow — same pattern, faster cadence, larger blast radius.
The Archive Was Not Hindsight
I want to close by naming something the discipline tends to underweight. The pattern visible in the chart above did not become visible in 2026 with Marijn’s data. It was being documented in real time, with corroborating evidence, in this archive for nineteen years. The 2008 Glenda Leatherman piece. The 2010 Wookey response. The 2018 “Which Came First: Procurement’s Transformation or Technology’s?” post. The 2023 technology adoption curve piece. The May 2025 spreadsheet decline analysis and the May 2025 “what replaces AI” piece. The December 2025 humans-at-the-wheel argument. The May 2026 IKEA-box piece. Each one was contemporaneous. Each one named a piece of the pattern as it was emerging, with the language available at the time.
That is the function the contemporaneous archive performs. It is not memory. It is not retrospective construction. It is the documented record of structural recognition occurring in real time, against the prevailing vocabulary of the moment, with the predictions on the page and the dates on the posts. It is what makes the pattern argument hard to dismiss when the fourth-wave version of the data finally arrives.
Marijn’s data is wave four. It is real, it is important, and it is exactly what the archive said wave four was going to look like.
The question is whether wave five gets recognized before it arrives.
Phase 0™ · Hansen Fit Score™ (HFS™) · RAM 2025™ · Strand Commonality™ · Real-World Condition Substrate™ Hansen Models™ · Founder: Jon W. Hansen · hansenprocurement.com
-30-
Procurement Insights
Marijn’s Data Is Wave Four: Why the AI Adoption Gap Was Always Going to Look Like This
Posted on May 8, 2026
0
Marijn Overvest’s 121-leader survey landed on May 8th. The headline numbers — 47% of procurement professionals using AI every working day, 8% in organizations that have formally embedded it, 83% operating without an enforced AI policy — are striking. They are also not new. They are wave four.
I want to start with the data point and then step back from it, because the data point is not the story. The story is that this is the fourth time this pattern has played out in the last thirty-five years, and the discipline is naming it as if it were the first.
Marijn’s survey found that procurement teams have already estimated they could delegate 10.6 hours per week to AI — more than a full workday. The teams are not waiting for governance. They are not waiting for policy. They are not waiting for the ProcureTech vendor’s roadmap or the consultancy’s recommended framework. They are using consumer AI tools right now, today, on contract data and supplier data and pricing analyses, with no enforced guardrails sitting between the prompt and the output.
That is the AI wave. It is the most recent wave. It is not the first.
Wave One: Spreadsheets (1990–present)
The first article in the Procurement Insights archive about spreadsheets is a March 2008 guest post by Glenda Leatherman, a change management consultant working on a three-way telecom merger. The methodology she described for integrating three procurement organizations under regulatory pressure on Day 1 of a merger was not built around the merging companies’ ERP systems. It was built around spreadsheets. Each company would populate an agreed schema — names, titles, User IDs, manager hierarchies, supplier contracts, expiration dates, spend categories. On Day 1 the spreadsheets would be exchanged through third-party attorneys, and the integration would proceed.
Read that paragraph again. In 2008, at the height of the SAP/Oracle/PeopleSoft era, when three large telecoms were merging in a regulated industry, the practical operational tool that ran the integration was Excel. Not because the consultants didn’t know about ERP. Because the spreadsheet was the only tool fast enough, flexible enough, and accessible enough to actually move the work forward in the time available.
Seventeen years later, the Procurement Insights archive on spreadsheet adoption shows estimated organizational use moving from roughly 98% in 2005 to 85–90% in 2025. The decline is real but mild. Two decades of ProcureTech investment, hundreds of platforms launched, billions of dollars deployed — and the spreadsheet is still in roughly nine out of ten procurement workflows.
The error data tells the deeper part of the story. Spreadsheets carry an estimated 88% error rate at the file level. ProcureTech implementations carry a 70–80% failure rate at the initiative level. The spreadsheet wins the error battle by a wide margin — not because it is more accurate per cell, but because the human at the keyboard is in the loop, exercising judgment, catching mistakes, adjusting course. The technology-led system at the enterprise level fails to ship at all, or ships and is not adopted, or ships and is reverted.
That is wave one. The bottom-up tool was good enough, individual enough, and adaptable enough to outrun every sanctioned alternative for thirty-five years. It is still doing it.
Wave Two: BYOD (2007–present)
I have a personal anchor for this one. In the early 1980s, I was banging on the keyboard of my old Kaypro running CP/M before the DOS-based PC had landed in the corporate workplace. The IT department viewed the personal computer as a passing phase — something the workforce would get over once the central mainframe people stopped indulging it. Forty-three years later, the personal computer has been everything except a passing phase, and the IT department’s relationship to user-driven hardware has been a continuous rolling defeat.
The 2007 iPhone was the version of that story most people remember. By 2010 the corporate IT function was openly losing the perimeter — employees were carrying their own devices into work, accessing email from their own phones, syncing corporate calendars to their own iPads. The discipline named the resulting problem “BYOD” and built mobile device management platforms to recover the perimeter. The platforms helped. The pattern did not change. The perimeter never came back. The user kept the device.
Wave two looked operationally different from wave one — hardware instead of software, mobility instead of analysis — but the structural pattern was identical. Bottom-up adoption raced ahead. Enterprise sanction lagged behind. The gap stayed open. The vocabulary the discipline used to describe it (“BYOD policy,” “mobile device management,” “shadow IT perimeter”) was operationally legitimate and structurally insufficient.
Wave Three: SaaS App Sprawl (2012–present)
By the mid-2010s, SaaS had collapsed the cost and timeline of deploying enterprise software. A platform that used to require a six-figure license, a multi-quarter implementation, and procurement-committee sign-off now took a credit card and a thirty-minute signup. Departments stopped waiting for IT. Procurement stopped being consulted on departmental SaaS purchases — and over time, procurement teams started running their own shadow SaaS stacks too. The discipline meant to govern enterprise software acquisition became one of the functions undermining it. By December 2022, Forrester and Airtable found that the average large business was running 367 software applications, with workers spending 2.4 hours per day — nearly a third of the work week — sorting through silos to find the data they needed.
I was watching this in real time and writing about it in 2010, when SAP’s John Wookey took the stage at SAP Sapphire and described what he called the SaaS conundrum. Customers had bought SAP. SAP didn’t fit. Customers wanted upgrades. Upgrades were too expensive. Customers got frustrated and bought SaaS instead, departmentally, in parallel, without telling SAP. Wookey’s proposed answer was that SAP would now offer the on-demand functionality the customers had already gone elsewhere to get, and the customers would presumably consolidate back to SAP.
I responded at the time in a post called “SaaS Sprawl, One-Stop Shopping, and Free 8-Tracks To Boot”. The argument was that the snakes were already out of the playpen and SAP could not put them back. The functional silos that had given large ERP vendors their pricing power had already dissolved into enterprise-wide collaborative connectivity. The market had crossed a threshold, and the threshold did not run in the direction Wookey wanted to go. That post is fifteen years old. The CIODive 367-apps statistic from December 2022 is the empirical confirmation.
Wave three took roughly ten years to fully manifest. The discipline named it shadow IT, master data management, SaaS governance, application rationalization. None of those vocabularies asked the structural question. The structural question was whether the organization could absorb three hundred and sixty-seven independent decision-making applications without the implicit governance the original ERP architecture had provided — and the answer, demonstrably, was no.
Wave Four: AI (2022–present)
ChatGPT launched in November 2022, the same month CIODive published the 367-apps finding. Three and a half years later, Marijn’s survey arrives at 47% daily AI use against 8% formal embedding and 83% absence of enforced policy.
I wrote about exactly this gap last December in “Why Spreadsheets Keep Beating AI: The Case for Humans at the Wheel”. Spreadsheets succeed because the human is structurally in the loop — there is no way to use a spreadsheet without the human at the wheel. AI fails when humans are not ready to lead it, because the technology demands governance, clean data, and clear ownership the organization does not have, and the technology accelerates whatever it accelerates regardless of whether the underlying conditions are sound. The result is what I have been calling the velocity trap: the AI does not fix the dysfunction; it scales it.
The same point in slightly different framing: in May 2025 I asked, “If AI replaces spreadsheets, what will eventually replace AI for procurement?” The implicit answer is that AI is not the end state. AI is wave four. There will be a wave five. The question that matters is not which technology the next wave brings, but whether the discipline can recognize the recurring pattern in real time rather than documenting it after the fact.
The Pattern, Charted
Two lines, illustrative not measured, because no single dataset tracks bottom-up adoption against sanctioned adoption across thirty-five years. What we have is each wave’s individual evidence base — the spreadsheet usage data, the BYOD perimeter studies, the 367-apps statistic, Marijn’s 47%/8%/83%. The chart’s job is to surface the pattern, not the precision. The pattern is the persistent gap of roughly thirty to thirty-five percentage points between what people are actually doing with technology and what the organization has formally sanctioned them to do.
The gap does not close. It has not closed once in thirty-five years. Every wave has produced a fresh round of governance literature promising that this time the policy framework would catch up to the practice. None of them have. The reason is structural: bottom-up adoption requires only a credit card, a download, or a prompt. Top-down sanction requires committee, budget, alignment, audit, and absorption capacity. The adoption clock runs at minutes. The sanction clock runs at quarters. The clocks are not converging.
The Cadence Has Compressed
The wave-one bolt-on era took roughly fifteen years for its structural failures to surface in human-recognizable forms — Hershey’s order-fulfillment collapse, FoxMeyer’s bankruptcy, Cadbury’s chocolate-bar overproduction, the King County maelstrom. The wave-three SaaS app sprawl era surfaced its failures in three to five years. The wave-four AI era will not get those buffers, and I made the case for this directly in “Bolt-Ons. App Sprawl. Replicated Lakehouses. Same Open IKEA Box.”. AI workflows operate at machine cadence. They propagate through context layers that scale failure rather than contain it. Failures that took fifteen years to manifest in 1998 will manifest in fifteen weeks in 2026.
This is the part most current AI-readiness commentary misses. The structural risk of shadow adoption has been with us for thirty-five years. What changes in 2026 is not the existence of the risk. What changes is the speed at which the risk converts into consequences the organization can no longer absorb gracefully. The grace period is gone.
What Phase 0™ Is Built To Do
Policy, by itself, has never closed the gap. Four waves of governance literature should have settled that question by now. What closes the gap is structural readiness assessed before the technology lands — Phase 0™, in the Hansen Models™ vocabulary, but the deeper point is independent of any one framework. The deeper point is that the human strand, the incentive architecture, the workflow physics, and the absorption capacity of the organization have to be diagnosed before the technology gets a seat at the table, not after.
Spreadsheets succeeded because humans were structurally required to be at the wheel. AI fails when humans are not ready to be. Phase 0™ is the bridge between the two. It does not change what the technology can do. It changes whether the organization is structurally positioned to absorb what the technology does. Without that bridge, AI becomes the spreadsheet’s velocity-amplified shadow — same pattern, faster cadence, larger blast radius.
The Archive Was Not Hindsight
I want to close by naming something the discipline tends to underweight. The pattern visible in the chart above did not become visible in 2026 with Marijn’s data. It was being documented in real time, with corroborating evidence, in this archive for nineteen years. The 2008 Glenda Leatherman piece. The 2010 Wookey response. The 2018 “Which Came First: Procurement’s Transformation or Technology’s?” post. The 2023 technology adoption curve piece. The May 2025 spreadsheet decline analysis and the May 2025 “what replaces AI” piece. The December 2025 humans-at-the-wheel argument. The May 2026 IKEA-box piece. Each one was contemporaneous. Each one named a piece of the pattern as it was emerging, with the language available at the time.
That is the function the contemporaneous archive performs. It is not memory. It is not retrospective construction. It is the documented record of structural recognition occurring in real time, against the prevailing vocabulary of the moment, with the predictions on the page and the dates on the posts. It is what makes the pattern argument hard to dismiss when the fourth-wave version of the data finally arrives.
Marijn’s data is wave four. It is real, it is important, and it is exactly what the archive said wave four was going to look like.
The question is whether wave five gets recognized before it arrives.
Phase 0™ · Hansen Fit Score™ (HFS™) · RAM 2025™ · Strand Commonality™ · Real-World Condition Substrate™ Hansen Models™ · Founder: Jon W. Hansen · hansenprocurement.com
-30-
Share this:
Related