Purchasing embraced spreadsheets for a reason. It is the same reason most AI deployments are not producing what was expected — and it has nothing to do with the technology in either era.
Jon W. Hansen, FCIPS | Procurement Insights | August 2026
THE SHORT VERSION FOR BUSY EXECUTIVES
Going back to 1983, the purchasing department had virtually no say in what technology platform their company implemented. We were an afterthought, waiting in a long line of competing departments to be told the equivalent of you’ll get nothing and like it.
Looking back, is it any wonder we embraced spreadsheets — and continue to?
That is not a joke about Excel. A spreadsheet is what a function builds when its operating reality has no way into the technology decision. It is a local instrument, outside the sanctioned stack, holding the model the official system does not contain. Four decades of spreadsheets is four decades of evidence that the requirement never reached the decision.
The hierarchy that produced them is still in place. What has changed is the cost of it. A platform that ignores operating reality underperforms. An autonomous agent reasoning from the same incomplete model executes against it continuously, at speed, with every indicator on its dashboard improving.
A DEEPER DIVE
What I wrote in 2007, and why it still reads as current
In June 2007, in part six of the Dangerous Supply Chain Myths series, I put the mechanism this way:
However innovative the technology was or is, its ultimate benefits could only be realized in an environment where the unique operating requirements of key stakeholders were fully recognized and incorporated into the core strategy.
Unfortunately, most purchasing departments inherit their software as an adjunct downstream byproduct of either an original finance or IT-centric initiative. This has usually meant that their input had been relegated to the realm of the afterthought versus providing decisive and proactive input when it matters the most — prior to a decision being made.
Read that again with the current wave in mind, and notice that it is not a complaint about status. It is a claim about sequence.
The operating requirement does not fail to arrive. It arrives after the decision — as implementation feedback, as change management, as an issue log. By then the architecture is fixed, the contract is signed, and the requirement can only be accommodated as an exception rather than incorporated as a condition.
In the same post I gave the thing a name: the hierarchical implementation mechanism, which I described as still firmly entrenched in the marketplace. It is the same arrangement I had walked into in 1983 — by 2007 it had simply acquired an architecture diagram. And I wrote a sentence that reads oddly now for someone who has spent the last month arguing about AI placement:
It is ironic that the actual software you use has very little to do with the success of your procurement initiative.
That was June 2007, about SOA.
The post also carried the sequence rule, stated plainly:
It is at this very point when a foundational understanding exists and a consensus has been reached between stakeholders that technology — whether existing or proposed — can be introduced as a means of driving greater efficiency in the supply chain.
Foundation first, consensus on operating reality second, technology third. Nineteen years before I built a stack to show where AI belongs, the sequencing argument was already on the page.
And it ended with a condition: unless this hierarchical practice is changed. It has not been.
What I asked in 2025, and what today’s post adds to it
In April 2025 I returned to the same problem and asked how strong procurement’s voice and influence had become in ProcureTech selection. I said the answer would determine the trajectory of implementation success for the next decade.
That question still stands. Procurement needs a seat at the table — and so does every other stakeholder whose operating reality bears on whether the decision works. Finance, IT, service delivery, the front line. None of them holds the whole picture, and none of them can be left downstream of the architecture without cost.
What today’s post adds is scope. The 2025 question was asked about one function because that is the function I came from and the one whose exclusion I had watched since 1983. The same exclusion runs across every function whose operating reality is not represented when the architecture is settled.
And the 2007 post had already said what makes a seat useful rather than decorative:
It is at this very point when a foundational understanding exists and a consensus has been reached between stakeholders that technology — whether existing or proposed — can be introduced.
Not when the right people are in the room. When the operating reality has been established. A seat matters because of what the person in it brings, and what they bring only counts if it arrives before the decision rather than after it.
What forty years of spreadsheets actually document
Every spreadsheet running alongside a sanctioned system is a record of the same event: somebody’s operating reality did not fit the model, and they built a private instrument to hold the difference.
(With the emergence of AI, the challenges of spreadsheet independence are significantly compounded by the proliferation of AI model access outside the main enterprise — personal AI is becoming the new shadow spreadsheet.)
Nobody logs that as a finding. It appears as shadow IT, as a data-governance problem, as user resistance, as something to be migrated in the next phase. It is treated as a symptom of poor adoption.
It is the opposite. It is the most reliable evidence available that the declared model and the real operation have diverged, produced continuously, by the people closest to the work, at their own cost.
I have seen the same shape in every case I have traced. Technicians batching parts orders to four in the afternoon because they were rated on calls responded to. Buyers disqualifying a supplier over a port strike the data feed had not heard about. A finance function reconciling part cost and freight as two clean line items with the loss living between them.
None of it was misconduct. All of it was people holding something true that the system had no field for.
And the spreadsheet has a twin that costs more. In that same 2007 post I described a conversation with a senior executive who had two full-time staff dedicated to making a PeopleSoft application work inside their procurement organization — which confined their effort to the application rather than to understanding and refining the practice itself.
I put it this way at the time: the vast majority of purchasing organizations are now in the software business, as they attempt to adapt their practice to an application they would not have chosen to use in the first place.
Those are the two available responses to an inherited system. Build a private instrument beside it, or assign people to bend your practice into it. One is filed as shadow IT. The other is filed as headcount. Neither is filed as evidence that the requirement arrived too late.
Why the 1983 hierarchy is more expensive now than it has ever been
For most of those four decades, the cost of the sequence was underperformance. The platform went in, it did less than promised, the spreadsheets stayed, and the organization absorbed the gap.
Autonomy changes the arithmetic.
An agent given a decision inherits the model it was given. If the operating requirement never entered the architecture, the agent does not know it is missing — and it does not hesitate, does not build a workaround, and does not keep a private spreadsheet. It executes the incomplete model faster and more consistently than any human ever did, and every metric inside its scope improves while it does.
That is why the current failure figures look the way they do. RAND’s study of AI project failure opens by relaying other estimates — that more than 80 percent of AI projects fail, roughly twice the rate of IT projects without AI. RAND is characterising figures it did not measure, so read it as the large majority fail rather than as a percentage.
What RAND did measure is the part that matters. From its own interviews with sixty-five data scientists and engineers, the root causes are overwhelmingly organizational rather than technical: misaligned purpose, weak data foundations, integration treated as an afterthought, fading sponsorship. Not one of those causes sits in the model.
Integration treated as an afterthought. That is the 2007 sentence, in someone else’s words, nineteen years later.
So what would moving past it actually require?
Not a seat at the table. That is the 2025 framing and it accepts the table.
It requires that operating evidence and the technology decision become structurally inseparable — that the architecture cannot be settled without the operating reality having been established first, because the decision depends on it rather than being informed by it.
Three questions that test whether an organization has moved past the 1983 sequence. None of them are technical, and they sit alongside the placement questions in the Human/Tech Stack™.
Was the operating reality established before the architecture, or after? If the requirement document was written from the declared process, the sequence has not changed.
What is the front line currently holding outside the sanctioned system, and why? Every spreadsheet, side database and personal tracker is an answer to a question nobody asked. Find them before deciding what to automate, not during rollout.
And if the operating evidence contradicts the strategy, what happens? If the answer is that it gets logged as a change request, the evidence is downstream. If the answer is that the architecture changes, it is not.
Are we finally ready?
I have been asking a version of this question since before I entered procurement, and the honest answer is that I do not know.
What I do know is that the hierarchy has survived four technology eras — mainframe and ERP, e-procurement, SaaS and cloud, and now AI — and that each era arrived with the claim that this time the technology was different. Each time it was. The sequence was not.
The difference now is that the cost of leaving it alone has changed. An unexamined operating model used to produce a system that underdelivered. It now produces a system that acts on its own, confidently, at scale, from a picture with a department missing from it.
Forty years of spreadsheets were people telling us where those gaps were. We filed it under adoption.
We will know the sequence has changed when the first requirement document gets written from what the front line is holding outside the system, rather than from the process the organization believes it runs.
Jon W. Hansen, FCIPS — Procurement Insights | Hansen Models™ | Independent. Unsponsored. Archive-based.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Related
To Ensure AI Implementation Success, Are We Finally Ready to Move Past the 1983 Decision Hierarchy Framework?
Posted on August 27, 2026
0
Purchasing embraced spreadsheets for a reason. It is the same reason most AI deployments are not producing what was expected — and it has nothing to do with the technology in either era.
Jon W. Hansen, FCIPS | Procurement Insights | August 2026
THE SHORT VERSION FOR BUSY EXECUTIVES
Going back to 1983, the purchasing department had virtually no say in what technology platform their company implemented. We were an afterthought, waiting in a long line of competing departments to be told the equivalent of you’ll get nothing and like it.
Looking back, is it any wonder we embraced spreadsheets — and continue to?
That is not a joke about Excel. A spreadsheet is what a function builds when its operating reality has no way into the technology decision. It is a local instrument, outside the sanctioned stack, holding the model the official system does not contain. Four decades of spreadsheets is four decades of evidence that the requirement never reached the decision.
The hierarchy that produced them is still in place. What has changed is the cost of it. A platform that ignores operating reality underperforms. An autonomous agent reasoning from the same incomplete model executes against it continuously, at speed, with every indicator on its dashboard improving.
A DEEPER DIVE
What I wrote in 2007, and why it still reads as current
In June 2007, in part six of the Dangerous Supply Chain Myths series, I put the mechanism this way:
Read that again with the current wave in mind, and notice that it is not a complaint about status. It is a claim about sequence.
The operating requirement does not fail to arrive. It arrives after the decision — as implementation feedback, as change management, as an issue log. By then the architecture is fixed, the contract is signed, and the requirement can only be accommodated as an exception rather than incorporated as a condition.
In the same post I gave the thing a name: the hierarchical implementation mechanism, which I described as still firmly entrenched in the marketplace. It is the same arrangement I had walked into in 1983 — by 2007 it had simply acquired an architecture diagram. And I wrote a sentence that reads oddly now for someone who has spent the last month arguing about AI placement:
That was June 2007, about SOA.
The post also carried the sequence rule, stated plainly:
Foundation first, consensus on operating reality second, technology third. Nineteen years before I built a stack to show where AI belongs, the sequencing argument was already on the page.
And it ended with a condition: unless this hierarchical practice is changed. It has not been.
What I asked in 2025, and what today’s post adds to it
In April 2025 I returned to the same problem and asked how strong procurement’s voice and influence had become in ProcureTech selection. I said the answer would determine the trajectory of implementation success for the next decade.
That question still stands. Procurement needs a seat at the table — and so does every other stakeholder whose operating reality bears on whether the decision works. Finance, IT, service delivery, the front line. None of them holds the whole picture, and none of them can be left downstream of the architecture without cost.
What today’s post adds is scope. The 2025 question was asked about one function because that is the function I came from and the one whose exclusion I had watched since 1983. The same exclusion runs across every function whose operating reality is not represented when the architecture is settled.
And the 2007 post had already said what makes a seat useful rather than decorative:
Not when the right people are in the room. When the operating reality has been established. A seat matters because of what the person in it brings, and what they bring only counts if it arrives before the decision rather than after it.
What forty years of spreadsheets actually document
Every spreadsheet running alongside a sanctioned system is a record of the same event: somebody’s operating reality did not fit the model, and they built a private instrument to hold the difference.
(With the emergence of AI, the challenges of spreadsheet independence are significantly compounded by the proliferation of AI model access outside the main enterprise — personal AI is becoming the new shadow spreadsheet.)
Nobody logs that as a finding. It appears as shadow IT, as a data-governance problem, as user resistance, as something to be migrated in the next phase. It is treated as a symptom of poor adoption.
It is the opposite. It is the most reliable evidence available that the declared model and the real operation have diverged, produced continuously, by the people closest to the work, at their own cost.
I have seen the same shape in every case I have traced. Technicians batching parts orders to four in the afternoon because they were rated on calls responded to. Buyers disqualifying a supplier over a port strike the data feed had not heard about. A finance function reconciling part cost and freight as two clean line items with the loss living between them.
None of it was misconduct. All of it was people holding something true that the system had no field for.
And the spreadsheet has a twin that costs more. In that same 2007 post I described a conversation with a senior executive who had two full-time staff dedicated to making a PeopleSoft application work inside their procurement organization — which confined their effort to the application rather than to understanding and refining the practice itself.
I put it this way at the time: the vast majority of purchasing organizations are now in the software business, as they attempt to adapt their practice to an application they would not have chosen to use in the first place.
Those are the two available responses to an inherited system. Build a private instrument beside it, or assign people to bend your practice into it. One is filed as shadow IT. The other is filed as headcount. Neither is filed as evidence that the requirement arrived too late.
Why the 1983 hierarchy is more expensive now than it has ever been
For most of those four decades, the cost of the sequence was underperformance. The platform went in, it did less than promised, the spreadsheets stayed, and the organization absorbed the gap.
Autonomy changes the arithmetic.
An agent given a decision inherits the model it was given. If the operating requirement never entered the architecture, the agent does not know it is missing — and it does not hesitate, does not build a workaround, and does not keep a private spreadsheet. It executes the incomplete model faster and more consistently than any human ever did, and every metric inside its scope improves while it does.
That is why the current failure figures look the way they do. RAND’s study of AI project failure opens by relaying other estimates — that more than 80 percent of AI projects fail, roughly twice the rate of IT projects without AI. RAND is characterising figures it did not measure, so read it as the large majority fail rather than as a percentage.
What RAND did measure is the part that matters. From its own interviews with sixty-five data scientists and engineers, the root causes are overwhelmingly organizational rather than technical: misaligned purpose, weak data foundations, integration treated as an afterthought, fading sponsorship. Not one of those causes sits in the model.
Integration treated as an afterthought. That is the 2007 sentence, in someone else’s words, nineteen years later.
So what would moving past it actually require?
Not a seat at the table. That is the 2025 framing and it accepts the table.
It requires that operating evidence and the technology decision become structurally inseparable — that the architecture cannot be settled without the operating reality having been established first, because the decision depends on it rather than being informed by it.
Three questions that test whether an organization has moved past the 1983 sequence. None of them are technical, and they sit alongside the placement questions in the Human/Tech Stack™.
Was the operating reality established before the architecture, or after? If the requirement document was written from the declared process, the sequence has not changed.
What is the front line currently holding outside the sanctioned system, and why? Every spreadsheet, side database and personal tracker is an answer to a question nobody asked. Find them before deciding what to automate, not during rollout.
And if the operating evidence contradicts the strategy, what happens? If the answer is that it gets logged as a change request, the evidence is downstream. If the answer is that the architecture changes, it is not.
Are we finally ready?
I have been asking a version of this question since before I entered procurement, and the honest answer is that I do not know.
What I do know is that the hierarchy has survived four technology eras — mainframe and ERP, e-procurement, SaaS and cloud, and now AI — and that each era arrived with the claim that this time the technology was different. Each time it was. The sequence was not.
The difference now is that the cost of leaving it alone has changed. An unexamined operating model used to produce a system that underdelivered. It now produces a system that acts on its own, confidently, at scale, from a picture with a department missing from it.
Forty years of spreadsheets were people telling us where those gaps were. We filed it under adoption.
We will know the sequence has changed when the first requirement document gets written from what the front line is holding outside the system, rather than from the process the organization believes it runs.
Jon W. Hansen, FCIPS — Procurement Insights | Hansen Models™ | Independent. Unsponsored. Archive-based.
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Share this:
Like this:
Related