Gartner just published essentially the same capability ladder it released a year ago. The ladder isn’t the problem. The sequence is.
This blog carries the following words as a badge of honor:
Truth Is Believing. Accuracy Is Knowing.
In September 2025, Gartner shared an AI Agent Assessment Framework — a capability ladder running from conventional chatbot up to autonomous agent. In June 2026, the same progression reappeared, now under the banner “the era of agentic AI is here.” As far as the contemporaneous record shows, that is two dated releases of the same ladder. It may go back further still.
Let me be clear about something up front, because it is where most commentary gets this wrong.
The ladder is right.
As a map of what the technology can do, Gartner’s framework is accurate — and it gets more accurate every year. I am not criticizing Gartner, or any analyst, consultancy, or ProcureTech provider, for getting the technology piece right. They have.
What I am saying is that the technology piece is the final piece of the success equation. Not the first.
The capability climbed. The outcomes didn’t.
Set the advancing capability next to the outcomes it was supposed to produce, and the gap is hard to miss — and most of the evidence is Gartner’s own.
Gartner forecast that at least 30% of GenAI projects would be abandoned after proof of concept by the end of 2025, and that over 40% of agentic AI projects will be canceled by 2027 — for the reasons Gartner itself names: escalating cost, inadequate risk controls, unclear business value, and poor data quality. MIT’s 2025 study put 95% of GenAI pilots at no measurable P&L return, and — this is the part worth sitting with — attributed the divide not to model quality, but to approach.
The technology got dramatically more capable. The results did not improve to match. If capability alone determined outcomes, that would be impossible.
Four technology eras, one unsolved variable. The capability curve rises; the determinant of success — the agent-based DND model — has not moved since 1998.
This is not new, and that is the point
I made this same argument in print in October 2006, in Summit magazine. I called the error a “technology-centric approach to a process-driven requirement,” argued for process understanding before technology selection, and named the disconnect between finance, IT, and users as what derails initiatives.
The technology eras since then have changed beyond recognition — e-procurement, SaaS, RPA, GenAI, and now agentic AI. The failure pattern has not. Across every era, regardless of how much the capability advanced, success has continued to elude most organizations. The determining variable was never on the ladder.
The proof runs the other way — and it is twenty-eight years old
In 1998, a Canadian Department of National Defence engagement moved delivery performance from 51% to 97.3% in three months, and held it for roughly seven years, at around 23% cost savings.
The technology was primitive by today’s standard. It succeeded because the substrate was aligned first — the operating conditions, the process, the people — and the technology was the last piece dropped onto an already-aligned foundation. That is why a 1998 tool on an aligned substrate outperformed what 2026 capability routinely fails to deliver.
Technology changes capability. Substrate determines survivability.
The question is no longer whether Gartner is right
It is. The question is when we start sequencing the equation properly.
The substrate first — what I call Phase 0™ — and the technology last. Whatever Gartner, any analyst, any consultancy, or any ProcureTech provider publishes next, success will continue to elude companies until the Phase 0™ substrate issue is addressed. The technology is the final piece. It only pays off on a foundation that was aligned before it arrived.
The capability curve is easy to see — it is right there on the ladder, five neat levels. The organizational curve — incentives, decision rights, process alignment, the conditions that decide whether any of it lands — is the one missing from the diagram. It has been missing for twenty years. The DND result (and others like Virginia’s eVA) shows what happens when you put it back.
“But don’t we already do this?”
It’s a fair question, and the one most practitioners will raise here. Most organizations do run a discovery phase — process mapping, stakeholder interviews, requirements gathering. So how is Phase 0™ different?
Look at what conventional discovery is for. It’s run to configure a technology that has already been chosen. The mapping documents how the organization works so the tool can be set up to match it. The interviews feed a configuration document. That isn’t Phase 0™ — it’s implementation prep, and it still puts technology first.
Phase 0™ runs before the technology decision, and it asks a different question. Not “how do we work, so we can configure the tool?” but “will the way we work actually deliver the outcome — and what has to change before any tool can help?”
At DND, conventional process mapping would have documented the existing order flow and automated it faithfully — and preserved the 51%. The variable that moved delivery to 97.3% wasn’t in the process map; it was in whether the process should have worked that way at all. Standard discovery captures what is, so the technology can replicate it. Phase 0™ interrogates whether what is can succeed — and fixes it first, which sometimes means discovering you don’t need the system you were about to buy.
That’s the difference: not whether you look at process, but when and why. Before the technology decision, to determine whether any initiative can land — versus after it, in service of the tool already selected. One serves the technology. The other determines whether the technology has a chance.
Today’s takeaway: The high rate of initiative failure across four decades and every technology era has never been a technology problem. It has been, and remains, a substrate problem. Address it first, and the technology — whichever era’s technology it happens to be — finally does what it promised.
Jon Hansen is the founder of Hansen Models™ and publisher of Procurement Insights, an independent research practice operating on a nineteen-year archive with zero vendor sponsorships.
-30-
Related
Technology Is the Final Piece — Not the First
Posted on June 5, 2026
0
Gartner just published essentially the same capability ladder it released a year ago. The ladder isn’t the problem. The sequence is.
This blog carries the following words as a badge of honor:
Truth Is Believing. Accuracy Is Knowing.
In September 2025, Gartner shared an AI Agent Assessment Framework — a capability ladder running from conventional chatbot up to autonomous agent. In June 2026, the same progression reappeared, now under the banner “the era of agentic AI is here.” As far as the contemporaneous record shows, that is two dated releases of the same ladder. It may go back further still.
Let me be clear about something up front, because it is where most commentary gets this wrong.
The ladder is right.
As a map of what the technology can do, Gartner’s framework is accurate — and it gets more accurate every year. I am not criticizing Gartner, or any analyst, consultancy, or ProcureTech provider, for getting the technology piece right. They have.
What I am saying is that the technology piece is the final piece of the success equation. Not the first.
The capability climbed. The outcomes didn’t.
Set the advancing capability next to the outcomes it was supposed to produce, and the gap is hard to miss — and most of the evidence is Gartner’s own.
Gartner forecast that at least 30% of GenAI projects would be abandoned after proof of concept by the end of 2025, and that over 40% of agentic AI projects will be canceled by 2027 — for the reasons Gartner itself names: escalating cost, inadequate risk controls, unclear business value, and poor data quality. MIT’s 2025 study put 95% of GenAI pilots at no measurable P&L return, and — this is the part worth sitting with — attributed the divide not to model quality, but to approach.
The technology got dramatically more capable. The results did not improve to match. If capability alone determined outcomes, that would be impossible.
Four technology eras, one unsolved variable. The capability curve rises; the determinant of success — the agent-based DND model — has not moved since 1998.
This is not new, and that is the point
I made this same argument in print in October 2006, in Summit magazine. I called the error a “technology-centric approach to a process-driven requirement,” argued for process understanding before technology selection, and named the disconnect between finance, IT, and users as what derails initiatives.
The technology eras since then have changed beyond recognition — e-procurement, SaaS, RPA, GenAI, and now agentic AI. The failure pattern has not. Across every era, regardless of how much the capability advanced, success has continued to elude most organizations. The determining variable was never on the ladder.
The proof runs the other way — and it is twenty-eight years old
In 1998, a Canadian Department of National Defence engagement moved delivery performance from 51% to 97.3% in three months, and held it for roughly seven years, at around 23% cost savings.
The technology was primitive by today’s standard. It succeeded because the substrate was aligned first — the operating conditions, the process, the people — and the technology was the last piece dropped onto an already-aligned foundation. That is why a 1998 tool on an aligned substrate outperformed what 2026 capability routinely fails to deliver.
Technology changes capability. Substrate determines survivability.
The question is no longer whether Gartner is right
It is. The question is when we start sequencing the equation properly.
The substrate first — what I call Phase 0™ — and the technology last. Whatever Gartner, any analyst, any consultancy, or any ProcureTech provider publishes next, success will continue to elude companies until the Phase 0™ substrate issue is addressed. The technology is the final piece. It only pays off on a foundation that was aligned before it arrived.
The capability curve is easy to see — it is right there on the ladder, five neat levels. The organizational curve — incentives, decision rights, process alignment, the conditions that decide whether any of it lands — is the one missing from the diagram. It has been missing for twenty years. The DND result (and others like Virginia’s eVA) shows what happens when you put it back.
“But don’t we already do this?”
It’s a fair question, and the one most practitioners will raise here. Most organizations do run a discovery phase — process mapping, stakeholder interviews, requirements gathering. So how is Phase 0™ different?
Look at what conventional discovery is for. It’s run to configure a technology that has already been chosen. The mapping documents how the organization works so the tool can be set up to match it. The interviews feed a configuration document. That isn’t Phase 0™ — it’s implementation prep, and it still puts technology first.
Phase 0™ runs before the technology decision, and it asks a different question. Not “how do we work, so we can configure the tool?” but “will the way we work actually deliver the outcome — and what has to change before any tool can help?”
At DND, conventional process mapping would have documented the existing order flow and automated it faithfully — and preserved the 51%. The variable that moved delivery to 97.3% wasn’t in the process map; it was in whether the process should have worked that way at all. Standard discovery captures what is, so the technology can replicate it. Phase 0™ interrogates whether what is can succeed — and fixes it first, which sometimes means discovering you don’t need the system you were about to buy.
That’s the difference: not whether you look at process, but when and why. Before the technology decision, to determine whether any initiative can land — versus after it, in service of the tool already selected. One serves the technology. The other determines whether the technology has a chance.
Today’s takeaway: The high rate of initiative failure across four decades and every technology era has never been a technology problem. It has been, and remains, a substrate problem. Address it first, and the technology — whichever era’s technology it happens to be — finally does what it promised.
Jon Hansen is the founder of Hansen Models™ and publisher of Procurement Insights, an independent research practice operating on a nineteen-year archive with zero vendor sponsorships.
-30-
Share this:
Like this:
Related