Jon W. Hansen, FCIPS · Procurement Insights | Hansen AI Labs™ · October 2026
For years, Phase 0™ was the stumbling block. It was seen as a delay to the sales and implementation process: a pause standing between the decision and the deployment. Yet with Phase 0™, I would rarely, if ever, expect an enterprise to abandon an initiative. A properly conducted Phase 0™ will prevent an unsuitable initiative from proceeding, or reframe it before significant deployment capital is committed. Once the initiative is deployed, the Firm Fence keeps a valid initiative aligned with operating reality as that reality changes. The initiative would end only if the business purpose itself ceased to exist. Today, the end client is asking for that pause. Skipping the perceived delay now carries a more immediate and far-reaching cost than the delay ever did. Phase 0™ is the necessary short pause executives are requesting.
A CIO’s question
Leena Sharma, Global CIO at Enerpac Tool Group, came back from Gartner IT Symposium/Xpo and shared her takeaways. Her conclusion: most AI struggles are not about technology. They are about discipline. One figure she reported says it all: 71% of organizations plan to deploy AI agents this year, but only 21% have the governance to manage them.
Then she asked: What’s your rule for deciding when to stop an AI initiative?
My answer starts with a case from 1998.
Where the problem shows up is rarely where it starts
In 1998, the services provider managing the Department of National Defence’s MRO procurement was about to lose a multimillion-dollar contract. Next-day delivery, and with it the ability to close service calls, stood at 51% against a contract that required 90%.
Everyone could see where the flow was breaking: procurement. So my assignment was to automate procurement. That is what almost everyone was doing at the time, and it is still the most common response today. Find the point where the flow appears to break, and put better technology there. That is the tech-first, equation-based response.
I had learned long before that the place where a problem surfaces is usually not the place where it starts. So instead of mapping forward to the choke point, I traced backward from the failure, and asked a question that had nothing to do with procurement: what time of day do the orders come in, and what happens as a result?
There was a policy in place telling the service technicians to order spare parts immediately after each service call. But the technicians were measured on call response: get to the customer fast, and complete as many calls as possible. Stopping to submit a part order after each call meant fewer calls in a day. So they held their orders and submitted them together at about four in the afternoon.
Did procurement still receive the orders before the end of the business day? Yes. So the conclusion everyone reached was that procurement needed a faster ordering system. In an equation-based world, from point A to point B, that makes perfect sense.
But the operation did not run on a single strand.
The same six functions: one drawn as a sequence, one as it operates. Illustrative, not measured data.
Discovering the unintended consequences, and the hidden strands
By using an agent-based model rather than a tech-led, equation-based map, I identified several hidden strands.
The first was cost. A part that cost $100 when it was ordered at 9:00 a.m. could cost as much as $900 when it was ordered at 4:00 p.m.
The second was delivery. Eighty percent of the parts were sourced from suppliers in the United States, and those suppliers were not receiving the orders until after 4:00 p.m. From that point, three more strands came into play: supplier inexperience, courier cutoff times and customs clearance requirements. Together, they produced the 51% next-day delivery performance.
When I raised this, I received some strange looks, and the same question: why not just automate procurement so it can process the orders faster? My answer was this. You are about to lose a multimillion-dollar contract. If you buy and implement a new and better ordering platform, six to nine months from now you will be right back where you started. You do not have a technology problem. You have an agent visibility problem. And it does not involve only your internal agents, service and purchasing. Your suppliers, couriers and customs are agents too, and each of them crosses your operation in ways you are not seeing.
That is what Phase 0™ caught: the shadow behind the choke point, the orders being held until four o’clock.
When misaligned incentives create the boomerang effect
I won’t go into the rest of the mechanics of what became the Phase 0™ model we use today. But it is worth revisiting the technicians’ call-response incentive, because it shows why change management is a byproduct of a misunderstood process, rather than the solution for it.
Up front, the incentive worked. Customers were happy, because their calls were responded to quickly and very few had to wait for someone to look at their problem.
But a quick diagnosis does not fix the original problem. It only identifies it. The fix still depended on the part arriving.
Phase 0™ showed that the source of the delay was not procurement. It was the technicians meeting their response incentive. And that incentive came back around like a boomerang. Orders held until four o’clock meant parts arriving late, and 51% next-day delivery meant the technicians could not close the calls they had opened so quickly. Call close rates were the critical result, and they were putting the contract at risk.
That is why equation-based mapping would have had the service provider spending time and money to automate procurement, only to end up right back at 51%.
There is much more to this story, including how Phase 0™ identified the loopback learning mechanisms for recognizing and coordinating multiple strands at once. That is a post of its own.
Strand Commonality™ in one sentence
Strand Commonality™ holds that seemingly unrelated strands have unique operating attributes that collectively contribute to a single outcome.
I first became aware of it in 1983, programming in dBase II, when a new and better line of code would break another line that had been working perfectly. I formulated it as a theory in the late 1990s, in research funded through Canada’s Scientific Research and Experimental Development (SR&ED) program, and applied it to the DND operation. Every agent, human or machine, is a strand. The outcome is decided where they cross.
Two different stop decisions
That brings me back to Leena’s question, because it contains two decisions.
When should the AI stop acting? The AI works inside what I call a Firm Fence: fixed limits on what it is for, what results it must produce, and what it is allowed to decide on its own. It pauses or changes course in three situations:
The result it produces falls outside the range it was expected to produce.
The next action would go beyond the authority it was given.
New information shows something affecting the outcome that the system was not accounting for: a hidden strand, like the four o’clock orders.
When that happens, the system does not shut down. It works out why the result changed, adds what it learned to the next decision, and either adjusts within its authority or hands the decision to a person. In 1998, when a storm threatened a supplier’s region, the system flagged that supplier’s ranking and put the choice to the buyer, who recorded the reason for the decision.
When should the enterprise stop the initiative? Almost never, if Phase 0™ was done properly, because an unsuitable initiative is stopped or reframed before significant money is committed. After that, the initiative should end only if its business purpose no longer exists.
Why executives now ask for the pause
For years, Phase 0™ was the step most often negotiated away, because it was seen as slowing the sale and delaying the start date.
AI has changed the cost of skipping it. In 1998, automating procurement would have taken six to nine months to bring the operation back to where it started. Today, an AI agent acting on the same incomplete picture can spread the same failure across plants, supply chains, customers and finances in days. That is the real meaning of Leena’s 71% versus 21%.
Governance before scale is the right instinct. But governance needs something to govern against: a clear view of how the operation actually runs, who the agents are, and where their strands cross. That is what Phase 0™ provides.
The market’s view of Phase 0™ has shifted: it is no longer an unwanted delay, but a necessary pause to prevent initiative failure.
Phase 0™ Was Once Seen as a Delay. Today It Is the Short Pause Executives Are Asking For.
Posted on October 28, 2026
0
Jon W. Hansen, FCIPS · Procurement Insights | Hansen AI Labs™ · October 2026
For years, Phase 0™ was the stumbling block. It was seen as a delay to the sales and implementation process: a pause standing between the decision and the deployment. Yet with Phase 0™, I would rarely, if ever, expect an enterprise to abandon an initiative. A properly conducted Phase 0™ will prevent an unsuitable initiative from proceeding, or reframe it before significant deployment capital is committed. Once the initiative is deployed, the Firm Fence keeps a valid initiative aligned with operating reality as that reality changes. The initiative would end only if the business purpose itself ceased to exist. Today, the end client is asking for that pause. Skipping the perceived delay now carries a more immediate and far-reaching cost than the delay ever did. Phase 0™ is the necessary short pause executives are requesting.
A CIO’s question
Leena Sharma, Global CIO at Enerpac Tool Group, came back from Gartner IT Symposium/Xpo and shared her takeaways. Her conclusion: most AI struggles are not about technology. They are about discipline. One figure she reported says it all: 71% of organizations plan to deploy AI agents this year, but only 21% have the governance to manage them.
Then she asked: What’s your rule for deciding when to stop an AI initiative?
My answer starts with a case from 1998.
Where the problem shows up is rarely where it starts
In 1998, the services provider managing the Department of National Defence’s MRO procurement was about to lose a multimillion-dollar contract. Next-day delivery, and with it the ability to close service calls, stood at 51% against a contract that required 90%.
Everyone could see where the flow was breaking: procurement. So my assignment was to automate procurement. That is what almost everyone was doing at the time, and it is still the most common response today. Find the point where the flow appears to break, and put better technology there. That is the tech-first, equation-based response.
I had learned long before that the place where a problem surfaces is usually not the place where it starts. So instead of mapping forward to the choke point, I traced backward from the failure, and asked a question that had nothing to do with procurement: what time of day do the orders come in, and what happens as a result?
There was a policy in place telling the service technicians to order spare parts immediately after each service call. But the technicians were measured on call response: get to the customer fast, and complete as many calls as possible. Stopping to submit a part order after each call meant fewer calls in a day. So they held their orders and submitted them together at about four in the afternoon.
Did procurement still receive the orders before the end of the business day? Yes. So the conclusion everyone reached was that procurement needed a faster ordering system. In an equation-based world, from point A to point B, that makes perfect sense.
But the operation did not run on a single strand.
The same six functions: one drawn as a sequence, one as it operates. Illustrative, not measured data.
Discovering the unintended consequences, and the hidden strands
By using an agent-based model rather than a tech-led, equation-based map, I identified several hidden strands.
The first was cost. A part that cost $100 when it was ordered at 9:00 a.m. could cost as much as $900 when it was ordered at 4:00 p.m.
The second was delivery. Eighty percent of the parts were sourced from suppliers in the United States, and those suppliers were not receiving the orders until after 4:00 p.m. From that point, three more strands came into play: supplier inexperience, courier cutoff times and customs clearance requirements. Together, they produced the 51% next-day delivery performance.
When I raised this, I received some strange looks, and the same question: why not just automate procurement so it can process the orders faster? My answer was this. You are about to lose a multimillion-dollar contract. If you buy and implement a new and better ordering platform, six to nine months from now you will be right back where you started. You do not have a technology problem. You have an agent visibility problem. And it does not involve only your internal agents, service and purchasing. Your suppliers, couriers and customs are agents too, and each of them crosses your operation in ways you are not seeing.
That is what Phase 0™ caught: the shadow behind the choke point, the orders being held until four o’clock.
When misaligned incentives create the boomerang effect
I won’t go into the rest of the mechanics of what became the Phase 0™ model we use today. But it is worth revisiting the technicians’ call-response incentive, because it shows why change management is a byproduct of a misunderstood process, rather than the solution for it.
Up front, the incentive worked. Customers were happy, because their calls were responded to quickly and very few had to wait for someone to look at their problem.
But a quick diagnosis does not fix the original problem. It only identifies it. The fix still depended on the part arriving.
Phase 0™ showed that the source of the delay was not procurement. It was the technicians meeting their response incentive. And that incentive came back around like a boomerang. Orders held until four o’clock meant parts arriving late, and 51% next-day delivery meant the technicians could not close the calls they had opened so quickly. Call close rates were the critical result, and they were putting the contract at risk.
That is why equation-based mapping would have had the service provider spending time and money to automate procurement, only to end up right back at 51%.
There is much more to this story, including how Phase 0™ identified the loopback learning mechanisms for recognizing and coordinating multiple strands at once. That is a post of its own.
Strand Commonality™ in one sentence
Strand Commonality™ holds that seemingly unrelated strands have unique operating attributes that collectively contribute to a single outcome.
I first became aware of it in 1983, programming in dBase II, when a new and better line of code would break another line that had been working perfectly. I formulated it as a theory in the late 1990s, in research funded through Canada’s Scientific Research and Experimental Development (SR&ED) program, and applied it to the DND operation. Every agent, human or machine, is a strand. The outcome is decided where they cross.
Two different stop decisions
That brings me back to Leena’s question, because it contains two decisions.
When should the AI stop acting? The AI works inside what I call a Firm Fence: fixed limits on what it is for, what results it must produce, and what it is allowed to decide on its own. It pauses or changes course in three situations:
When that happens, the system does not shut down. It works out why the result changed, adds what it learned to the next decision, and either adjusts within its authority or hands the decision to a person. In 1998, when a storm threatened a supplier’s region, the system flagged that supplier’s ranking and put the choice to the buyer, who recorded the reason for the decision.
When should the enterprise stop the initiative? Almost never, if Phase 0™ was done properly, because an unsuitable initiative is stopped or reframed before significant money is committed. After that, the initiative should end only if its business purpose no longer exists.
Why executives now ask for the pause
For years, Phase 0™ was the step most often negotiated away, because it was seen as slowing the sale and delaying the start date.
AI has changed the cost of skipping it. In 1998, automating procurement would have taken six to nine months to bring the operation back to where it started. Today, an AI agent acting on the same incomplete picture can spread the same failure across plants, supply chains, customers and finances in days. That is the real meaning of Leena’s 71% versus 21%.
Governance before scale is the right instinct. But governance needs something to govern against: a clear view of how the operation actually runs, who the agents are, and where their strands cross. That is what Phase 0™ provides.
The market’s view of Phase 0™ has shifted: it is no longer an unwanted delay, but a necessary pause to prevent initiative failure.
For the full Firm Fence rules, see Phase 0 Can Be Delayed. It Cannot Be Escaped: The Firm Fence Rules AI Deployments Keep Skipping (15 October 2026).
Truth Is Believing. Accuracy Is Knowing. Outcome Is Proof.™
-30-
Share this:
Like this:
Related