When to Kill an AI Initiative (And When to Persevere Through the Valley)
Created on 2026-02-06 09:44
Published on 2026-03-15 10:00
The hardest decision in AI transformation and a framework for making it
The CFO wants answers.
Six months into the AI initiative, the numbers are disappointing. Productivity has not improved. Adoption is below target. The promised benefits have not materialized.
The CFO asks the question that nobody wants to answer: Should we kill this?
This is the moment where most organizations fail.
Some organizations kill initiatives that were on track to succeed. They mistake the J-Curve valley for genuine failure. They give up at month six when success would have arrived at month nine.
Some organizations persevere with initiatives that will never succeed. They interpret sunk costs as reason to continue. They hope that more time will somehow produce results that were never possible.
Both mistakes are expensive. The first wastes the investment already made and damages credibility for future initiatives. The second compounds bad investment with more bad investment.
This article provides a framework for the hardest decision in AI transformation: when to kill and when to persevere.
Why This Decision Is So Hard
Before offering the framework, let me acknowledge why this decision is genuinely difficult.
The J-Curve obscures reality.
The J-Curve means productivity declines before it improves. This is normal for successful initiatives.
But it is also normal for failing initiatives. They also show early decline. The difference is that they never recover.
Early decline cannot distinguish between success-in-progress and failure-in-progress. The J-Curve creates a period of genuine uncertainty.
Sunk costs distort judgment.
You have invested significantly. Money, time, attention, political capital.
Killing the initiative means acknowledging that investment was wasted. Human psychology resists this acknowledgment. We prefer to believe our investments will pay off.
Sunk cost fallacy leads to throwing good money after bad. “We have invested too much to stop now.”
Sponsors protect their initiatives.
Every initiative has sponsors who have staked credibility on success.
Sponsors have incentives to present optimistic interpretations. To hide problems. To delay the decision to kill.
The information you receive about initiative status may be filtered through sponsor self-interest.
Success is uncertain, failure is certain.
If you kill an initiative, failure is certain. The investment is lost. The benefits will not come.
If you continue, success remains possible. Maybe things will improve. Maybe the problems are temporary.
This asymmetry biases toward continuation. Certain failure feels worse than possible success, even when the probability of success is low.
Nobody wants to be the executioner.
Killing an initiative is unpleasant. It means admitting failure. It means disappointing people. It may mean ending someone’s project, team, or career trajectory.
The human desire to avoid unpleasant tasks biases toward delay. “Let’s give it another quarter.”
The J-Curve Is Not a Guarantee
Let me be clear about something important.
The J-Curve describes successful initiatives. Productivity declines, then improves, then exceeds baseline.
But the J-Curve does not mean every initiative will eventually succeed. Some initiatives decline and keep declining. Some initiatives decline and flatten. Some initiatives were never going to work.
The J-Curve creates a valley that successful initiatives pass through. It is not a guarantee that every initiative in the valley will emerge successfully.
The question is: How do you distinguish between initiatives that are in the valley on the way to success and initiatives that are in the valley on the way to failure?
The Warning Signs That Suggest Killing
Certain patterns suggest an initiative is not going to succeed. When you see these patterns, killing becomes the right decision.
Warning Sign 1: Declining adoption.
Healthy initiatives show increasing adoption over time. More people using AI. More frequent usage. More diverse applications.
Unhealthy initiatives show declining adoption. Fewer people using AI. Less frequent usage. Retreat to only the earliest adopters.
Declining adoption means people who tried AI are choosing not to continue. They have concluded it does not help them.
This is a fundamental signal. If people who use AI conclude it is not valuable, no amount of training or encouragement will change that. The initiative is failing.
Warning Sign 2: Negative quality indicators.
Healthy initiatives show AI-assisted outcomes that are better than non-assisted outcomes. Faster. More accurate. Higher quality.
Unhealthy initiatives show AI-assisted outcomes that are worse. Slower when learning time is included. Less accurate. Lower quality.
If AI is making work worse rather than better, the value thesis is broken. The initiative is failing.
Warning Sign 3: Sponsor abandonment.
Healthy initiatives have engaged sponsors. Sponsors who pay attention. Sponsors who remove obstacles. Sponsors who fight for resources.
Unhealthy initiatives have disengaged sponsors. Sponsors who have moved on to other priorities. Sponsors who are not returning calls. Sponsors who are positioning themselves for distance from failure.
Sponsors have information about initiative prospects. When sponsors abandon, they are signaling what they believe. Follow their behavior, not their words.
Warning Sign 4: Persistent blockers.
Healthy initiatives encounter blockers and resolve them. Data access problems are solved. Integration issues are addressed. Resistance is overcome.
Unhealthy initiatives have the same blockers at month six that they had at month one. Nothing has been resolved. The same obstacles prevent progress.
If blockers have persisted for months without resolution, they are probably not going to be resolved. The organizational capability or will to resolve them does not exist.
Warning Sign 5: Strategic context has changed.
Sometimes the business reason for an initiative no longer holds.
The customer problem the AI was meant to solve is no longer a priority. The process the AI was meant to improve is being redesigned or eliminated. The competitive threat the AI was meant to address has changed.
When strategic context shifts, initiatives designed for the old context may no longer make sense, even if they were on track to succeed in that old context.
Warning Sign 6: Team morale collapse.
Healthy initiatives have teams that believe in what they are doing. Even when things are hard, they see a path to success.
Unhealthy initiatives have teams that have lost belief. They are going through the motions. They no longer think success is possible. They are waiting to be assigned to something else.
Team morale is information. Teams close to the work often know before leadership whether something is going to work.
The Signs That Suggest Perseverance
Other patterns suggest an initiative is on track despite disappointing current metrics. When you see these patterns, perseverance is the right decision.
Perseverance Sign 1: Growing adoption despite challenges.
Even when productivity metrics are negative, growing adoption is a positive signal.
More people trying AI. More frequent usage. Expansion beyond initial users to new groups.
Growing adoption means people are finding value even if that value is not yet appearing in aggregate metrics. This is a leading indicator of eventual success.
Perseverance Sign 2: Improving quality over time.
Even when quality is not yet above baseline, improvement is a positive signal.
Week over week, are AI-assisted outcomes getting better? Is accuracy improving? Is speed increasing?
Improvement means learning is happening. Learning means the curve will eventually inflect.
Perseverance Sign 3: Leading indicators are positive while lagging indicators are negative.
This is the classic J-Curve pattern.
Adoption is growing. Engagement is increasing. Capability is developing. But productivity has not yet improved.
When leading indicators are positive, give lagging indicators time to follow.
Perseverance Sign 4: Blockers are being resolved.
The initiative encounters obstacles, but obstacles are being addressed.
Data access problems that existed at month one are solved by month three. Integration issues are being worked. Resistance is being overcome.
Progress on blockers indicates the organization is capable of making this work.
Perseverance Sign 5: Team believes despite difficulty.
The team is working hard. Things are challenging. But they believe success is possible.
Team belief is not naivete. Teams close to the work have information that leadership lacks. When they believe, it often means they see something real.
Perseverance Sign 6: Learning is accumulating.
Even when outcomes are disappointing, the organization is learning.
Learning what works. Learning what does not. Learning about the organization, the data, the processes.
Accumulated learning has value even if the specific initiative changes. It accelerates future initiatives.
The Decision Framework
Let me synthesize these signals into a practical decision framework.
At defined checkpoints, evaluate honestly:
Establish evaluation checkpoints in advance. Month three. Month six. Month nine.
At each checkpoint, gather data systematically. Do not rely on sponsor reports alone. Collect information from multiple sources.
Assess warning signs and perseverance signs:
How many warning signs are present? How severe are they?
How many perseverance signs are present? How strong are they?
Create an honest picture of initiative health.
Apply the decision matrix:
Strong warning signs + weak perseverance signs = Kill.
The initiative is failing and shows no signs of recovery. Continued investment is throwing good money after bad.
Weak warning signs + strong perseverance signs = Persevere.
The initiative shows signs of eventual success despite current challenges. Give it time.
Mixed signals = Investigate deeper.
When signals are mixed, you need more information. Dig deeper into what is actually happening. Do not decide based on ambiguous signals.
Strong signals in both directions = Restructure.
Sometimes initiatives have genuine strengths and genuine problems. Consider restructuring rather than binary kill or continue.
Weight leading indicators more heavily than lagging indicators during the expected valley period:
If you expected the J-Curve to inflect at month nine, evaluate primarily on leading indicators until month nine.
Leading indicators predict. Lagging indicators confirm.
During the valley, leading indicators tell you whether to expect recovery. Lagging indicators only confirm what you already knew: you are in the valley.
Make the decision, do not defer it:
When the evidence suggests killing, kill.
Deferring the decision to kill often means continuing to invest while the probability of success declines.
“Let’s give it one more quarter” is often code for “I do not want to make this decision.” Make the decision.
How to Kill Gracefully
When you decide to kill, how you kill matters.
Preserve learning:
Before you kill, capture what you learned.
What worked? What did not? What did you discover about your organization, your data, your processes, your people?
This learning has value even if the initiative failed. It accelerates future initiatives. It prevents repeating mistakes.
Create a formal post-mortem. Document findings. Share them broadly.
Preserve relationships:
Do not blame individuals for organizational decisions.
The decision to kill is an organizational decision. It reflects initiative design, resource allocation, and strategic context, not individual failure.
Protect the people who worked on the initiative. Thank them for their effort. Assign them to other work without stigma.
Blaming individuals for killed initiatives makes future initiative work impossible. Nobody will take risks if risk-taking is punished.
Preserve credibility:
Acknowledge what happened honestly.
“This initiative did not produce the results we expected. Here is what we learned. Here is what we will do differently next time.”
Honest acknowledgment builds credibility. Pretending the initiative succeeded when it did not destroys credibility.
Do not spin the failure as success. Do not hide the kill as a “strategic pivot.” Be direct.
Preserve capability:
Killing an initiative does not mean abandoning AI entirely.
Distinguish between “this specific initiative did not work” and “AI does not work for us.”
The first is a learning that informs future initiatives. The second is a conclusion that may not be warranted.
Make clear that the organization remains committed to AI readiness even though this specific initiative is ending.
How to Persevere Effectively
When you decide to persevere, how you persevere matters.
Address the problems:
Perseverance is not passive continuation. It is active problem-solving.
What obstacles remain? Address them.
What capabilities are missing? Develop them.
What support is lacking? Provide it.
Perseverance without action is just hoping things get better. Things do not get better without action.
Manage the narrative:
During the valley, narrative matters.
Explain to stakeholders what is happening. “We are in the expected valley period. Leading indicators suggest we are on track for recovery. Here is what we are seeing.”
Narrative management is not spin. It is education. Stakeholders who understand the J-Curve can support perseverance. Stakeholders who do not understand will pressure for premature termination.
Protect resources:
During the valley, initiatives are vulnerable to resource reallocation.
Budget cuts. Staff reassignments. Attention diversion.
Fight for the resources needed to reach the inflection point. Make the case for continued investment. Do not allow the initiative to be starved.
Set a time limit:
Perseverance is not indefinite.
Set a clear timeline. “If we do not see productivity improvement by month twelve, we will reevaluate.”
Time limits create accountability. They prevent indefinite perseverance with failing initiatives. They force eventual decision.
The Sunk Cost Trap
Let me address sunk costs directly.
You have invested significant resources. The initiative is not working. Should you continue because of what you have already invested?
No.
Sunk costs are sunk. They are gone regardless of what you do next. The decision should be based on future value, not past investment.
The question is not: “How much have we invested?”
The question is: “Given what we now know, would we invest more to continue?”
If the answer is no, stop. The sunk cost is irrelevant.
If the answer is yes, continue. But make sure the answer is actually yes based on future prospects, not past investment.
How sunk cost thinking sounds:
“We have invested too much to stop now.”
“We cannot waste all that investment.”
“We just need a little more to get over the finish line.”
“All that work will be for nothing if we stop.”
These statements focus on past investment, not future value. They are sunk cost traps.
How future-value thinking sounds:
“Given what we now know about this initiative, what is the probability of success?”
“If we invest X more, what is the expected return?”
“Is this the best use of additional resources, or would those resources produce more value elsewhere?”
These statements focus on future value. They lead to better decisions.
The Restart Option
Sometimes the right answer is neither kill nor persevere. It is restart.
When restart makes sense:
The concept was right but the execution was wrong. A different approach might succeed where this approach failed.
The learning accumulated is valuable, and a restructured initiative could leverage that learning.
The blockers have been identified and could be addressed in a redesigned initiative.
The team has developed capability that would be wasted by killing entirely.
How restart differs from perseverance:
Restart involves explicit acknowledgment that the current approach is not working.
Restart involves redesign, not just continued effort.
Restart may involve changed scope, changed team, changed approach.
Restart is a new initiative informed by the old one, not continuation of the old one.
The restart conversation:
“This initiative as designed is not working. We recommend terminating the current initiative but launching a new initiative that incorporates what we learned. Here is how the new initiative will be different. Here is why we believe it can succeed where the current approach could not.”
This is more honest than pretending the current approach will eventually work. It is more productive than killing entirely.
The Decision Conversation
Let me give you language for the conversation with leadership when you recommend killing.
Opening:
“I need to recommend that we terminate the AI initiative. This is not a recommendation I make lightly, and I want to explain the reasoning.”
The evidence:
“When I look at the data, I see several concerning patterns. Adoption has declined from peak of 45% to current 28%. People who tried the AI are choosing not to continue. Quality metrics show AI-assisted work is still below non-assisted work after six months. And the blockers we identified in month one remain unresolved.”
The distinction from J-Curve:
“I know we discussed the J-Curve and the expectation of early decline. But the pattern we are seeing is not consistent with eventual recovery. The leading indicators that would predict recovery, growing adoption, improving quality, resolving blockers, are not present.”
The recommendation:
“Based on this evidence, I recommend we terminate the initiative, preserve the learning through formal post-mortem, and consider a restructured initiative that addresses what we have learned.”
The ask:
“I am asking for your support on this decision. It is not a decision I wanted to make, but I believe it is the right one.”
Knowing when to kill and when to persevere is the hardest decision in AI transformation.
Both mistakes are expensive. Killing too early wastes investment and damages credibility. Persevering too long compounds bad investment.
The J-Curve creates genuine uncertainty. Early decline does not distinguish success-in-progress from failure-in-progress.
But there are signals. Warning signs that suggest killing. Perseverance signs that suggest continuing. A framework for making the decision.
Use the framework. Make the decision. Do not defer.
The courage to kill failing initiatives preserves resources and credibility for initiatives that can succeed.
Have you faced this decision? What signals helped you decide?
The AI Readiness Scorecard helps you assess initiative readiness before investment, reducing the probability that you face the kill decision later.
Comment “SCORECARD” below and I will send you access.
The hardest decision is sometimes the right decision. Have the courage to make it.