The J-Curve: Why AI Investments Look Like Failures Before They Succeed
Created on 2026-02-06 09:06
Published on 2026-02-17 09:30
What every CFO needs to understand about the productivity dip that kills promising initiatives
Three months into your AI initiative, your CFO walks into your office with a spreadsheet.
The numbers are not good.
Productivity is down. The metrics you promised to improve have gotten worse. The team is spending more time on AI-related tasks than on their actual work. The investment is not paying off.
The CFO asks the reasonable question: Should we pull the plug?
If you say yes, you will kill an initiative that was three months away from breakthrough results.
If you say no without understanding why the numbers look bad, you will lose credibility and possibly budget.
This is the moment where most AI initiatives die. Not because they were failing. Because they looked like they were failing.
Welcome to the J-Curve.
The Shape of Transformation
Economist Erik Brynjolfsson and his colleagues at Stanford have studied what happens when organizations adopt transformative technologies.
Their finding challenges everything we assume about technology investments.
Productivity does not improve immediately. It declines first.
When you plot productivity over time after AI deployment, the line does not go up. It goes down, then up. The shape looks like the letter J.
In the early months, productivity falls. People are learning new tools. Processes are being adapted. Mistakes are being made. The organization is investing effort that has not yet paid off.
Then, at some point, the curve inflects. Productivity starts rising. Eventually, it rises above where it started. The investment pays off.
But between deployment and payoff, there is a valley. Brynjolfsson’s research suggests this valley can be deep. In organizations that are not prepared, productivity can decline by up to 60 percentage points before recovering.
Sixty percent.
That is not a small dip. That is a crater.
Why the J-Curve Exists
The J-Curve is not a flaw in AI technology. It is a feature of how organizations absorb change.
Learning takes time.
People cannot master new tools instantly. Even intuitive AI applications require learning. What prompts work? How do you evaluate outputs? When should you trust the AI and when should you override it?
This learning happens while people are also trying to do their regular work. Attention is divided. Mistakes happen. Productivity suffers.
The learning investment is real. It has real costs. Those costs appear before the benefits.
Processes must adapt.
Existing workflows were designed for human-only work. When AI enters, workflows need adjustment.
Where does AI output go? Who reviews it? What happens when AI is wrong? How do handoffs work?
These questions get answered through trial and error. The trial and error period is messy. Productivity suffers while answers emerge.
Integration is imperfect.
AI systems do not plug seamlessly into existing technology infrastructure. Data must flow. Systems must connect. Outputs must be formatted correctly.
Initial integration is never perfect. Problems surface. Fixes take time. Workarounds create inefficiency.
The integration investment is real. It appears as cost before it enables benefit.
Behavior change is slow.
People have habits. They have comfortable ways of working. AI requires new habits.
Behavior change does not happen by decree. It happens through practice, reinforcement, and social proof. Until new behaviors are established, old behaviors persist alongside new tools.
The gap between having AI and using AI well is where productivity suffers.
The Valley of Death
I call the bottom of the J-Curve the Valley of Death.
This is where AI initiatives go to die.
The Valley of Death is typically around month three to month six after deployment. The initial excitement has faded. The learning investment is accumulating. The integration problems are surfacing. The productivity decline is visible.
And the benefits have not yet appeared.
This is when CFOs ask hard questions. This is when sponsors lose confidence. This is when budgets get cut.
The tragedy is that many initiatives killed in the Valley of Death were months away from success.
They were not failing. They were following the predictable pattern of transformation. But because nobody expected the J-Curve, nobody planned for it.
The initiative died not from failure but from impatience.
What the 95% Do Wrong
MIT’s research found that 95% of organizations get zero return from AI investments.
A significant portion of these failures happen in the Valley of Death. Not because the technology did not work. Because organizations quit before the curve inflected.
They set wrong expectations.
Leadership expects immediate improvement. The business case shows productivity gains starting in month one. The board is told that AI will pay for itself quickly.
When the J-Curve appears, it looks like failure. The expectations were wrong, so reality seems broken.
They measure too early.
Eager to demonstrate value, organizations measure results immediately. They compare month two to the pre-deployment baseline. The comparison looks terrible.
Early measurement captures only the cost side of the J-Curve. The benefit side has not arrived yet.
They lose sponsor confidence.
Executive sponsors have limited political capital. They spend that capital to get the initiative approved. When early results disappoint, their capital depletes.
By month three, sponsors may be looking for exits. They do not want to be associated with a visible failure. They withdraw support.
They cut too soon.
Facing disappointing metrics and wavering sponsors, organizations cut initiatives that were working.
The cut happens at the bottom of the J-Curve. The inflection point was approaching. The benefits were about to materialize.
But nobody knew that because nobody understood the J-Curve.
What the 5% Do Differently
The 5% who succeed with AI understand the J-Curve. They plan for it. They manage through it.
They set realistic expectations.
Before deployment, they educate leadership about the J-Curve. They explain that productivity will decline before it improves. They set a timeline for when benefits will materialize.
This is not lowering expectations. It is calibrating expectations to reality.
When the dip arrives, it is expected. Leadership knows this is part of the process, not a sign of failure.
They protect the initiative.
Understanding that the Valley of Death is coming, they build protection in advance.
They secure commitments that the initiative will not be evaluated until after the expected inflection point. They create governance that prevents premature termination. They build coalitions of supporters who will defend the initiative through the valley.
They measure leading indicators.
Lagging indicators like productivity and revenue improvement do not move until after the inflection point. Measuring them early produces discouraging results.
Leading indicators move earlier. Adoption rates. Usage patterns. Capability development. Learning velocity.
The 5% measure leading indicators in the early months. They track signals that the curve will eventually inflect. They use these indicators to maintain confidence while lagging indicators are still negative.
They celebrate learning.
In the Valley of Death, the primary output is learning. What works? What does not? How does AI actually perform in your context?
The 5% treat learning as valuable output even when productivity is down. They document what they are learning. They share it widely. They create narratives about progress that are not dependent on premature productivity metrics.
They manage the narrative.
The J-Curve is not intuitive. People expect improvement, not decline.
The 5% manage the narrative actively. They explain what is happening. They contextualize the metrics. They help stakeholders understand that the current state is expected and temporary.
Narrative management is not spin. It is education. Stakeholders who understand the J-Curve can support the initiative through it.
How to Plan for the J-Curve
If you are about to deploy AI, or have recently deployed, here is how to plan for the J-Curve.
Educate before you deploy.
Before the initiative launches, educate everyone who will see metrics or make decisions.
Explain the J-Curve. Show Brynjolfsson’s research. Describe what will happen and when. Make the dip expected rather than surprising.
This education is not optional. Stakeholders who do not understand the J-Curve will make bad decisions when it appears.
Build the protection structure.
Design governance that protects the initiative through the valley.
Secure commitments that the initiative will not be evaluated on lagging indicators until a specified date. Create review processes that consider leading indicators, not just outcomes. Establish decision rights that prevent premature termination.
This structure should be in place before deployment. Building it after the dip appears looks defensive.
Define leading indicators.
Before deployment, define the leading indicators you will track.
Adoption: Are people actually using the AI tools?
Engagement: How frequently and deeply are they using them?
Capability: Are people developing judgment, not just tool proficiency?
Learning: What are you discovering about what works?
These indicators move before lagging indicators. Track them from day one.
Create the learning narrative.
From the beginning, frame the initiative as a learning journey.
Document what you learn. Share it widely. Create regular communications that emphasize learning progress.
When someone asks “is it working?” you should be able to answer with learning evidence, not just outcome metrics.
Set the inflection timeline.
Based on your organizational context and the research, estimate when the inflection point will occur.
For most organizations, this is between month six and month twelve. Organizations with stronger Human Layers may inflect earlier. Organizations with weaker foundations may take longer.
Communicate this timeline. Make it explicit. “We expect to see productivity improvement beginning in month nine.”
This gives stakeholders a concrete expectation to hold onto through the valley.
The Conversation with Your CFO
Let me give you language for the conversation that is coming.
When your CFO arrives with the discouraging spreadsheet, here is what you say:
“This is exactly what we expected. Let me explain.”
Then walk through the J-Curve:
“Research by Stanford economist Erik Brynjolfsson shows that productivity declines before it improves when organizations adopt transformative technology. The decline can be significant. This is normal, not a sign of failure.”
Show the shape:
“We are here in the curve. We are in the investment phase, where we are building capability, adapting processes, and generating learning. The return phase has not started yet.”
Point to leading indicators:
“While productivity is down, look at what is happening. Adoption is at 73%. Users are engaging more deeply each week. We are learning rapidly about what works. These are the indicators that predict eventual success.”
Reference the timeline:
“We set expectations before launch that productivity improvement would begin around month nine. We are in month three. We are on track.”
Ground in research:
“MIT found that 95% of organizations get zero return from AI. A significant reason is that they quit in the valley before the curve inflects. The 5% who succeed understand this pattern and persist through it.”
Ask for commitment:
“What I need is continued support through month nine. At that point, we will have real outcome data to evaluate. Pulling the plug now means killing an initiative that is performing exactly as research predicts it should.”
This conversation is different from a defensive scramble. It is an educated explanation grounded in research. It gives your CFO a framework for understanding what is happening.
CFOs are not irrational. They are responding to incomplete information. Give them complete information and many will respond appropriately.
The Hidden Costs of Quitting Early
Organizations that quit in the Valley of Death do not just lose the investment already made. They incur hidden costs that compound over time.
Credibility damage.
Failed initiatives create organizational scar tissue. Leaders who sponsored failed projects become cautious. Teams who worked on failed initiatives become cynical.
The next AI initiative faces higher skepticism. Approval is harder. Support is weaker. The credibility damage from the failed initiative makes future success less likely.
Capability loss.
The learning that accumulated in the Valley of Death walks out the door or gets forgotten.
What did people learn about AI in your context? What worked? What did not? This knowledge had value. When the initiative is killed, the knowledge dissipates.
The next initiative starts from zero rather than building on accumulated learning.
Competitive disadvantage.
While you quit and restart, your competitors continued. They pushed through their Valley of Death. They reached the inflection point. They are now accumulating compound advantages.
The gap that opens during your restart period may never close.
Organizational antibodies.
Failed initiatives create antibodies. The organization develops resistance to similar future initiatives.
“We tried AI. It did not work.”
This narrative, even when wrong, becomes organizational truth. It blocks future attempts. It becomes a self-fulfilling prophecy.
The hidden costs of quitting early often exceed the visible costs of the investment itself.
How Deep Is Your Valley?
The depth of your J-Curve valley depends on your Human Layer readiness.
Organizations with strong Human Layers have shallower valleys. Leadership is aligned. Data flows where it needs to go. People can judge AI outputs. Processes are designed for adaptation. Governance is clear. Culture supports experimentation.
These organizations absorb AI more easily. The learning investment is lower. The process adaptation is faster. The valley is shallow and brief.
Organizations with weak Human Layers have deeper valleys. Leadership is confused. Data is siloed. People accept AI outputs without judgment. Processes are accidental. Governance is unclear. Culture punishes failure.
These organizations struggle to absorb AI. Everything takes longer. Mistakes are more common. The valley is deep and extended.
This is why I emphasize building the Human Layer before deploying AI.
The Human Layer investment reduces the depth of the J-Curve. It shortens the time in the valley. It increases the probability of reaching the inflection point.
Organizations that skip Human Layer work and rush to deployment face the deepest valleys. Many do not survive them.
The Board Presentation
If you need to present to your board during the Valley of Death, here is a framework.
Normalize the pattern.
Open with the J-Curve research. Establish that productivity decline is expected and predicted by the best research available.
Show where you are.
Place your initiative on the curve. Explain that you are in the investment phase. Be specific about timeline to inflection.
Present leading indicators.
Boards understand leading versus lagging indicators. Show them the leading indicators that predict eventual success.
Quantify the learning.
What have you learned? What do you know now that you did not know at deployment? Frame learning as tangible output even when productivity metrics are down.
Project the outcome.
Based on your progress and the research, project what outcomes will look like after inflection. Give the board something to look forward to.
Request specific support.
Do not leave the room without specific commitments. Budget protection through month X. Evaluation postponement until date Y. Executive sponsorship continuation.
Make the ask explicit.
The J-Curve is not failure. The J-Curve is transformation.
Every organization that succeeds with AI passes through the valley. The question is whether you understand what is happening and persist through it.
The 95% who fail often quit at month three.
The 5% who succeed reach month nine.
The difference is not luck. It is understanding.
Have you experienced the J-Curve in your AI initiatives? How did you manage through the valley?
The AI Readiness Scorecard helps you assess how deep your valley is likely to be. Organizations with stronger Human Layers face shallower curves.
Comment “SCORECARD” below and I will send you access.
The valley is coming. The question is whether you will be prepared for it.