Chapter 010　Which Companies Succeed with AI, and Which Fail

We deployed AI. The company does not feel any different. As of August 2026, we hear this from executives in every industry. The tools are installed. Usage is climbing. The shape of the business is what it was last year. In the same period, other companies deployed the same technology and the outline of their business changed. What separated them was not engineering strength and not budget. In this chapter we dissect where that difference is created, at the level of practice.

1 The question — why it arises now, in this form

The question of whether an AI deployment succeeded has changed in kind over the past few years. The earlier question was whether AI actually worked. It doubted the technology. That question is settled. Prose, code, analysis, design proposals — AI produces all of them at a usable standard. Today’s question is different. Why has the company not changed, when what we installed was supposed to work? The shift matters. The first was a technical problem. The second is a management problem. A technical problem can be solved by a technical function. A management problem can be solved only by the executive team. The awkward part is that this failure does not look like failure. The deployment is complete. The front line is not complaining. There is a slide deck showing measured effects. And yet the revenue structure, the definition of the business, and the shape of the organization have not moved at all. We call this the quiet failure. No loss is booked, so nobody stops it. Two years pass, then three. In that time another company using the same technology has arrived somewhere entirely different. That is why the question arises now, in this form. What do we call success, and what do we call failure? An overwhelming number of companies have that definition wrong.

2 Conventional answers and their limits — the error of

measuring “did we master it” Three ways of measuring an AI deployment circulate today. All three are widely used. None of them measures success or failure. The first answer: “Rising deployment and usage rates mean success” This is the most widespread measure. Did every employee get an account? What share are weekly active? Are monthly queries growing? The indicator has meaning early. A tool nobody uses produces nothing. But the indicator has a ceiling. Usage measures how much AI was touched inside the existing work. Unless the work changes, a hundred percent usage rate leaves the company a company that does its existing work slightly faster. Usage is a necessary condition. It is nowhere near a sufficient one. The second answer: “Return on investment is the objective measure” Next most common is measurement by financial indicator. Convert hours saved into currency and compare against the investment. Payback inside the threshold counts as success; outside it, failure. It looks rigorous. It measures only what can be measured. AI produces two kinds of effect. It lowers the cost of existing work, and it creates value that did not previously exist. The first can be measured. The second cannot be measured in advance, because there is no method for estimating revenue in a market that does not exist. The result is that companies evaluating by payback approve only measurable proposals. Measurable proposals are cost reductions in existing work. The “objective criterion” then functions as a device that pushes the enterprise back inside its existing business. We made this point in 100 Questions on Management in the Age of AI, #061, “The Real Reason the Return on AI Investment Stays Invisible.” The return is not invisible. Only the proposals whose return is visible are being approved. The third answer: “Get the data and the talent in place and success follows” The third is an argument about preconditions. Build the data platform, hire specialists, raise literacy across the workforce. Once the conditions are in place, results will come. The direction is right. The order is wrong. A data platform can be designed only once you know what you want to know. The same holds for talent. Companies that assembled both before settling the purpose end up holding excellent infrastructure and excellent people, with nobody able to say what to do. The three conventional answers share one assumption. They measure success by whether the company mastered AI. The reason that assumption fails is plain. The ability to use AI well will soon stop differentiating anyone (→ Vol. I, Ch. 001). Models arrive from shared clouds and are refreshed every few months. Yesterday’s sophisticated technique is next month’s default. Make a criterion out of something that stops differentiating, and success itself becomes unmeasurable. So what do we measure instead? Our answer is single. The success of an AI deployment is decided not by whether the company could use AI, but by what the company became through it.

3 Redefinition — AI deployment is one means of

Enterprise Redefinition Future Value Theory places AI deployment as follows. An AI deployment is one means of carrying out Enterprise Redefinition. Not the objective. A means. And the success of a means can be judged only against the objective. One distinction belongs here. Digital transformation ends. Once the platform is built and the work is digitized, the program is complete. An AI deployment ends too. Once what should be deployed is deployed, the program is over. Enterprise Redefinition does not end. As long as assumptions keep going obsolete, the enterprise keeps redefining itself. First Principle 6 states it. Enterprise Exists to Redefine Itself — continuous self-redefinition is its essence. Completing an AI deployment therefore does not complete Enterprise Redefinition. Companies that confuse the two stop thinking on the day the deployment closes. Put the order back The Future Value Chain fixes the order in which value appears. Purpose → Learning → Redefinition → Creation → Enterprise Value Where in that chain do failing deployments begin? In most cases far down it — AI is connected directly to an existing business process sitting just short of Creation. AI connected without passing through Purpose achieves the existing purpose at speed. If the existing purpose is “protect this quarter’s gross margin,” AI works at full strength to protect this quarter’s gross margin. The technology is functioning correctly. It is simply not pointed at the future. This is the structure at the root of AI failure. AI has no purpose of its own. So it amplifies the purpose the enterprise already holds. An old purpose gets amplified as an old purpose. First Principle 4 states it. AI Optimizes. Humans Define. Deploy AI without updating the definition, and the old definition is what gets optimized. AI Integration Capability Here we can name the capability that separates success from failure. Future Value Creation Capability is expressed by this formula. FVCC = Purpose × Learning × Redefinition × AI Integration × Ecosystem × Capital Allocation × Trust This is multiplication, not addition. The relationship is multiplicative: weakness in any single capability weakens the whole, so purpose, AI, and capital are each individually insufficient, and Future Value emerges only when all seven reinforce one another. The fourth term is AI Integration. We treat it as a capability and call it AI Integration Capability. It is not the capability to deploy AI. It is the capability to integrate AI into the chain of Purpose, learning, redefinition, and capital allocation. The distinction is decisive. Deployment places AI beside the organization. Integration builds AI into the structure of the organization. AI placed beside the work makes the work somewhat faster. AI built into the structure changes what counts as the work. In practice, the multiplicative property reads like this. However high the AI Integration term, if the Purpose term is zero, Future Value Creation Capability is zero. The phenomenon of a whole company fluently using the newest models while the enterprise changes in no respect is explained exactly by this formula. The converse holds. With a clear Purpose but AI Integration at zero, the whole is still zero. In the Age of AI, choosing not to integrate AI is equivalent to leaving one term of the capability blank. The definition of success From this we define the success of an AI deployment. The five dimensions of Enterprise Redefinition are Purpose, Business, Organization, Capital, and Leadership. Success is the state in which, through the AI deployment, at least one of those five dimensions has actually been rewritten. Failure is the state in which AI is running everywhere and none of the five has been rewritten. Not usage. Not payback years. Whether something was rewritten. Under this definition many celebrated “success stories” are reclassified as failures, and unglamorous cases are reclassified as successes. That reclassification is precisely the practical value of the definition.

4 Structure — five types of failure, and the cause of each

Failures look innumerable. The types are few. We observe the same five repeatedly. In each case we go past the symptom to the cause.

4.1 Stuck at pilot — the purpose of the trial was mistaken

The most common type. A pilot ran. Results were good. It never reaches production. The following year another pilot begins on another theme. The cause lies in the purpose of the trial. Most pilots test whether AI can be used in a given task. That is a technical test. What a production decision requires is a test of what changes in the business when this task is done by AI. That is a business test. Pass the technical test without running the business test, and the deployment proposal cannot clear approval. The grounds for approving it do not exist. The second cause is the absence of a stopping rule. Without one, every pilot concludes that “a certain effect was confirmed.” Everything passes, and everything stalls. The remedy is simple. Write the business-side decision conditions before designing the trial. What has to change, and by how much, to justify company-wide rollout? What has to fail to trigger cancellation? A trial without those two decided should not be started.

4.2 Departmental optimization — the department is the unit of

deployment The second type. AI goes in department by department, each department shows an effect, and nothing changes company-wide. The cause is that the department is the deploying unit. Departments are appraised on departmental metrics. Any AI use a department designs will therefore improve departmental metrics. That is rational behavior. But departmental boundaries are also work boundaries and data boundaries. Optimize repeatedly inside a boundary and you lose every occasion to doubt the boundary. The largest value in the Age of AI comes from redesigning processes that cross departments. There, no owner exists. Recall here a mandatory note on the Enterprise Redefinition Maturity Model. The model evaluates organizational coherence rather than isolated excellence. One department’s outstanding AI use does not raise enterprise maturity. Maturity is assessed across all five dimensions in balance: organizations with exceptional technological capability but weak leadership redesign cannot achieve higher maturity. The second cause is that departmental optimization looks like success in the short run. Results reports arrive from each department, and the executive meeting registers that results are being produced. At the moment of registering it, the structural problem becomes invisible.

4.3 Missing purpose — it started because other companies

started The third type. Why the deployment is happening was never settled. This one is hard to see in yourself. Ask why, and an answer comes back. Productivity. Competitiveness. The labor shortage. None of these is a purpose. They are abstract nouns shaped like a purpose. A purpose is something you can judge as achieved or not achieved. “Improve productivity” cannot be judged. “Rebuild quotation preparation so that no person handles it, and move sales time into defining new customers’ problems” can be judged. The cause lies in the motive for starting. A competitor started. The board asked. A shareholder asked. A deployment begun from that motive has achieved its purpose at the moment it begins, because beginning was the purpose. First Principle 1 states it. Purpose Precedes Profit — profit is the result of a purpose society has embraced. An AI deployment that breaks the order does not arrive at profit either.

4.4 Preserving the existing work — AI is laid on top of an

unchanged process The fourth type is the most widespread and the least noticed. AI goes in. The business process stays as it was. The assignee stays, the approval stages stay. A person checks every AI output, and one more step joins the process. Total hours barely fall. Sometimes they rise. Two causes. First, the decision to change the work is heavier than the decision to deploy. Deployment is settled by budget. Abolishing work touches people and the organization. Because nobody wants to touch it, only the deployment happens. Second, the design has stayed inside the Human-in-the-Loop idea, in which humans supervise AI from inside the operational loop. A design that inspects AI’s output item by item breaks down as volume grows. What is needed is Human-on-the-Loop Management — the move to a position that designs the whole system (→ Vol. I, Ch. 009). This type is also the classic presentation of Level 2 of the Enterprise Redefinition Maturity Model, the Improvement Enterprise. In the paper’s own words, such organizations become “increasingly efficient while remaining fundamentally unchanged.” Changing the work and cutting headcount are not the same thing. On that short circuit, see 100 Questions on Management in the Age of AI, #046. Once a process is abolished, where the freed capability is reallocated is a management decision.

4.5 Misalignment with the appraisal system — the system

rewards old behavior The fifth type is institutional. Many companies encourage AI use while leaving the appraisal system untouched. Cases handled. Hours worked. Volume of documents produced. As long as those are what is appraised, an employee who deletes a process with AI is worse off on the scorecard. They reduced their own workload, after all. The cause is that AI deployment is treated as an operations initiative and never reaches the owner of the HR system. The function driving deployment and the function designing appraisal are different. The only forum where both sit on the same agenda is the executive meeting. There is a mirror-image misalignment. A system that rewards only short-term cost reduction. Under it, attempting a use that creates new value never pays. Failure is penalized, and success has nowhere to be booked. Systems are stronger than doctrine. However much an executive speaks of redefinition, if the system rewards old behavior the organization will choose old behavior. The root common to all five Line the five up and one root appears. All five treat AI deployment as a matter of work. As long as it is a matter of work, the decision rights sit with the front line and the IT function. They can improve the work. They do not hold the authority to change the definition of the business. The people who hold that authority are not on the agenda, and a change requiring that authority is being expected anyway. That is the shared structure of failure.

5 What it looks like in practice — the structure of

success, and how to diagnose yourself

5.1 Four structures common to the companies that succeed

As far as we observe, organizations whose definition actually changed through an AI deployment share four features. Industry and scale differ. These four are present without exception. First, Purpose comes first. Before the deployment discussion, what the company is trying to solve has been settled. One medical device company decided first to move from selling devices to carrying responsibility for diagnostic accuracy itself. Once that was decided, where to put AI followed automatically. In the reverse order, the placement is never settled. Second, capital allocation actually moves. This is the easiest sign to read. Put this year’s budget beside last year’s; if the destinations have not moved, nothing has changed yet. First Principle 3 states it. Capital Exists to Create Possibility, not merely to maximize return. Capital that only defends the existing business is not creating possibility. Capital is not only money. People, time, and executive attention are capital. After AI deletes a process, where were the freed people and hours reallocated? Seriousness shows up there. Third, the metrics change. Companies that succeed decide the change of metrics in the same meeting as the deployment. One logistics company stopped appraising by volume processed and added the number of proposals to redesign a process. Without changing the metric, behavior does not change. Fourth, the executive changes the question. In a failing company’s executive meeting, the question is how to use AI. In a succeeding company’s executive meeting, the question is: if AI can do this, what kind of company are we? The first is a question AI can answer. The second is a question AI cannot pose. The Leadership Formula shows the structure. Leadership = Purpose × Question Design × Capital Allocation × System Architecture × Trust Note that Question Design and Capital Allocation stand side by side. Changing the question and moving capital are two faces of one managerial act. Change the question without moving capital, and the organization concludes that only the vocabulary changed. It concludes correctly. 5.2 Diagnosing yourself with the Enterprise Redefinition Maturity Model To locate where an enterprise currently stands, we use the Enterprise Redefinition Maturity Model (ERMM). The five levels are as follows. Level 1, the Reactive Enterprise. It primarily responds to external events. Transformation occurs only after significant deterioration in performance. Level 2, the Improvement Enterprise. It actively pursues operational excellence. Improvement, however, remains incremental, and existing business models are rarely questioned. Level 3, the Transformation Enterprise. It recognizes that business models require substantial redesign. It continues, however, to view redesign as a project rather than a permanent organizational capability. Level 4, the Continuous Redefinition Enterprise. Enterprise redesign becomes embedded within normal management processes. Such organizations increasingly redesign themselves before external disruption requires it. Level 5, the Future Value Enterprise. Rather than reacting to external change, these organizations actively shape future industries. One warning, stated strongly. Reaching Level 5 as rapidly as possible is not the objective. Different industries may require different levels of organizational adaptability. The purpose of diagnosis is not ranking. It is finding where your own enterprise lacks coherence. Progression is not linear, either. Organizations frequently display characteristics from multiple levels simultaneously. Diagnose each of the five assessment dimensions. Use the paper’s assessment questions as written. Purpose — “Does the organization periodically re-examine its Core Purpose and adapt its expression without unnecessarily weakening organizational identity?” Business — “Does the business model continuously evolve?” Organization — “Can structures adapt rapidly to technological change?” Capital — “Are resources allocated toward Future Value rather than historical success?” Leadership — “Do leaders continuously redesign the enterprise rather than merely manage operations?” The procedure has four steps. First, score the five dimensions separately. Do not combine them; always read them apart. Second, identify the lowest dimension. Third, measure the gap to the highest. Fourth, decide the single move that closes that gap. Why read the lowest value? The paper notes the reason. An organization may possess Level 4 AI capability while remaining Level 2 in leadership. That organization’s real standing is not Level 4. Most companies that fail with AI have this shape.

5.3 Why most companies stop at Level 2

In our observation, the majority of companies working on AI stall at Level 2, the Improvement Enterprise. And they do not notice that they have stalled. There are four reasons. First, Level 2 is comfortable. Improvement always produces a result, and the result can be shown as a number. Numbers are welcomed at the executive meeting. Level 2 is experienced as success, not as failure. Second, existing profit is protected. Improvement takes the existing business model as given, so it does not damage current earnings. Redefinition almost always worsens the short-term numbers. When an option that does not worsen them is available, people take it. Third, agreement is easy to obtain. An initiative that does not question the existing business has no department opposing it. An initiative that questions it always has an affected party. The cost of building agreement differs by an order of magnitude. Fourth, the completion of DX looks like an endpoint. As the paper records, at Level 2 digital transformation becomes systematic and AI adoption expands. The program completes. At the moment it completes, the enterprise mistakes completion for arrival. But digital transformation ends and Enterprise Redefinition does not. The company finishes the thing that ends and believes it has begun the thing that does not. That is the Level 2 trap. How does a company get out? Three conditions. Condition 1 — move the object of doubt from the work to the business model. What separates Level 2 from Level 3 is not the quality of improvement. It is whether the existing business model was questioned. Change “how do we make this process faster” into “is this business right in this form.” Swapping one agenda item is enough to cross the boundary. Condition 2 — move capital allocation first. Start from discussion and Level 2 gravity pulls you back, because the case for defending the existing business is always the more concrete one. Move the allocation first, and the discussion begins from an allocation that has already moved. Condition 3 — make redefinition a standing item of management. The limit of Level 3 is that redesign is still treated as a project. Projects end. The only way to keep it from ending is to embed it in the management process. None of the three conditions has anything to do with AI performance. That leads directly to this chapter’s conclusion. What separates success from failure in an AI deployment sits on the enterprise’s side, not on AI’s.

6 Questions for the executive

The argument, in one line. The company that succeeds with AI is not the company that mastered AI. It is the company that used AI as the occasion to rewrite its own definition. The company that fails is not the company that could not use AI. It is the company that could, and then chose the use to fit the company it was yesterday. Four questions to close. None is abstract. Each can be answered at your next executive meeting. Question 1 — Which of the five dimensions is your AI deployment meant to rewrite? Purpose, Business, Organization, Capital, Leadership. If none of them can be named, the deployment is an operations initiative. Operations initiatives are not bad. They simply must not be called Enterprise Redefinition. Question 2 — This year, where were the people and the hours AI freed up reallocated? Do not answer with the amount saved. Answer with the destination. If the destination is “slack in existing work,” the enterprise is at Level 2. If the destination is a question nobody was responsible for before, the enterprise has begun to move. Question 3 — Does your appraisal system reward or punish the person who deleted a process with AI? Few executives can answer immediately. Ownership of the system is split, and it sits outside the executive’s line of sight. But systems move organizations more forcefully than doctrine does. Question 4 — Which of the five dimensions is lowest? And is the executive included in that assessment? The most common answer is Leadership. In companies with high AI capability whose maturity still does not rise, the executive is often the lowest value. A diagnosis that excludes yourself is not a diagnosis. None of the four questions asks about AI. All four ask about your own company. An AI deployment is not the procurement of technology. It is the occasion to ask again what your company is, using technology as the mirror. It is possible to complete a deployment without looking at what the mirror shows. A few years later you arrive back in front of the same question. First Principle 6 states it. Enterprise Exists to Redefine Itself — continuous self-redefinition is its essence. AI is the most powerful means yet available for accelerating that redefinition. What the redefinition moves toward is not decided by AI. It is decided by the executive. Before the deployment approval is signed, that one point has to be settled.

In brief

The company that succeeds with AI is not the one that mastered AI; it is the one that rewrote its own definition.
AI holds no purpose and amplifies the purpose already in place, so a deployment without an updated definition accelerates an old purpose.
Success is judged not by usage or payback years but by whether one of the five dimensions was actually rewritten.
The root common to the five types of failure is a single one: the AI deployment has been cut loose from Enterprise Redefinition.

Key concepts

Enterprise Redefinition / Enterprise Redefinition Maturity Model (ERMM) / Future Value Chain / Future Value Creation Capability (FVCC) / Purpose

The chain of ideas

Purpose → Learning → AI Integration → Redefinition → Future Value

Related first principles

Principle 1 — Purpose Precedes Profit. Principle 3 — Capital Exists to Create Possibility. Principle 4 — AI Optimizes. Humans Define. Principle 6 — Enterprise Exists to Redefine Itself.

Related chapters

Vol. V, Ch. 044 “What Is the Enterprise Redefinition Maturity Model (ERMM)?” — the five levels and their diagnostic criteria in full
Vol. VI, Ch. 053 “AI and Enterprise Redefinition” — the relationship between AI and redefinition, taken head-on
Vol. II, Ch. 011 “What Is an Organization in the Age of AI?” — designing an organization that gets past departmental optimization
Vol. II, Ch. 013 “What Is Corporate Culture in the Age of AI?” — why systems and behavior fall out of alignment

Papers and companion volumes

Kadowaki, N. (2026a). Future Value Theory: A Management Framework for Enterprise, Capital, and Society in the Age of AI. VURA Working Paper Series. SSRN: https://ssrn.com/abstract=7120980 / Zenodo: https://doi.org/10.5281/zenodo. 21255662
Kadowaki, N. (2026b). Enterprise Redefinition: Toward an Enterprise Evolution Theory for the Age of AI. VURA Working Paper Series. (Published on Zenodo; under review at SSRN)
100 Questions on Management in the Age of AI, #061 “The Real Reason the Return on AI Investment Stays Invisible” / #046 “Before You Call It an AI Restructuring: What the Executive Should Think Through”

Chapter 010 Which Companies Succeed with AI, and Which Fail