Why Consulting Wins the Meeting and Loses the Year

and value follows. They can believe that the remaining work is training, tooling, and a few policies, then the promised gains show up in the ledger. The chart is not wrong. The inference most people draw from it is.

The piece is not about AI adoption. It is about conversion

Major consulting firms are world class at producing insight artifacts that sound like action. They are far less consistent at installing the conversion system that takes an insight and turns it into action, then into control, then into outcomes, then into value. They can tell you what is happening. They can often tell you why it is happening. They frequently stop before they show you how an enterprise will keep producing the desired result after the consultants leave. The same research brief that reports rising AI use also shows the pattern. It frames the executive as a decision maker who can integrate AI into how they decide, review, and course correct. It points to real concerns around legal exposure, explainability, and data quality. It offers a sensible posture. It still leaves the reader holding the largest unpaid bill in the room, which is the missing middle between knowing and governing. What would have to be true for this outcome to keep repeating. That line is the entire test. Not for the brief alone, but for the consulting model that produces it, and for the executive class that keeps buying the same kind of artifact and calling it progress.

The client becomes middleware, and the bill is paid in imagination

A consulting engagement rarely fails because the people involved are unserious. The people are serious. The decks are polished. The analysis is often correct. The failure happens because the artifact stops where the work becomes politically expensive and architecturally permanent. The consulting firm delivers an insight and a recommended set of moves. The client then has to imagine how that insight converts to action through a process that is not their own. That is not a small detail. It means the organization must translate an external vocabulary into its internal operating language, then execute it inside permission boundaries that were built to prevent uncontrolled action. It must do this while still running the business, while still defending budgets, while still surviving quarterly scrutiny, while still managing risk that the deck does not own. The imagination bill is paid in inference. Somebody has to infer who owns the decision. Somebody has to infer what evidence qualifies as sufficient. Somebody has to infer what exceptions are allowed. Somebody has to infer which approvals can be bypassed and which cannot. Somebody has to infer what happens when the model is wrong, when the data is stale, when the action triggers a regulatory consequence, when the customer reacts in a way the scenario never predicted.

That somebody is almost always an operator with a full calendar and a real P and L. They become the translation layer between the insight artifact and the enterprise that must execute it. The client becomes middleware. This is why so much consulting work produces a season of motion and then drift. The motion is real. The drift is predictable. The artifact never installed a native architecture of control that can keep producing the result without foreign energy.

The prevailing belief, steel manned

The prevailing belief is reasonable. It deserves to be stated cleanly, because people keep acting as if it is true. Complex enterprises are hard to coordinate. They contain silos, competing incentives, and risk gates that slow action. External consultants provide clarity, speed, and cross functional authority. They can see patterns internal teams normalize. They can benchmark what good looks like. They can propose a target operating model and a change plan. If the client adopts the plan, results follow. If the client fails, it is because the client did not execute, or leadership did not commit, or the organization resisted change. There is truth in that. Many enterprises do need an external view. Many internal teams do normalize waste. Many leaders do hide behind process when they fear consequence. Many engagements fail because the client refuses to make hard decisions. That belief still collapses under one question. What converts the insight into repeatable outcomes under variance, inside the client’s actual permission system, after the external authority leaves. Most consulting artifacts do not answer that question because the answer is not an artifact. The answer is architecture.

Insight does not create action. Permission does

Executives love insight because it feels like control without risk. You can approve an insight. You can endorse a vision. You can sponsor a roadmap. You can do all of that without changing who is allowed to do what on Tuesday morning when the plant is down, the supplier is late, the customer is escalating, and legal is watching. Action is not an idea. Action is permission. Enterprises survive by restricting action. That restriction is not always wise, but it is rarely accidental. It is a response to risk, blame, and history. Permission boundaries are how the organization encodes what it fears, what it values, and who it trusts. A consulting deck that proposes action without redesigning permission is not a plan. It is a request for heroics.

This is why AI adoption statistics can rise while decision speed and outcome quality do not. An executive can use AI daily, even hourly, and still be unable to act, because the bottleneck is not cognition. The bottleneck is permission. AI can expand inference. It cannot grant authority. It can suggest the move. It cannot sign for the move. It cannot carry the liability for the move. That is where the missing middle lives. Inference is accelerating. Permission is not. The research brief’s chart is a perfect tell. It shows active use rising, but it does not show a corresponding compression in the time between signal and committed action for the decisions that matter most, capital allocation, product liability, safety exposure, major supplier changes, pricing moves, plant shutdown calls. Those decisions are governed by permission ceilings for good reason. They are also the decisions where latency is most expensive. If a firm cannot redesign the permission layer, it will become a more intelligent spectator.

Control is not governance, and oversight is not authority

Many consulting engagements respond to risk by proposing governance. That instinct is safe. It is also often useless. Governance is oversight. Control is an operating property. Control means the enterprise can take action within defined bounds, detect variance, correct fast, and learn in a way that updates the system, not just the story. Control requires signals, thresholds, and embedded authority to act. It also requires an evidence record that can survive legal scrutiny and internal politics. Oversight can watch a system fail in high definition. It cannot make the system succeed. This is why so many transformations produce more meetings and slower decisions. The organization adds oversight to compensate for uncertainty. Oversight multiplies interfaces. Interfaces multiply coordination cost. Coordination cost becomes latency. Latency becomes vulnerability. A firm can publish a research brief warning about legal risks, explainability, and data quality, and it can be fully correct. If the prescription stops at oversight and policy, the enterprise remains unchanged at the point that matters. It still cannot convert a recommendation into an authorized action fast enough to win. The missing middle is not ethics language. The missing middle is the control plane.

The alien process problem, and why politics kills it

You asked for the blunt version. Here it is.

Consultants often impose an artificial process on top of the enterprise rather than build an architecture the enterprise can operate as its own. The process can produce visible progress while the external authority is present. It can also be unsustainable, because it is alien to the host. Alien does not mean foolish. Alien means it requires foreign energy to run. It shows up as constant workshops, constant translation, constant facilitation, constant escalation, constant reminders. It shows up as a new vocabulary that employees must speak in meetings while they still do the real work in the old language. It shows up as parallel reporting that lives beside the real operating cadence rather than inside it. It shows up as borrowed authority that forces cross functional alignment that the internal system could not force without casualties. Then the engagement ends. The foreign energy leaves. The system returns to its native permission boundaries, its native incentives, and its native political reality. The imposed process dies. The political nature of the enterprise kills it because politics is not merely ego and turf. Politics is the enforcement layer of permission. It allocates budget, assigns blame, protects careers, and constrains risk. It decides which actions are allowed, and which actions are career ending if they go wrong. When a consulting process bypasses that reality rather than redesign it, it creates short term motion and long term rejection. The bypass itself becomes the reason the system cannot adopt the change. People learn that the change only works when the consultants are in the room. That lesson spreads faster than the slide deck. AI can make this worse. AI can accelerate the production of insight artifacts and recommended actions, and it can do it at low cost. If the enterprise still requires human intermediaries to translate, approve, and defend each action across layers, AI becomes an insight amplifier feeding a permission bottleneck. More suggestions hit the same gate. The gate does not widen. The enterprise becomes more informed and more frustrated. That is the alien middleware problem in its final form. The client becomes the translation and approval layer between machine inference and organizational action. The client pays in calendar time, political capital, and personal risk.

The mechanism that makes the failure testable

If you want this to be more than rhetoric, you need a falsifiable mechanism. Here it is. When consulting delivers insight without installing a native conversion architecture, the client’s decision latency does not compress for high consequence decisions. It either stays flat or expands, because new oversight and new coordination requirements are added on top of the old permission model. The organization may see isolated wins where the consultants can force

alignment, but those wins do not repeat reliably after the engagement ends. The enterprise returns to drift, and leaders respond by buying the next artifact. That mechanism is testable in operations and finance. Measure the time between signal and committed action for a set of high consequence decisions before the engagement, during the engagement, and six months after the engagement ends. If the time does not compress after the foreign authority leaves, the engagement did not change the enterprise. It temporarily overrode it. Measure the cost of coordination. Count the interfaces required to execute the decision. Count the approvals. Count the escalations. Count the meetings. Track how many hours senior operators spend translating the plan into internal permission language. If those counts rise, the enterprise did not gain control. It gained overhead. Measure outcome attribution. Ask whether the outcomes can be linked to the actions taken, and whether the actions taken can be linked to specific evidence and permission grants. If the story cannot survive audit, it will not survive politics. The research brief’s data points matter here, not because they prove value, but because they show the class of executive behavior that will collide with this mechanism. If 92 percent of active users use AI daily, then inference is becoming a constant companion to executive thought. If permission remains unchanged, the gap between what leaders can see and what they can do grows. That gap is where risk, resentment, and failure breed.

Two diagnostic paragraphs executives can read aloud

If your firm hires a consulting team next quarter, and they bring a deck full of insight and a plan full of actions, can you point to the single permission boundary that will be rewritten so that a front line leader can act within defined limits without escalation. Can you name the role that owns that boundary, and can you name the evidence record that will defend it when something goes wrong. If you cannot name those three things, do you actually have a plan, or do you have an artifact that requires heroics to become real. When the consultants leave, what remains that can run without them. Is there a decision record that ties evidence to action, action to outcomes, and outcomes to value in a way that survives legal scrutiny and internal blame. Is there embedded authority to correct course without a steering committee. Is there a cadence that lives inside the operating system of the enterprise rather than beside it. If the only thing that remains is a roadmap and a governance calendar, what would you expect to happen other than drift.

The counterexample, and why it does not rescue the model

There are consulting engagements that work. Some firms embed operators, stay through variance, and build systems that persist. Some do the politically expensive work of redesigning

decision rights, simplifying interfaces, and installing control loops that survive leadership turnover. That is the counterexample. It matters because it proves the thesis is not that consultants are incapable. It proves the thesis is about incentives and default behavior. The dominant consulting product remains insight heavy, artifact centered, and outcome light, because that product is scalable, defensible, and billable without taking ownership of the conversion layer. The exceptions are real. They are exceptions because they require the consulting firm to do work that looks less like publishing and more like engineering.

The prediction that should embarrass the industry

Here is the prediction that will test whether anyone is serious. By the end of 2027, most large enterprises will report materially higher executive use of AI in strategic decision support, and many will cite daily use by senior leaders. Over the same period, the median cycle time from signal to authorized action for the top tier of high consequence decisions will not fall in a material way, and in many firms it will rise, because oversight and coordination will grow faster than permission boundaries are rewritten. That prediction is falsifiable. Pick a decision class, track the clock, and publish the result. If the clock compresses while auditability improves, the enterprise is installing control. If the clock stays flat while AI use rises, the enterprise is buying inference without permission.

What a complete engagement would have to deliver

A complete engagement would not stop at insight. It would deliver a conversion system that the enterprise can run as its own. It would tie insight to a specific decision owner and a deadline that forces tradeoffs. It would redesign permission boundaries so that actions can be taken within defined limits without constant escalation. It would build an evidence record that can survive legal scrutiny and internal politics, because accountability does not disappear when speed increases. It would install control loops that detect variance and correct fast, because outcomes do not hold still for quarterly governance. Most of all, it would make the sequence repeatable. Insight to action to control to outcomes to value would not be a slogan. It would be an operating property. It would be visible in the clock, visible in the audit trail, visible in the P and L. That is the gap. Inference, permission, architecture. If a consulting firm cannot cross that gap, it is not delivering transformation. It is delivering a story that the client must translate into reality.

The consequence, and the sharper question that remains

AI will make insight cheaper. It will make recommendations faster. It will make benchmarking easier. It will make decks prettier. None of that guarantees value. Value requires conversion, and conversion requires permission and control. So the question for the board is no longer whether the enterprise is adopting AI, or whether it has a governance committee, or whether leaders feel more informed. The question is whether the enterprise can authorize and execute high consequence action faster than its environment changes, without losing accountability. If the answer is no, then the firm is buying intelligence and renting control. That is not a strategy. It is a delay.

References

Capgemini Research Institute. Inside the C-Suite: How AI is quietly reshaping executive decisions. Research brief. 2026. Based on the Capgemini Research Institute survey, AI and the future of decision-making. August to September 2025. N equals 500 C-level executives. Michael Carroll. The One-Degree Dispatch. “Deepfakes. Due Process. And the New Burden of Proof.” LinkedIn newsletter essay. January 18, 2026. Michael Carroll. One-Degree decision doctrine. Essays and working papers on permission architecture, decision latency, inference boundaries, and conversion control. 2025 to 2026. W. Edwards Deming. Out of the Crisis. 1986. Ronald Coase. “The Nature of the Firm.” 1937. Oliver E. Williamson. Markets and Hierarchies. 1975. Herbert A. Simon. Models of Man. 1957. Jay W. Forrester. Industrial Dynamics. 1961. Stafford Beer. Brain of the Firm. 1972. W. Ross Ashby. An Introduction to Cybernetics. 1956. Daniel Kahneman. Thinking, Fast and Slow. 2011. Judea Pearl. Causality: Models, Reasoning, and Inference. 2009.