You don't have an AI problem. You have a decision problem.

Ask any senior leader and they will tell you the same thing: running a large organisation has never felt more complex. The number of decisions has multiplied. The variables feeding into each one have expanded. The cost of getting them wrong has never been higher. And yet the operating model most organisations rely on to make those decisions was designed for a world that looked nothing like today.

That is the real problem with enterprise AI. Not the models. Not the data. Not the tooling. The system that is supposed to turn intelligence into action was designed for a world that no longer exists: fewer decisions, simpler trade-offs, longer timelines. Organisations now run across global supply chains, interdependent systems, and compressed timeframes. The decision load has grown dramatically. The infrastructure for handling it has not kept pace.

I have worked inside this problem for twelve years, across financial services, energy, pharma, and national infrastructure. The sectors differ. The pattern does not. More AI goes in. The operating model stays the same. And the gap between insight and action only widens.

Why?

We are still asking the wrong question

The implicit premise of almost every AI investment I encounter is that if you make the inputs smarter, the outputs get better. That is a reasonable assumption about a system that works. Most enterprise systems were not built for the volume or complexity of decisions they are now being asked to handle.

Better models improve the quality of analysis. They do not change how a decision gets made, who owns it, or what the organisation learns from it afterwards. That part of the system has been left untouched.

The bottleneck has never been the intelligence. It is the absence of a system designed to translate that intelligence into accurate, accountable decisions at scale. Improving model quality without addressing that gap does not solve the problem. It often amplifies it: better information flowing faster into a process that still lacks the structure to use it well.

This is where Decision Intelligence becomes critical. Not as a feature or a point solution, but as an operating layer: one that augments human judgment on complex, high-stakes choices, automates routine decisions with precision and consistency, and creates the conditions for organisations to learn from every commitment they make. That is the capability most enterprises are still missing, and the reason AI investment alone has not closed the gap.

The operating model still has not changed

Most enterprise operating models were designed for a different era. One where decisions were fewer, systems were simpler, and coordination could happen through people and sequential process. That world is gone.

The shift that is actually underway is not from "no AI" to "AI." It is from organisations that run on processes to organisations that run on decisions. That sounds like a subtle distinction. It is not.

In a process-driven organisation, the volume and complexity of decisions gets absorbed into sequential workflows. Work moves from function to function. Trade-offs get resolved late, if at all. Outcomes are rarely connected back to the reasoning that produced them. As decision volume grows and complexity increases, these organisations do not adapt. They slow down, or they cut corners.

In a decision-driven organisation, the decision is the unit of work. Inputs connect at the point of commitment. Trade-offs are made explicitly. The organisation can learn from what it chose. As complexity scales, the system scales with it.

McKinsey's 2025 State of AI report makes this visible in the numbers. Nearly nine in ten enterprises are now regularly using AI. Yet just 39% report any measurable impact on enterprise-level earnings. Adoption is not the problem. The operating model is.

The same pattern holds for agentic AI specifically. Gartner projects that by 2028, 33% of enterprise software will include AI agents, up from less than 1% in 2024. Yet Gartner also estimates that more than 40% of those initiatives are at risk of being abandoned. Not because the technology failed. Because organisations are deploying agents into a process-driven operating model that was never designed to handle the decision load they are trying to automate. 

What is actually missing

To operate as a decision-driven organisation at scale, you need a layer most enterprises simply do not have. Not another tool. An environment where decisions can be explored before they are made, executed within the real system, and learned from afterwards.

What strikes me, having worked across some of the most consequential decision environments in the world, is how differently high-stakes fields outside of enterprise approach this problem. During the COVID-19 pandemic, we saw this directly; decisions about where to allocate critical medical resources were informed by forecasting systems that stress-tested outcomes before any commitment was made. In our work with pharmaceutical organisations, we have seen how modeling decisions computationally across thousands of scenarios before a single patient is enrolled changes the quality of commitment entirely. In those fields, the ability to stress-test a decision before committing to it is not considered advanced practice. It is considered a basic responsibility.

Enterprise has not caught up. The decisions that lock in cost structures, determine resource allocation, and shape outcomes for years are still largely made without that capability. The reasoning behind major commitments exists in meeting notes, in email threads, in the memories of the people who were in the room. When the volume and complexity of decisions increases, that gap does not stay the same size. It widens.

What we have seen consistently across the organisations Faculty has worked with is that when this changes, something important follows. The organisation begins to learn in a structured way. What was decided. What happened as a result. Where expectations diverged from reality. It compounds. Each decision makes the next one slightly faster and better informed. Over time, that becomes a structural advantage in handling complexity that is genuinely difficult for competitors to replicate.

Why this is hard, and why scale is the remaining problem

If this were purely a technical problem, it would already be solved.

The reason scale has remained the hard problem is organisational. You can build the right system in a controlled environment. Getting it to take root inside a global enterprise, across teams, cultures, hierarchies, and legacy systems, is an entirely different discipline. It requires connecting systems that were never designed to work together, aligning functions that have always operated independently, and redefining how decisions are made and owned at every level of the organisation. That is where most initiatives fail. Not because the idea is wrong. Because the organisation cannot absorb the change at the speed the technology enables.

This is the problem Faculty has been working on for a decade. When we founded the company, our mission was the safe and widespread adoption of AI. The "safe" part we understood clearly: rigorous, explainable, built for real-world conditions. "Widespread" is the harder half, and it remains so. Building decision intelligence that works in a controlled setting is one thing. Embedding it across an organisation whose decision volume and complexity keeps growing is another entirely.

What has changed is the capacity to close that gap at scale: deep applied AI capability, combined with the transformation reach to change how real organisations actually operate. Neither half is sufficient alone. That combination is what this moment requires.

What comes next

In five years, the competitive moat in every sector will not be which company has the best AI. It will be which company has built an operating model capable of handling the full volume and complexity of the decisions it faces, and has accumulated the institutional memory to make each one better than the last.

The organisations that get the most from AI over the next decade will not be the ones that adopted it fastest. They will be the ones that were willing to treat the operating model itself as the thing that needed to change. Moving from fragmented processes to coordinated decision systems. From isolated intelligence to integrated execution. From one-off improvements to something that genuinely compounds.

That is not a technology problem. It is a design choice. The infrastructure to make it is available now. The question is who builds it with the intention it requires, and who waits until the gap is too wide to close.