The Silo Ceiling: Why Most Organisations Will Never Reach Tier 3 AI

This article is part of our “CXO & Founders Introduction to” series, designed to clarify complex topics that frequently arise in the media, boardrooms, and workplaces. Rapid advances and a constant stream of announcements can make it hard to stay informed. Our aim is to explain core concepts in straightforward language and cut through jargon. Where helpful, we simplify descriptions and examples to enhance clarity and support understanding.

Ask a leadership team whether they are using AI, and the answer is almost always yes. Teams draft documents with it. Analysts turn spreadsheets into dashboards. Developers ship code faster. Productivity is up, and the board deck says so.

Then ask a second question: what has it changed about how the organisation works?

The room usually goes quiet.

That silence is not unusual. In McKinsey's 2026 State of AI survey, 80% of respondents said AI had improved their individual productivity. Only 37% could attribute any enterprise EBIT to it, roughly the same as the year before.¹ Productivity is rising. The business, for the most part, is not moving.

In our work with regional leaders, we see the same pattern. Most organisations believe AI integration means using AI to complete tasks. Some are starting to connect those tasks into workflows. Very few have recognised that a third tier exists, one where AI works across the organisation towards its goals.

We call these three tiers Task, Flow and Orchestrate.

This article explains each tier, why most organisations stall at Flow, and why the barrier is not technology. It is the structure of the organisation itself.

The tiering model outlines the value that AI delivers as it is implemented throughout your organisation. Each successive tier builds on the previous one and requires increasing commitment from the organisation, not the technology itself.

This represents the entry point for most users. Individuals leverage AI to convert information from one format to another. For example, an analyst may transform a spreadsheet into an interactive web page. A manager may condense a lengthy report into a concise, one-page summary, and an assistant may extract actionable items from a meeting recording.

Most AI use sits here today, and rightly so. Task is easy to adopt, the benefit is visible straight away, and the risk is low. Its limit is that the value stays with the person who used it.

At the second tier, individuals integrate multiple tasks, enabling the output of one to serve as the input for the next. For example, an analyst who creates a web page might incorporate an email sign-up feature, ensuring subscribers automatically receive weekly updates when the page changes. Although each task is familiar, their combination enables a single person to run a streamlined, end-to-end process.

This flow approach delivers tangible operational improvements. However, it typically remains confined to individuals or teams pursuing their objectives within their specific business domains. Many organisations mistakenly believe this represents the pinnacle of AI integration.

At the third tier, tasks and flows are integrated across the entire organisation in pursuit of a unified objective. Processes no longer halt at departmental boundaries. Instead, functions such as product, marketing, sales, and finance collaborate within a single, cohesive workflow, coordinated by AI.

Orchestrate is not merely a tool enhancement. It represents AI operating at the organisational level, transcending the confines of individual or team-based efforts.

These tiers are cumulative: flows are constructed from tasks, and orchestration arises from interconnected flows. As a result, organisations cannot bypass any tier.

While most AI maturity models assess progress based on levels of autonomy (such as assistant, colleague, or agent), this framework evaluates the breadth of impact: whether value is delivered to an individual, a team, or the entire organisation. This perspective equips leaders with a critical question for any AI initiative: How far does its influence extend?

If the tiers are cumulative, the obvious question is how far most organisations have climbed. The research gives a consistent answer: most are at Task, some have reached Flow, and very few are close to Orchestrate.

Task is everywhere. Most organisations now have AI in daily use. According to Deloitte's 2026 State of AI in the Enterprise report, 37% of organisations use AI "at a surface level with little or no change to underlying business processes."² The tools are in place, but how work gets done has not changed.

Flow is emerging. The same Deloitte research finds that only 30% of organisations are redesigning key processes around AI.³ Yet this is where results start to diverge. McKinsey's 2026 survey shows that nearly three-quarters of AI high performers have fundamentally redesigned their workflows, compared with a quarter of everyone else.⁴

Orchestrate is rare. BCG's 2025 study of 1,250 firms found that only 5% achieve AI value at scale.⁵
IBM's 2025 CEO Study reports that only 16% of AI initiatives have scaled across the enterprise.⁶ Depending on the measure, somewhere between one in twenty and one in six organisations are operating anywhere near the third tier.

The overall pattern matters more than any single metric. Investment in AI is not the primary challenge. Adoption is already widespread. The real gap emerges when AI must extend beyond an individual or a single team, leading to the central question this article addresses: What hinders further progress?

To understand what holds organisations back, it helps to start with how they grow.

Picture a baker with a stall at a busy market. Every morning he bakes, he sells, and he speaks to every customer himself. If customers want more croissants, he knows. Tomorrow, he bakes more, and he knows exactly who will buy them. When he orders flour and butter from his suppliers, he knows what demand looks like and how far ahead to plan. The whole business lives in one head. The baker is connected to every part of it.

Then the business grows. He opens a second stall, then a shop, then several. He can only be in one place at a time, so he hires people. At first, if he is lucky, he finds people like himself. As the business keeps growing, he needs specialists. Accounting is not his passion, so he hires an accountant. Managing people is not his passion either, so he hires a manager. As the shops multiply, he appoints store managers. In time, the business becomes a franchise.

This is the natural course of a growing business. It needs more people so it can keep doing what it does best, which for the baker is baking.

But delegation has a cost. The baker no longer speaks to his customers directly. He no longer manages his own supply. He relies on his managers to think as he would and to hear what he would have heard.

To make sure everyone works the way he intends, he introduces process. As the business grows, those processes become essential. They are the only way to manage growth and to see what is happening across the business.

This is what Melvin Conway observed in 1968.⁷ An organisation cannot communicate one way and be structured another. The way information has to flow dictates the structure, and every system the organisation builds mirrors that structure. Conway's Law is not a theory about bad management. It is the inevitable outcome of growth.

It does have a cost, though. As the distance between leadership and the customer grows, so do the layers needed to govern it. Some of those layers exist only to manage the process itself. They were the price of growth while communication had to pass through people.

AI changes how communication can flow. Structure has to follow.

As the baker's business grew, it split into departments: product, marketing, sales and finance. Each department now owns its own data, its own permissions and its own decisions. That is Conway's Law written into the systems.

Now add AI. The product team adopts it with enthusiasm. One person summarises customer feedback. Another drafts release notes. Before long, the team has chained these tasks into a Flow that runs from insight to launch. It is impressive work, and it delivers real gains.

Then it hits a ceiling.

The inputs that drive product's results do not sit in product. They sit in marketing: the campaigns, the audiences, the spend. The product team cannot see that data, and it has no right to change it. However capable its Flow becomes, it cannot influence the signals that determine whether its work succeeds.

Think back to the croissants. At the market stall, the baker heard demand directly and acted on it the next morning. In the grown organisation, that signal now sits on the other side of a departmental wall.

This is the silo ceiling. It is the point where Flow stops, and it is why so many organisations stall at the second tier.

Forrester has warned that AI agents do not rise above the organisation. They mirror it, and in doing so amplify its silos.⁸ This is often presented as a reason for caution. In our view, it describes the second tier exactly. AI implemented inside a silo will always be bounded by that silo.

That gives leaders a simple test:

If your AI workflow cannot read or change an input that another department owns, you are in Flow, however sophisticated it looks.

The data supports this. In IBM's 2025 CEO Study, half of CEOs said rapid investment had left them with "disconnected, piecemeal technology," and only 16% of AI initiatives had scaled across the enterprise.⁹

Breaking through the silo ceiling is not a technology decision. It is a structural one.

If the silo ceiling is structural, the way through it is structural too.

Organisations are built around people. Departments exist because people need managers, specialists need a home, and communication between people has limits. The baker could only be in one place at a time, and so could everyone he hired.

AI does not share those limits. It can carry information across departments without waiting for a meeting, a report or an approval chain. When work is organised around tasks rather than around the people who used to perform them, a process no longer has to stop at a departmental boundary.

This is where Orchestrate begins. Go back to the product team. At the third tier, campaign data from marketing, conversion data from sales and cost data from finance all feed the same process. AI coordinates what passes between them. The croissant signal reaches the oven again, but this time through the whole organisation rather than one person at a stall.

Conway's Law still applies. If communication changes, structure must follow. Conway himself noted that the structures of large systems "tend to disintegrate during development."¹⁰ As AI changes how information flows, structures built around human communication begin to loosen. Layers that existed only to manage the process lose their purpose.

Others have reached the same conclusion. Economist Erik Brynjolfsson has long argued that AI changes tasks rather than whole jobs, which pulls roles apart and rebuilds them.¹¹ Ethan Mollick of Wharton observes that "management is, in part, built around human limitations."¹² McKinsey describes a future of "flat networks of empowered, outcome-aligned agentic teams,"¹³ and Microsoft proposes replacing the org chart with a "Work Chart" built around goals rather than functions.¹⁴

This is the real shift at the third tier. AI is no longer something the organisation uses. It begins to reshape the organisation itself, in much the same way the baker once held his whole business together.

If Orchestrate is possible, why have so few organisations reached it? The barriers are real, and they deserve a fair hearing.

Accountability. Someone must answer for decisions. When a process runs across departments and AI coordinates it, leaders need to know who owns the outcome. Traditional structures give a clear answer, even if they are slow.

Regulation. Rules on data and oversight are tightening. The EU AI Act requires that people can effectively oversee high-risk AI systems,¹⁵ and organisations in the UAE must respect local data rules on where information is held and who may access it.¹⁶ Both shape how far data can move between departments.

Data ownership and internal politics. Data that belongs to a department also gives that department influence. Opening it to the rest of the organisation can feel like giving that influence away, so permissions are often guarded more for reasons of power than of risk.

Trust. Leaders are right to ask whether AI will behave as intended when it acts across the business. Trust is earned through evidence, and that takes time.

None of these is a technical limit. Each is a choice about governance and organisational design. Who owns an outcome, who may see which data and how oversight works are all decisions leaders make, and they can be made differently.

That is the central point of this article. The tools needed for Orchestrate largely exist. What is missing is the willingness to redesign the organisation around them, and that responsibility sits with the CXO.

Leaders do not need a new strategy to begin. They need a clear view of where their organisation stands today. We suggest four steps.

1. Label every AI initiative. List each one and mark it as Task, Flow or Orchestrate. Be honest. Most portfolios are mostly Task, with a handful of Flows. That is not a failure. It is a baseline.

2. Find the ceiling. For each Flow, ask where it stops. Which data does it need but cannot see? Which decisions does it depend on but cannot influence? Those points mark the silo ceiling in your organisation.

3. Redesign governance around outcomes. Choose one shared outcome, such as customer retention or margin on a product line. Give it a single owner, and grant that outcome access to the data and decisions it needs across departments. Start small, prove it, then extend.

4. Measure what matters. Track organisational outcomes, not productivity per task. Time saved by individuals is useful, but it is not the goal. A simple rule applies: if an initiative does not serve a defined company objective, do not start it.

The audit will not move an organisation to Orchestrate overnight. It will show leaders exactly where the ceiling sits, and that is the first step towards breaking through it.

The baker never needed a strategy for communication. He stood at his stall, heard what his customers wanted and acted on it the next morning. Growth took that away from him, and process was the price he paid to keep the business together.

AI offers a way to win some of it back. Task makes people faster. Flow makes teams better. Only Orchestrate changes how the organisation itself works, and it asks something of leaders that no tool can provide: the willingness to let structure follow communication.

Most organisations will stop at the silo ceiling. Not because the technology falls short, but because the structure above it stays in place. The ones that break through will be those whose leaders choose to redesign it.