Three strategic shifts that separate AI leaders from the rest

By Nizar Hneini and Gianfrancesco Martinico

AI fails not because the technology is unready, but because organizations continue applying new capabilities to old structures: automating obsolete processes, optimizing within silos, deploying tools that any competitor can replicate. Understanding the failure modes is necessary, but it is not sufficient. The more pressing question is what a different approach actually looks like and what it requires of leadership.

Shift 1: Ask what should be deleted, not what should be automated

The instinct to redesign before automating is sound but incomplete. The more powerful starting question is whether a process should exist at all.

Most organizational processes were not designed to create value. They were designed to manage constraints, human error, slow systems, unreliable data, many of which have long since disappeared. Applying AI to these processes does not unlock value; it preserves dysfunction at greater speed and cost. The fact that funding so frequently stops after the pilot phase, a pattern our “ AI across the Gulf: From ambition to scalable impact ” report identifies, often reflects this: organizations measure activity rather than value, because the underlying process was never designed to generate it.

The diagnostic is straightforward: trace each workflow back to its origin and ask what specific constraint it was solving, and whether that problem still exists. This step consistently surfaces surprises: approval layers introduced after a single incident decades ago, never revisited; reconciliation workflows built for legacy systems replaced years ago; review cycles designed around data quality issues that modern governance resolved long ago.

The harder question follows: would the business notice if this process simply disappeared? Running each workflow through four filters provides the necessary clarity:

1. Regulatory necessity
2. Direct customer value
3. Financial protection
4. Competitive differentiation

Most organizational processes fail all four. They persist not because they create value, but because no one has stopped to question them.

What survives this scrutiny falls into three categories. Some processes are fundamental, they define what the business does and require reimagining from scratch for an AI-native environment, not a layer of automation on top. Others made sense in a prior technological era but are now redundant: multi-step approval chains that AI can collapse into a single action. And some carry no current value whatsoever and should be deleted entirely. What organizations consistently discover is that the truly irreplaceable processes are far fewer than assumed.

Only once that ground is cleared does it become possible to build something substantively better: processes designed around outcomes rather than inherited steps, running activities in parallel rather than in sequence, and reserving human judgment for genuine strategic exceptions.

Shift 2: Stop trying to fix silos. Dissolve them.

The standard response to silo problems is better collaboration: more cross-functional meetings, shared KPIs, matrixed reporting structures. This misdiagnoses the problem and treats the symptom rather than the cause.

Functional silos did not emerge from poor organizational design. They emerged from real constraints: humans coordinate complexity only within small groups, information moved slowly, and specialization was the only viable path to efficiency at scale. The functional hierarchy was a rational solution for its era. It was simply never designed to optimize for end-to-end business outcomes.

AI removes those constraints. It coordinates thousands of variables simultaneously, optimizes across the whole organization rather than locally within a department, and executes in milliseconds. The architecture that made sense in the pre-digital era has become the primary bottleneck to value creation.

The productive question is not how to get functions to collaborate more effectively. It is what the organization would look like if it were structured around outcomes rather than capabilities. The answer points toward cross-functional teams accountable for business results: a team responsible for total capability cost and availability, drawing on workforce strategy, vendor management, financial analysis, and operations as a single unit; a team mandated to automate routine transactional workflows entirely, freeing talent for higher-value work; a team that sets the parameters within which AI operates autonomously and manages the exceptions requiring genuine human judgment. This shift goes beyond reorganization. It is a different answer to the foundational question of how expertise creates value in an AI native organization.

Before pursuing this transition, however, organizations must be candid about their actual capacity to make it. Not their cultural readiness in the abstract, but the concrete mechanics that will govern pace. How quickly can skills be redeployed? How modular are existing processes? How readily can data move across functions? Organizations that cannot answer these questions with precision tend to underestimate the friction ahead and overestimate how quickly meaningful structural change can occur.

Shift 3: Build the organizational genome before commoditization arrives

AI commoditization is approaching faster than most leadership teams recognize. The same powerful models, scalable infrastructure, and technical talent will soon be broadly accessible to every organization in every market. At that point, having AI will not be a differentiator. What will distinguish organizations is what their AI knows that con competitor’s system does.

This is the organizational genome: the contextual knowledge that transforms a generic AI system into one that actually understands a specific business. Not just the data an organization holds, but the judgement behind how it is used; why certain customer relationships work in your market; what past product launches revealed that no dataset can capture; how the organization navigates genuinely ambiguous decisions.

This kind of knowledge does not live in documents. It lives in the judgment calls of experienced practitioners, the stories behind past decisions, and the patterns that only become visible after years operating in a specific market. Capturing it requires different methods: AI systems that learn by observing expert performance in real time, communities of practice where tacit knowledge flows through ongoing dialogue, feedback loops where AI-generated decisions receive expert review that becomes new training data.

The practical difference this makes is concrete. An AI system informed by this kind of intelligence does not optimize a vendor negotiation for the lowest price in isolation. It incorporates the organization's history with that vendor, its preferred partnership dynamics, and its long term strategic intent. That’s not generic. That’s proprietary and it compounds over time in ways that no competitor can quickly replicate.