The Org Chart Nobody Drew

Why AI fails not because of technology, but because the organization was never redesigned for it.

My nine-year-old recently asked me why her LEGO spaceship keeps falling apart. I told her the pieces are fine. The blueprint is wrong. She looked at me like I was stating the obvious.

She is right. It is obvious. And yet, most companies deploying AI are making the exact same mistake: they keep buying better pieces while ignoring the blueprint.

Moderna merged HR and IT into a single function. Pfizer gave every department an AI mandate from the CEO. DeepL fired 21% of its workforce and called it a structural shift. Three companies. Three radically different redesigns. One shared insight that McKinsey, BCG, MIT, and RAND all confirm with hard data:

AI does not fail because of technology. It fails because the org chart was never redrawn.

The numbers that should alarm every board

Let me put this bluntly.

McKinsey's State of AI 2025 reports that 88% of organizations now use AI in at least one function. Yet only 39% attribute any enterprise-wide EBIT impact to it. Just 6% qualify as "AI high performers" with measurable value creation.

MIT NANDA's GenAI Divide study (August 2025, 300+ initiatives) is even more sobering: 95% of GenAI pilots yield zero measurable P&L impact. Only about 5% of organizations extract material value.

RAND's Why AI Projects Fail (2024, 65 expert interviews) puts the project failure rate above 80%, double the rate of non-AI IT projects.

S&P Global's Voice of the Enterprise survey (N=1,006) shows that 42% of companies abandoned the majority of their AI initiatives in 2025, up from 17% the year before.

And here is the pattern that emerges when you triangulate these studies: the discriminating factor between the 5% that scale and the 95% that do not is organizational, not technical. The same LLMs, the same cloud platforms, the same APIs. Different outcomes. The difference is the org chart.

The 10/20/70 rule most companies invert

BCG surveyed 1,000 CxOs across 59 countries for their Where's the Value in AI? study (October 2024). Their finding is now one of the most cited numbers in AI strategy:

•       10% of AI value comes from algorithms

•       20% comes from technology and data

•       70% comes from people and process

Most companies spend their AI budgets in the inverse ratio. They pour money into model selection, infrastructure, and proof-of-concept sprints. Then they wonder why the pilot that worked beautifully in a sandbox dies the moment it touches real workflows, real teams, and real decision rights.

McKinsey's relative-weights regression across 25 organizational attributes corroborates this. The single strongest correlate of EBIT impact from AI? Fundamental workflow redesign. Not talent. Not technology. Not even data quality. Workflow redesign. CEO ownership of AI governance ranks second.

Three companies that redrew the blueprint

Moderna: When you merge HR and IT, everything changes

In late 2024, Moderna did something that made traditional pharma executives uncomfortable: they merged Human Resources and Information Technology into a single function called "People & Digital Technology."

This was not a rebranding exercise. It was a structural acknowledgment that in an AI-native organization, the people strategy and the technology strategy are the same strategy.

The results were visible fast. Moderna rolled out ChatGPT Enterprise to thousands of employees. Within two months, staff had built 750 custom GPTs. 40% of weekly users were building their own. The legal team hit 100% adoption. A purpose-built "Dose ID GPT" compressed dose-selection reasoning from weeks to hours.

Moderna identified roughly 100 "Generative AI Champions" through an internal contest, creating a distributed network of AI advocates across the company. CEO Stephane Bancel's framing was deliberate: "With a few thousand employees, we are scaling like a company of one hundred thousand."

The key lesson is not that Moderna deployed ChatGPT. Plenty of companies did that. The lesson is that Moderna changed who reports to whom, who is responsible for what, and how decisions flow through the organization. The technology was the easy part.

Pfizer: A CEO mandate that changed every function

Pfizer took a different path. Rather than merging departments, CEO Albert Bourla issued a top-down mandate: every function must identify AI use cases that deliver tangible value.

The centerpiece is "Charlie," a generative AI platform built with Publicis Marcel for the entire marketing workbench. Hundreds of people across central marketing use it. Thousands across brands interact with it. But Charlie is not just a tool. It is the organizational backbone of a redesigned commercial workflow: content creation, medical-legal-regulatory review, and multichannel deployment, all flowing through one AI-augmented pipeline that slots into Pfizer's existing Veeva-based promotional review process.

Beyond commercial, Pfizer rolled out Visier to democratize HR analytics. Clinical trials got centralized statistical monitoring. Each function was expected to own its AI transformation, not wait for a central lab to hand it down.

Pfizer has not published EBIT-level attribution for Charlie. But the organizational design choice is clear: distributed ownership under a central mandate. No single AI team runs everything. Every team runs its own AI. The CEO holds everyone accountable.

DeepL: The painful version of AI-native

If Moderna and Pfizer represent redesign through addition (new functions, new mandates), DeepL represents redesign through subtraction.

In May 2025, DeepL announced approximately 250 layoffs, more than 21% of its workforce. CEO Jarek Kutylowski's published memo was unusually candid. He asked: "What does it take to operate as an enduring global AI company at this pace of change, and are we built for that?" His honest answer was: "No."

The restructuring was explicitly framed as a shift to an "AI-native" organization: smaller teams, fewer management layers, founder-mode leadership, faster decisions. This is the Embedded/Distributed archetype in its purest form.

Whether this proves to be genuine value creation or what Sam Altman called "AI-washing" of layoffs (February 2025) is genuinely contested. Challenger, Gray & Christmas attributed 54,836 announced layoff plans in 2025 to AI, roughly 5% of total cuts that year. DeepL, Coinbase, Snap, Block, Cloudflare, and Meta all cited AI in recent reductions.

The lesson for pharma is nuanced: announcing an "AI-native reorg" without underlying workflow redesign is reputationally risky and operationally hollow. The restructuring must come with actual process change, not just headcount change.

What the frameworks agree on

Across nine independent data sources spanning McKinsey, BCG, Deloitte, Accenture, MIT CISR, MIT NANDA, RAND, Harvard Business School, and primary company disclosures, three conclusions converge:

1. Workflow redesign is the dominant lever. McKinsey's regression analysis and MIT NANDA's empirical study agree: redesigning how work flows through the organization is the single strongest predictor of AI value.

2. CEO or board ownership is necessary. McKinsey ranks it the second strongest correlate of EBIT impact. MIT NANDA finds that line-manager empowerment from above is a top predictor of success. Accenture's "Reinventors" (the 9% of firms with continuous reinvention capability) have widened their gap with peers by 2.4x in revenue growth.

3. Most pilots fail because the organization was not ready, not because the technology was not ready. RAND's top root cause is mis-scoped problems (industry leadership chasing the wrong use case), followed by data foundation gaps and tooling issues. Organizational readiness is necessary but not sufficient.

Why pharma is structurally different

Three features make life sciences a special case for AI organizational redesign.

The regulatory floor is higher. The FDA's January 2025 draft guidance and EMA's first AI methodology qualification opinion (March 2025) require lifecycle credibility assessment, model lineage, explainability, and clearly delineated roles. These are not optional governance nice-to-haves. They are regulatory requirements that effectively mandate organizational redesign in clinical, R&D, and diagnostic AI.

The Veeva-centric commercial stack constrains design choices. Most pharma omnichannel, MLR/PRC review, CRM, and content workflows run on Veeva (Vault, CRM, PromoMats). Any AI redesign in commercial needs to slot into Veeva's product evolution. Pfizer's Charlie explicitly integrates with Adobe Workfront/Experience Manager and Veeva-style content supply chains.

Clinical and scientific gravity pulls AI toward the highest-value decisions. Unlike retail or banking, pharma's primary value sits in scientific decisions: target selection, dose finding, trial design. These are exactly the decisions AI is now best at augmenting. Novartis's "Frontier" intelligent decision system and Moderna's Dose ID GPT both target this layer.

Four reference architectures for pharma

Based on the evidence, four organizational models emerge as defensible starting points:

Platform/Federated Learning Ecosystem (inspired by Eli Lilly's Catalyze360 + TuneLab): Best for mid-to-large pharma seeking external pipeline access and data network effects. Lilly's TuneLab opens 18 proprietary AI/ML models to selected biotech partners via federated learning. This shifts the company from data-hoarder to platform-orchestrator.

Workforce-wide GenAI + Structural Reorg (inspired by Moderna's People & Digital merger): Best for high-ambition, digitally native organizations willing to make structural changes. Requires executive courage and a workforce that can absorb rapid change.

Federated CoE + Decision-System Integration (inspired by Novartis's data mesh + Faculty AI Frontier): Best for large, established pharma with strong existing data assets and regulatory-sensitive R&D. The data mesh provides the plumbing; the AI decision system provides the value.

Functional, Marketing-Led Platform (inspired by Pfizer's Charlie): Best for commercially oriented organizations starting their AI transformation in marketing and MLR. Lower organizational risk, faster time-to-value, but narrower initial scope.

The five questions to ask before you redesign

Before choosing an operating model, every leadership team should answer five questions:

1. Are we adopting or scaling? Below 30% employee access, adoption is the constraint. Above 50%, scaling is. The org redesign question only becomes binding at scale.

2. Is our regulated risk envelope high? If yes (pharma R&D, clinical, MLR, or any GxP/HIPAA/EU AI Act high-risk classification), a federated CoE with named human oversight and a CAIO or empowered CDAO at C-suite level is not optional. It is the regulatory floor.

3. Do we have a defensible data foundation? If not, sequence data first (6 to 12 months), then AI. RAND's number-one root cause of failure is mis-scoped problems on weak data. Do not invert this sequence.

4. Do we own the workflows we want to transform? If yes, internal build is worth considering. If no (Veeva, SAP, Salesforce, EHR systems), buy or partner. MIT NANDA's evidence: external-vendor tools succeed roughly twice as often as internal builds.

5. Are line managers empowered to redesign their workflows? If no, no central CoE will produce EBIT impact. Fix this before buying any tooling.

The uncomfortable bottom line

BCG sequences the transformation journey as Deploy (0 to 6 months), Reshape (6 to 18 months), and Invent (18 months and beyond). Most companies are stuck in Deploy, running pilots and quick wins, while expecting Reshape-level results.

The 70% of programme budget that should go to people and process? Most of it belongs in the Reshape phase: end-to-end workflow redesign on 2 to 3 anchor processes (MLR review, clinical-trial design, medical-affairs content), embedded governance, human-oversight roles mandated by the EU AI Act and FDA guidance, and a reskilling programme that makes AI a capability, not a department.

The companies in the 5% that produce real P&L impact did not get there by picking better algorithms. They got there by redrawing the org chart that nobody wanted to touch.

My nine-year-old understood this about her LEGO spaceship in about ten seconds. It took the enterprise AI market about three years and $1.5 trillion in global spend to arrive at the same conclusion.

The blueprint matters more than the pieces. Always has.

Want to find out where your organization stands? Take my free AI Readiness Assessment. Link in the comments.

Christian Schulze is Founder & Management Consultant at ImpactWorks, specializing in AI Strategy & Implementation for pharma and life sciences. He helps leadership teams redesign their organizations for AI, not just their technology stacks.