TALENT on demand - Executive research
The Topology
of AI Transformation
Smaller is faster. Larger is structurally heavier.
So the question is not "how fast", it's "what shape".
Most large organisations are running the wrong play. They're trying to drag the whole enterprise up a single maturity ladder. Winners are running a different topology: an AI-native unit that grows and absorbs the core. This is the 1980 IBM PC playbook, applied to 2026 AI.
LEAPstart AI-native
SEEDone cell, then absorb
GRAFTmany cells, two speeds
The reframe

You're asking the wrong question

Most AI transformation programmes try to move the whole company up one ladder. That is the wrong frame. It treats size as an asset when size is the drag.

The wrong question
"How fast can we become AI-native, end to end?"
Assumes one ladder, climbed linearly, by every function, every team, every person. It implies that big organisations are further ahead because they have more scale and process depth. It ignores that a 40-year-old ERP estate, union agreements, and entrenched process ownership are not a foundation - they are a ceiling.
→
The right question
"What topology fits our size and our legacy?"
Do we leap to an AI-native model (small firms, clean slate)? Do we seed one AI-native cell and grow it into the core (mid-market)? Do we graft multiple cells onto the enterprise and run two operating speeds in parallel (large organisations)? Size determines the shape of the answer, not the speed.
The counterintuitive truth

Smaller is faster. Bigger is heavier.

Old-style capability maturity rewarded scale. AI-native maturity rewards the absence of legacy. The relationship has flipped.

A solo founder with a laptop and a handful of AI tools is already running an AI-native operating model. A 100,000-person enterprise needs 3 to 7 years to get to "organised" - let alone "native".

Small firm advantages

Near-zero legacy drag
  • No legacy data estates to integrate or clean
  • No entrenched process owners to displace
  • No works councils, no union agreements
  • Cloud-native stack, low switching cost
  • One or two decision-makers, minutes to yes
  • Headcount redesign is hiring differently, not layoffs
  • Identity is "I run a business with AI", not "my job is X"

Large firm drag

Legacy compounds with scale
  • Decades of ERP, bespoke integrations, on-prem systems
  • Process ownership wars across BUs and functions
  • Labour law, collective bargaining, consultation
  • Data fragmented across hundreds of systems
  • C-suite agreement is a multi-quarter exercise
  • Headcount redesign = restructuring with severance cost
  • Career ladders built around role, function, grade
The template that already works

Borrow the 1980 IBM PC playbook

The mainframe core could not have built the PC. So IBM didn't ask it to. Twelve people, a separate building, a different P&L, permission to break every rule. Then reverse absorption.

1980
Boca Raton, Florida

IBM's mainframe business had the wrong cost structure, wrong distribution model, wrong customers, wrong incentives to build a personal computer. If IBM had asked the core to build it, the core would have killed it - not out of malice, but out of metrics.

So IBM created a 12-person cell, 1,200 miles from HQ, with its own P&L and permission to violate IBM norms: off-the-shelf Intel CPU, Microsoft operating system, open architecture, retail distribution, no IBM blue-suits.

The PC shipped in 15 months. Then its operating model was grafted back into the company.

What this means for AI transformation

1. Ring-fence
The core will kill disruptive innovation by default. Not out of malice - out of metrics, incentives and identity.
2. Autonomy
Own P&L, own hiring rules, own stack, own metrics. Reports to CEO or board, not to a function head.
3. Prove first
Win real economics in a beachhead before scaling. Cells that don't prove themselves get killed, not protected.
4. Reverse absorption
The mothership adopts the cell's operating model, not the other way around. That is the whole point.
The three shapes

Three transformation shapes. Pick yours.

Not every company climbs the same ladder. There are three shapes - determined by size and legacy - and each needs a different playbook, timeline and partner.

01

LEAP

Start AI-native
No legacy to unwind. Build AI-native from day one: the operating model IS a digital workforce. Your ceiling is narrow, but the model is real. A few people plus agents delivering what used to need dozens.
WhoSmall firms, startups, spinouts
Timeline0 to 6 months
ScopeNarrow, single business
Core moveDesign around agents from day one
02

SEED

One cell, then absorb
Pick one beachhead (one product line, one function, one region). Make it AI-native. Run an enterprise hygiene programme in parallel. When the cell proves the economics, absorb its operating model into the core.
WhoMid-market, scale-ups
TimelineCell in 12 months, absorb in 2-3 years
ScopeOne function or line, then the rest
Core movePick beachhead, win, absorb
03

GRAFT

Many cells, two speeds
The IBM PC play at scale. Stand up multiple autonomous AI-native cells. Run an enterprise programme in parallel to lift the core. Graft the new operating model onto the core one function at a time. Some legacy never fully converts - and that is fine.
WhoLarge enterprises, regulated incumbents
Timeline3 to 7 years
ScopeMultiple cells + core programme
Core moveSeed cells, prove, graft back
The picker

Which shape fits you?

Size is a proxy for legacy load. Regulated industries (banking, insurance, health, public sector) carry more drag per head than unregulated (tech, media, services). The picker reads both.

Company size Light legacytech, media, services Medium legacyretail, manufacturing, logistics Heavy legacybanking, insurance, health, public
Smallunder 2,000 01 - LEAP 01 - LEAP 02 - SEED
Mid-market2,000 - 20,000 02 - SEED 02 - SEED 02 / 03 hybrid
Large20,000 - 100,000 02 / 03 hybrid 03 - GRAFT 03 - GRAFT
Giantover 100,000 03 - GRAFT 03 - GRAFT 03 - GRAFT (federated)
LEAP - start AI-native SEED - one cell then absorb GRAFT - many cells, two speeds
The function lens

Not all functions transform at the same speed

Each function has different data readiness, regulatory weight, and AI impact potential. The sequence matters.

IT/Service Desk
Wave 1
Procurement
Wave 1
HR Operations
Wave 1
Finance
Wave 2
Sales
Wave 2
Marketing
Wave 2
Customer Service
Wave 2
Supply Chain
Wave 3
Production
Wave 3
R&D
Wave 3
Legal/Compliance
Wave 4
Strategy/M&A
Wave 4
The sequence

Four waves, function by function

GRAFT does not happen all at once. The enterprise transforms in waves.

Wave 1
Foundation
Months 0-6
"Prove the model works"
Functions: IT/Service Desk, Procurement, HR Operations. Build quick wins in support functions with clean data and minimal regulatory friction.
Wave 2
Commercial
Months 6-18
"Touch the revenue line"
Functions: Finance, Sales, Marketing, Customer Service. Prove AI impact on revenue-facing functions. More complex data, higher regulatory weight.
Wave 3
Operations
Months 12-30
"Transform the engine"
Functions: Supply Chain, Production, R&D. Reshape core operations. Heavy data integration, long proving cycles.
Wave 4
Strategic
Months 18-48
"Transform the brain"
Functions: Legal/Compliance, Strategy/M&A, Business Unit P&L. Highest regulatory weight, longest timelines, most governance required. Run in parallel only with mature programme.
The operating model

Cells map to functions

The number of cells depends on company size. Each cell is dedicated to one or more functions.

10K employees
Light-to-medium legacy
Cells active 3-5
Focus Some cross-functional
20K employees
Medium-heavy legacy
Cells active 5-8
Focus One cell per function
50K+ employees
Heavy legacy, federated
Cells active 8-15+
Focus Dedicated per region/BU
Total programme: 15-80 cells created, grafted, and replaced over 3-5 years
The operating model

Two speeds, running in parallel

For SEED and GRAFT, the enterprise runs two speeds at once: an AI-native cell that builds the future, and a core programme that lifts the rest.

Speed 1 - the cell

AI-native from day one

  • Ring-fenced team, 10 to 30 people, own P&L
  • Own stack, own hiring, own metrics
  • Permission to violate enterprise norms
  • Separate location or floor, minimum
  • Reports to CEO or board directly
  • Kill criteria and absorption path set up front
↔
Speed 2 - the core

Hygiene at enterprise scale

  • Entire enterprise, waves by function or BU
  • Existing stack, existing governance
  • Data readiness, AI literacy, policy, use-cases
  • CTO / COO / CHRO coalition sponsored
  • 2 to 3 year rhythm, change-management heavy
  • AI copilots layered onto existing workflows
Reverse absorption: the cell's operating model is grafted onto the core function-by-function. The enterprise transforms as the share of the business running on the new model grows. Some legacy stays on the old model forever - that is an acceptable portfolio outcome, not a failure.
The partners

Two kinds of partner. You need both.

Cells and core programmes need different DNA. Strategy boutiques build cells fast. Big 4 and integrators run enterprise programmes at scale. Don't mix them up.

For the cells

Speed, design, AI-native DNA
McKinsey (QuantumBlack)
Cell architecture, AI-native operating model design, C-suite strategic clarity. Strong on the "why" and the end-state.
BCG (BCG X)
Build-operate-transfer cells, internal venture studios, cell incubation. Strong on productisation.
Bain (Vector)
AI-native commercial ops with PE-style P&L discipline. Strong on delivering results.
AI-native boutiques
Thoughtworks, a.team, Valtech and similar. They build the cell, staff it, hand it over.

For the core

Scale, governance, hygiene
Deloitte
Largest transformation practice. Data foundations, governance, responsible AI, workforce of the future.
PwC
Regulatory, risk, workforce strategy, cross-border compliance. Best fit for heavy-legacy sectors.
EY
Workforce analytics, people transformation, tax and legal restructuring of cells and spinouts.
KPMG
Governance, audit-ready AI, regulatory navigation. Where responsible AI is non-negotiable.
System integrators
Accenture, Capgemini, IBM, TCS, Infosys. Core platform build, data lake, managed services.
The make-or-break

Four questions that separate winners from museums

The cells that succeed all answer these four questions the same way. The ones that become a museum of pilots answer them the other way.

01

Does the CEO own it personally?

Winners: CEO is the sponsor, reviews quarterly, hires the cell leader personally, protects the cell from antibodies.
Museums: Delegated to a function head. Cell starves of air cover the moment priorities shift.
02

Is the cell actually ring-fenced?

Winners: Own P&L, own hiring rules, own stack, own metrics. Physically or legally separated. Different people.
Museums: "Virtual" team embedded in the core. Same approvals, same budgets, same people. Same results.
03

Is the absorption path designed up front?

Winners: Day 1 decision on when and how the cell's model gets grafted back. Named sponsor in the core for the receiving side.
Museums: "We'll figure it out when it works" - which means the cell becomes a permanent silo nobody can touch.
04

Will you actually kill what doesn't work?

Winners: Explicit kill criteria, reviewed at 6 and 12 months. Failed cells are closed, not relabelled.
Museums: Failed cells get renamed, re-scoped, re-launched. Five years later there are 12 pilots and zero absorbed.
Your next move
depends on your shape
Not every company needs the same programme. Start with the topology. Then we'll build the playbook.
01 - LEAP
Small firms, startups

Design your operating model around agents now

Stop hiring for old roles. Start with the AI-native org chart and hire against it. We help you architect the agent-first operating model in weeks.
02 - SEED
Mid-market, scale-ups

Pick the beachhead and stand up the cell

Choose one function or product line with clean data and an impatient leader. We help you scope the cell, hire the leader, and write the absorption plan on day one.
03 - GRAFT
Large enterprises

Run two speeds with a clear topology

Board mandate, Chief Digital Orchestrator, cell portfolio, enterprise programme. We help you architect the topology and sequence the grafts.

Let's pick your topology and get moving.

A 60-minute session with TALENT on demand. We read your size, your legacy load, your constraints - and give you the shape that fits.

TALENT ON DEMAND
philip.de_neve@talentod.com www.talentod.com
TALENT on demand
1 / 13