Change management and AI adoption: a three-by-three framework

Executive overview

Most AI initiatives fail not because of the technology, but because of poor change management, undefined outcomes, and misaligned teams. Companies rush into AI without asking what business problem they are actually solving.

The framework has three layers: align on outcomes before strategy, place yourself on a maturity curve, and workshop use cases hands-on before committing. The core insight: alignment on outcomes must precede data strategy — not the other way around.

Common mistakes in enterprise AI adoption

• Companies lead with technology, not the business outcome
• Hallucinations and data privacy risks ignored in early LLM deployments
• Customer or proprietary data entered into public LLM tools
• No defined guardrails or success metrics before building
• "Wide-eyed wonder" phase gives way to wasted spend without learning

The three-by-three framework

Layer 1 — Alignment, outcome, strategy

• Secure internal alignment top-down before any AI initiative
• Define the business outcome first; data strategy follows from that
• Strategy must specify what data is needed, not just what tools to use

Layer 2 — Maturity model

• Stage 1: out-of-the-box features — minimum disruption, fastest adoption
• Stage 2: configuration and customisation via flows
• Stage 3: blue-sky thinking — entirely new ways of working
• Companies often see enough ROI from Stage 1 to accelerate faster than expected

Layer 3 — Workshopping

• Write up all use cases on a wall without filtering; volume matters first
• Score each by business value — cool ideas with no ROI get parked
• Let people go hands-on with the product; use cases multiply after touching it
• Take three use cases off the wall and move them into a small pilot

Iterative experimentation and pilots

• Skin in the game is required — open-ended experiments produce nothing
• Set milestones tied to real business metrics, not technical output
• Pilots with 10–100 users generate enough signal to pivot, persevere, or park
• Small initial investments often scale 10x within months once early ROI is visible
• Iterative experimentation frequently surfaces deeper issues, especially data dependencies

Pivot, persevere, or park

• Power (role authority, budget control, technical veto) is the primary filter
• Iterative experimentation produces the evidence needed to defend a persevere decision
• Park triggers: ROI doesn't justify the change-management effort required
• Pivot triggers: data dependencies or timeline constraints make a use case infeasible
• Retail example: acquired-entity data silos forced a pivot to simpler, deliverable use cases
• Banking example: ROI fell short; a simpler alternative substituted

Quick wins versus long-term transformation

• Digital transformation is a three-to-six-year journey; annual milestones sustain executive support
• Quick wins on out-of-the-box features fund and justify the larger programme
• Each milestone should produce measurable ROI that feeds back into the roadmap
• Insurance contact-centre example: annual milestones track progress toward a fully AI-powered centre

Generational shift in AI expectations

• Incoming workforce treats AI as assumed infrastructure, not a transformation initiative
• Regulated enterprises that slow-walk adoption will be outpaced by the expectations of both employees and customers
• Customer loyalty is at an all-time low; switching costs are negligible
• Companies that know their customers best — through data and AI — will retain them

Advice for corporate AI executives

• Verify internal alignment is real, not assumed — silent blockers are common
• Build cross-functional teams with genuinely conflicting viewpoints to stay grounded
• Avoid groupthink driven by C-suite enthusiasm; pressure-test every direction against business fundamentals
• Constant self-check: does this make sense for the business?