Ground your AI strategy in reality.

1. 01

   It automates tasks, not roles

   AI takes over discrete, well-bounded tasks and augments the people around them. Whole-job replacement is the exception, not the plan.

2. 02

   Run costs scale with use

   Budget for operating AI, not just building it. The more it is used, the more it costs — value has to outgrow consumption.

3. 03

   Judgment must be made explicit

   AI applies the rules you can write down. Every decision it takes needs documented criteria, thresholds and human escalation paths.

4. 04

   Productivity is not savings

   Time saved becomes money saved only when processes, structures and service models are redesigned to capture it.

1. 01

   Who owns the decision?

   Agree where AI decides, where it recommends, and where a person must sign off.

2. 02

   Whose data, whose duty?

   Data quality becomes a shared obligation — name owners and standards up front.

3. 03

   Policy meets engineering

   Guardrails, data use and outputs are policy choices, not just configuration.

4. 04

   Shared credit, shared blame

   Agree in advance that wins and failures belong to the coalition, not one function.

Organisations that write these agreements down move faster — and their leaders emerge as the change authority.

1. 01

   Sponsors

   Fund the programme and defend it upward.

   What they need: A credible case, staged milestones, and value stories they can retell to the board and investors.

2. 02

   Builders

   IT, data and product teams who deliver.

   What they need: Clear ownership of data and integration, shared standards, and protection from scope whiplash.

3. 03

   Guardians

   Risk, compliance, legal and audit.

   What they need: Early involvement, documented guardrails, and co-ownership of governance — not veto by ambush.

4. 04

   Adopters

   Frontline teams whose work changes.

   What they need: Training, clarity on what changes for them, and visible personal wins — not just corporate ones.

1. 1 · GATE

   Assess

   Use-case triage against your data, systems and regulatory context. Output: a ranked portfolio and honest feasibility, not a wish list.

2. 2 · GATE

   Prove

   A production-grade pilot with success criteria agreed before build — accuracy, containment, cost and risk thresholds.

3. 3 · GATE

   Scale

   Multi-agent orchestration across functions, with integration, data contracts and human-in-the-loop controls hardened.

4. 4 · GATE

   Sustain

   Run-cost management, model monitoring, retraining cadence and a value dashboard the executive team actually reads.

Each phase ends at a gate: proceed, adjust, or stop — with accepted risks documented and owned.

1. 01

   Gate criteria

   Define what earns the next phase of investment — accuracy, containment, cost per interaction, risk posture.

2. 02

   Success metrics agreed up front

   No pilot starts without a written definition of success, its baseline, and who measures it.

3. 03

   A living risk register

   When you must proceed under pressure, do it with documented, accepted risks — signed, dated, owned.

4. 04

   Staged deployment

   Ship in controlled increments with rollback paths; speed without staging is how programmes lose a year.

5. 05

   One story to leadership

   Requirements, timelines and outcomes reported consistently by the whole coalition — no competing narratives.