AI strategy begins when a vision, use cases and operating choices hold together.
The customer promise shapes the portfolio. The portfolio exposes the capabilities you need. Those capabilities determine the operating model, funding logic and roadmap. Break one connection and the strategy becomes a collection of attractive slides.
This is the job of Design.
Imagine chooses a future worth pursuing. Design converts that future into a small system of decisions. Run tests the system against production evidence.
The handoffs between those three jobs are where the strategy is most vulnerable.
Make the choices agree
An executive team may approve a strong AI ambition on Monday and twenty unrelated pilots on Tuesday.
The ambition says customer intimacy. The portfolio funds back-office automation. The principles promise human oversight. The roadmap has no owner for it. Finance asks for returns by initiative while the technology team builds shared capabilities that no single initiative can fund.
Every document looks reasonable on its own.
Together, they contradict one another.
Design should make those contradictions visible before delivery begins. It creates traceability from a future business choice to the work, money, authority and evidence required now.
The sequence below is an operating design. Companies may use different workshops, artifact names and review cadences. Use the lightest form that preserves the decisions.
Make five decisions connect
- Write a North Star that can reject work. State the customer or business promise and the boundary around it. “Use AI everywhere” accepts everything. “Resolve standard claims in hours while preserving expert review for complex losses” tells teams what fits and what does not.
- Turn principles into trade-offs. A principle should settle a real tension. Human review for high-consequence cases. Reuse before custom build. Evidence before scale. A sentence that never changes a decision is only brand language.
- Build one governed portfolio. Compare candidate initiatives by their contribution to the North Star, expected value, workflow change, data readiness, risk and shared dependencies. Preserve a mix of near-term improvement and longer-term advantage, but keep the number small enough to fund properly.
- Design authority beside the work. Name who funds, starts, pauses, stops and rolls back each initiative. Define which capabilities are shared and which belong to a business unit. Governance added after the portfolio is chosen becomes an approval queue. Governance designed with it becomes execution infrastructure.
- Fund learning through explicit gates. A roadmap should state what the organization must learn next, what evidence unlocks more capital and what result ends the work. The length of a funding tranche follows the risk and learning cycle. Ninety days can be useful. It is not a law.
Each decision constrains the next. That is what gives the strategy shape.
A strong North Star with a weak portfolio produces scattered delivery. A strong portfolio without decision rights produces delay. A roadmap without evidence gates funds optimism. Coherence closes those gaps.
Pressure-test the design
Consider a fictional specialty insurer.
Its chosen future is simple: standard risks move from submission to a transparent quote within hours, while specialists keep authority over complex cases.
That promise rejects a generic chatbot programme. It points instead to submission intake, document extraction, risk triage, underwriter support and a clear explanation for brokers.
The principles settle the difficult choices. Automation stops when the evidence is weak. A human owns complex-risk decisions. Shared data and evaluation capabilities receive central funding. Each insurance line funds the workflow change needed to use them.
The portfolio now has a spine. Intake creates structured information. Triage uses it. Underwriter support builds on both. The explanation layer serves the whole journey. Dependencies become visible before four teams buy four incompatible tools.
The operating model names the people who can pause a bad recommendation and restore the previous process. The roadmap begins with the assumptions that could invalidate the design: document quality, broker acceptance, underwriter trust and the cost of changing the workflow.
No result is claimed here. This is an illustrative case showing how the five decisions should reinforce one another.
Find the first contradiction
Put your current AI ambition, principles, portfolio, operating model and roadmap on one table.
Trace one funded initiative across all five.
Which customer promise does it advance? Which principle shaped its design? What shared capability does it depend on? Who can stop it? What evidence earns the next funding decision?
Start with the first unanswered question.
You may discover that an initiative has no strategic parent. Stop it. You may find that three initiatives need the same data or evaluation layer. Fund the shared capability. You may find that the roadmap has dates but no learning decisions. Rewrite the gates.
AI strategy survives contact with reality when every important choice can be traced to the next one.
Remove the contradiction before adding more detail to the plan.
Your move.