CONTACT

Building AI Strategy Part 2: Design with Clarity, Kill the Noise.

Sep 19, 2025

 

AI is easy to admire and hard to operationalize.

Most firms do the first.

You feel the pressure. You carry the risk. You also own the opportunity.

The Design phase is where you convert future-back vision into a plan that survives contact with reality. It anchors your ambition to metrics, owners, and cadences. It forces hard trade-offs. And it does it with calm precision, not noise.

Clarity beats volume. Every time.

 

The Design Objectives: Turning Vision Into Reality

This article is the second in a three-part series. In Part 1, we time-traveled to your most credible future and identified how AI could serve customers, sharpen decisions, and shape advantage. Now we land the plane.

The Design phase answers three questions:

  1. Where will AI create the most value and durability?

  2. Who is accountable for outcomes, not just activity?

  3. How will we learn every 90 days and reallocate capital with courage?

To get there, we use a balanced set of tools: lean enough to move, strong enough to align. No framework cosplay. Just artefacts that executive teams actually use.

 

Step 1 — Align AI Vision With Business Goals

You cannot pursue everything. You shouldn't. Step 1 translates your future-back vision into business-first guardrails that give teams autonomy without drift.

North Star. One sentence that states the promise you intend to win with AI. It must be a decision rule, not a slogan. "Quote in minutes, bind in hours" is a decision rule. Teams can apply it.

AI Design Principles. Five to seven non-negotiables that filter decisions. Think human-in-the-loop for high-stakes flows, explainability where regulators demand it, privacy by design, and bias checks for protected classes. Principles keep you fast and safe.

Hoshin Kanri X-Matrix (one page). Connect breakthrough goals over two to three years, this year's objectives, priority initiatives, metrics, and owners. Make conflicts explicit. If an initiative pulls margin down while boosting NPS, name the trade-off. Hoshin prevents the classic failure mode: a "sum of use cases" pretending to be a strategy.

Why does this work? Executives need to see trade-offs on one sheet. The X-Matrix forces line of sight from ambition to action. It also builds literacy in the top team. When top management understands AI better, companies demonstrate a stronger AI orientation and improved implementation ability. Strategy literacy compounds execution speed.

Deliverables for Step 1.

  • North Star (one sentence)

  • AI Design Principles (one page)

  • AI X-Matrix v1 (one page)

Clarity is a constraint. That's the point.

 

Step 2 — Structure Your Use-Case Portfolio

After guardrails, you move to portfolio design. This is where dreams meet unit economics.

Generate candidates. In cross-functional sessions, surface 10–25 use cases. No fluff. Each idea must yield a Use-Case Canvas outcome, user, workflow, scope of change, data required, model fit, integration points, guardrails, and a clear human-in-the-loop moment.

Score value and feasibility.

  • Value = frequency times time saved or quality uplift times real money. Validate with a unit of work: one claim, one quote, one order, one case.

  • Feasibility = data availability plus integration complexity plus risk and regulatory load plus change effort. Numbers, not vibes.

Balance across the waves. Map each case to the adoption horizon:

  • AI Wave 1 — Efficiency. Automate routine tasks to reduce cost and cycle time.

  • AI Wave 2 — Quality. Augment decisions to improve accuracy, consistency, and customer outcomes.

  • AI Wave 3 — Transformation. Create new systems, products, or revenue models.

Your first-pass portfolio should contain immediate cash, visible differentiation, and one credible swing at new value. If you only chase AI Wave 1, you'll be efficient and irrelevant. If you only dream in AI Wave 3, you'll burn cash and miss learning.

Run the gate with finance and risk in the room. If a case lacks data access or a KPI owner, cut it or reframe it. Mercy kills save quarters of drift.

Deliverables for Step 2.

  • Prioritized Portfolio (top 8–12)

  • Use-Case Canvases (one page each)

  • Portfolio balance across AI Wave 1, AI Wave 2, AI Wave 3

Fewer bets. Higher signal. Faster learning.

 

Step 3 — Build the AI Control Tower

Strategy without an operating model is a press release. AI without governance is a liability.

Stand up a small AI Control Tower. It coordinates standards, approvals, and shared services: prompt libraries, evaluation harnesses, model registries, and red-team playbooks. Keep it tight and senior.

Define the RACI for the model lifecycle. Who chooses a model. Who trains. Who evaluates. Who deploys. Who monitors. Who can roll back in one hour, not one week.

Compliance by design. Document data lineage, consent, fairness checks, and audit trails. Define human-in-the-loop checkpoints and absolute override authority. Don't bolt compliance on at the end. Bake it in. Launch day is the wrong day to discover the gap.

Funding logic. Central budgets fund platforms and reusable components. Business units fund their use cases. P&L ownership sharpens focus.

Minimum viable platform.

  • API-first access to core systems

  • A secure retrieval layer for enterprise knowledge

  • A lightweight evaluation suite that runs on every release

You don't need a data palace. You need enough to ship the Portfolio and learn.

Deliverables for Step 3.

  • Operating-Model One-Pager (decision rights and interfaces)

  • Governance and Risk Playbook (checklists and process maps)

  • Platform Reference Diagram (as-is and to-be, with owners)

Scaling teams do this early. It's not bureaucracy. It's your speed enabler.

 

Step 4 — Design Your AI Roadmap

Plans don't deliver. Cadence do.

Your roadmap is a 90-day engine that links strategy to outcomes.

Link Hoshin to OKRs. For every annual objective in the X-Matrix, set Quarterly OKRs with outcome metrics the board recognizes: time, cost, quality, NPS, and risk. OKRs are the execution layer. Hoshin is the alignment spine. They're complementary.

Time-boxed funding. Work in 90-day tranches with graduation criteria. Discovery to controlled pilot to limited launch to scale. Tie money to milestones. Cut or double down based on evidence.

Run a learning agenda. For each use case, list the few assumptions that must be true: data quality thresholds, edge-case accuracy, and user adoption targets. Design the smallest test to falsify a bad idea or prove a good one deserves oxygen.

Make the portfolio visible. Keep a Portfolio Kanban everyone can see: Discovery, Build, Pilot, Rollout, Scale. Transparency is a discipline. When the board asks what moved this quarter, you point, not speculate.

Deliverables for Step 4.

  • Quarterly OKR set linked to the X-Matrix

  • 90-Day Roadmap and funding gates

  • Portfolio Kanban with owner and ETA

Commit. Ship. Learn. Reallocate. Repeat.

 

Flight Levels: Keep Every Layer Aligned

If you've run transformations before, you know the altitude problem. Leaders debate strategy while teams debate tickets. The work gets lost in the middle.

Flight levels solve that. Think of three altitudes working in lockstep:

Flight Level 3 — Strategy. The North Star, principles, and X-Matrix live here. The executive team owns breakthrough goals and the few annual objectives that matter.

Flight Level 2 — Coordination. This is the cross-functional handshake. Product, data, risk, finance, and operations align on the prioritized portfolio, the operating-model one-pager, and the governance playbook. Hand-offs are explicit. Dependencies are planned, not discovered.

Flight Level 1 — Operations. This is the runway. Teams work from use-case canvases, deliver against OKRs, follow the 90-day roadmap, and update the Portfolio Kanban weekly. They own the learning agenda and bring evidence to reviews.

When tensions surface—and they will—check your altitude. Are we debating direction, coordination, or execution? Solving a Flight Level 3 issue with Flight Level 1 activity is how strategies die.

 

What "Working" Looks Like

Take a global specialty insurer I worked with. The Imagine phase produces three credible scenarios and a North Star: "Quote in minutes, bind in hours for risks we understand, with transparency customers trust."

Design translates that into:

  • X-Matrix: Breakthrough goals to cut underwriting cycle time by 70 percent and improve loss ratio by one to two hundred basis points. Annual objectives to automate intake on most submissions, deploy AI-assisted triage across key lines, and launch an explainability layer for brokers. Ownership is explicit: COO on cycle time, CUO on loss ratio, CIO and CDO on data platform and evaluation suite.

  • Portfolio:

    • AI Wave 1: Automate submission intake to pull immediate cycle time and cost.

    • AI Wave 2: Underwriting triage and price-assist for decision quality.

    • AI Wave 3: Programmatic micro-products with risk-aware pricing.

  • Operating Model: A compact control tower, a model registry, fairness checks where required, and human override points defined. Funding split: central for platform, business for use cases.

  • Execution: 90-day tranches. OKRs tied to cycle time and loss ratio. Portfolio Kanban reviewed weekly by the COO and CFO. Fewer projects. Faster learning. Numbers the board recognizes.

Fewer slides. More shipped work.

 

Final Word: From Sideshow to Operating System

You don't need a dozen frameworks. You need clarity, courage, and calm. A focused design phase, anchored on the five artefacts above, turns AI from a sideshow into a core operating system. It concentrates capital and attention on the few bets that matter. It builds governance that speeds you up rather than slows you down. It gives teams a runway to execute and your board a reason to believe.

When someone asks, "So what's our AI strategy?" you won't flash a slogan. You'll point to your X-Matrix, your portfolio across AI Wave 1, AI Wave 2, and AI Wave 3, your operating model, and your 90-day roadmap.

Then you'll ask the only question that matters: Which KPI moves first?

Your move.