Capstone: The AI roadmap

The capstone assembles everything into the document you were implicitly assigned in lesson one: the AI roadmap — 12–18 months, board-ready, executable. Not a vision deck; an operating plan with owners and dates. Build it for your real organization (preferred, sanitized as needed) or complete the Alder case. Structure it as the answer to five questions a board actually holds:

1. 1'Why this, why now?' (1 page) — the reframe from lesson one applied to your company: the 2–3 business priorities AI leverages, the competitive context from your scan, and the honest calibration sentence (what AI can and can't do for you this year). No technology tour; the board doesn't need to know what a token is.
2. 2'What are we doing?' (2 pages) — the portfolio memo, matured: the funded initiatives with owners, dates, and kill criteria; the proportions sentence; the sequencing-for-compounding logic; the data-foundations investment framed as pipeline. Exhibit A is the 2×2. This section is artifact #2, and it should read like capital allocation, because it is. Sum the five cost lines into one program-budget number the board can approve — a low/likely/high year-one range, not false precision. Alder's illustrative year-one total: ~$0.6M low / ~$0.85M likely / ~$1.2M high, of which ~$400K is the new-envelope ask and the rest is reallocated (the deferred TMS upgrade plus existing budgets). When the board asks Jordan 'how much, total?', the answer is one sentence: 'Plan for roughly $0.85M in year one — call it $0.6M to $1.2M depending on how the ETA bet and the data cleanup land — about $400K of it new money, the rest reallocated.' A range, honestly bounded, beats a single number nobody believes.
3. 3'How are we sourcing it?' (1 page) — the build/buy verdicts for the top initiatives with their one-line rationales, the capability-location paragraph (where does our AI knowledge live?), the shared-dependency flags, and the POC discipline as standing policy ('we test on our data, against pre-written criteria, before we sign').
4. 4'How do we stay safe?' (1 page) — the governance charter, presented as an enabler: the appetite statement verbatim, the five structures in one diagram, the regulation tiering summary, and the calibration gauges that keep it lean. The risk-minded board member should finish this page recruited, not merely reassured.
5. 5'How does the organization come along — and how will we know?' (1.5 pages) — the stakeholder map's headlines (sponsors briefed for the middle, skeptics recruited, the jobs conversation already had — quote your own messaging), the role-based enablement plan tied to the rollout sequence, the champions network, and the three-layer metric stack with the quarterly review cadence. Close with the trajectory you're committing to: what the board will see at each quarterly, and the first review's date.

Every prior workshop ended in a finished artifact; so does this one. Below is the finished first page of Alder's roadmap — not the outline, the real thing — so you can see the standard the other pages must match. Write yours to this level of specificity, then run the whole document through the red-team.

ALDER LOGISTICS - AI ROADMAP FY26           J. Okafor, SVP Operations
=====================================================================
WHY THIS, WHY NOW

Our strategy hasn't changed: win regional contracts on reliability,
defend margin against the national carriers, keep our drivers. This
plan adds AI leverage to those three - it is not a fourth strategy.

Three priorities, three levers:
  - RELIABILITY  Predictive ETA on our top-20 lanes attacks the
                 churn we lose to missed promises. (Strategic bet.)
  - MARGIN       Exception triage, quote extraction, dispute drafts,
                 contract review - documents-in-motion, automated -
                 absorb volume growth without the 2 planned CS hires.
  - DRIVERS      Driver-log audit assist keeps compliance clean;
                 the human decision stays human, and documented.

Honest calibration (what AI can and can't do for us this year):
  Today's AI is reliably good at reading messy documents, drafting,
  triaging, and predicting from decent history. It is usually right,
  not always - so every initiative here is drafted-and-reviewed or
  triaged, never unattended on anything touching money, people, or
  safety. We are buying deployable wins, not a demo.

What we're asking for:
  ~$0.85M year one (range $0.6M-$1.2M); ~$400K new, rest reallocated
  from the deferred TMS upgrade. 8 funded initiatives, each with a
  kill date. First proof point: exception triage auto-routing 60%
  of the CS inbox by Sep 15.
=====================================================================

Presenting it (the last mile is spoken)

• Pre-wire the decision — no board vote is won cold in the room. Before the packet ships, brief the three people who move the vote one-on-one: the board chair (walk the shape, ask what would make them champion it), the risk-minded member (hand them the governance page early — recruit them, don't surprise them), and the CFO (confirm the funding envelope and the reallocation, so the number is pre-agreed). Walk into the meeting with the outcome already mostly decided; the meeting ratifies, it doesn't discover.
• Deliver it in 15 minutes, answer-first: the ask (funding, the governance mandate, the sponsor commitments), then the portfolio, then everything else as backup. Boards fund confidence built on visible rigor — the rigor is your appendix; the confidence is your sequencing.
• Pre-empt the three predictable challenges: 'why not faster?' (change capacity is the constraint — show the math), 'why not cheaper?' (the maintenance and enablement lines are why this will still be running in two years — point at the graveyard of programs that skipped them), and 'what if the technology shifts?' (the portfolio's options bucket and quarterly review are the adaptation mechanism — you built for drift).
• Run the red-team first: paste the full roadmap into your AI assistant as the hostile board member, the skeptical CFO, and the fearful employee-representative. Three passes, three revision rounds. A strategy document that hasn't survived simulated hostility shouldn't meet the real kind.

Certification rubric

• Opportunity & portfolio rigor (30%) — inventory built from real passes, scores defensible, portfolio proportioned and sequenced, every kill criterion written and dated.
• Economic honesty (25%) — five-line cost models with ranges, levers with capture rates and haircuts, baselines planned before launch, the hostile-CFO pass visibly survived.
• Sourcing & governance judgment (25%) — build/buy verdicts that apply the two-part test, a POC spec with pre-written criteria, a charter whose appetite statement can classify the whole portfolio, regulation tiering done.
• Change & measurement design (20%) — the jobs conversation drafted honestly, enablement role-based and rollout-sequenced, three-layer metrics with a quarterly review that can kill things.

Passing earns the AI Strategy Leader credential (ID format EDOVA-STR-2026-XXXX, independently verifiable at edova.ai/verify). It attests the full arc: turning 'what's our AI strategy?' into a funded, governed, measured, organizationally-honest operating plan.