CTO / Director of Technology
AI is now a board question, and you are the one who has to answer it with an engineering organisation, a budget and a risk register attached. This playbook is the executive layer over the delivery and architecture playbooks, drawn from the same anonymised deliveries.
25+years across IT operations, delivery and enterprise change
3production platforms built and operated
12% → 87%adoption in 90 days on a global programme
5 daysgoverned knowledge hub, idea to live
The job in one sentence
Turn the board's AI question into a governed portfolio your engineering organisation can actually ship, without betting the estate on any single vendor, model or pilot.
This seat has many names
Boards increasingly split this work into dedicated seats: Chief AI Officer, Head of AI, VP of AI. The reporting line moves; the decisions below do not. Whoever holds the seat, these are the calls that cannot be delegated.
The decisions only you can make
- Build, buy or partner, per use case. The three-question test: is the capability differentiating, is the data too sensitive to leave, do you have the team to sustain it? Fewer than you think pass all three. The full test, published.
- Portfolio, not projects. One honest backlog at executive level, balanced across quick wins, enablers and bets. Competing slide decks from different departments are how AI budgets die.
- Organisation topology. Central platform team, federated champions, or hybrid: each fits a different maturity and size. The three patterns, published.
- Governance as a rhythm, not a binder: a standing cadence that says a fast, safe yes, with one-way-door decisions escalated and everything else moving. The rhythm, published.
- The vendor and model strategy. Contracts pinned to workflow outcomes, not model versions; a named exit path from every vendor in the critical chain; one gateway so switching is an engineering task, not a renegotiation. The technical shape is in the architect playbook.
The integration tax
Most AI pilots die between demo and production, and the killer is rarely the model: it is identity, data access, monitoring, compliance sign-off and support ownership, the unglamorous integration work nobody budgeted. Budget the tax up front or watch pilots stack up unshipped. The full argument, published.
The pilot was never the hard part. The last mile into production is where AI budgets go to die quietly.
The product hats
- An AI portfolio is a product portfolio, and someone must wear the product owner and product manager hats. Unstaffed, they are why gates clog and pilots drift.
- The owner hat: one value hypothesis, one backlog position and one acceptance bar per use case, held by a named person who can say no.
- The manager hat: user discovery, the roadmap across horizons, and lifecycle honesty: scale it, iterate it, or retire it. The distinction, published.
- They can be one person on a small portfolio, but the hats stay distinct: ownership without discovery ships the wrong thing well; discovery without ownership ships nothing.
The experience layer
- Adoption dies at the interface. Users who cannot see why an answer is right either over-trust it or abandon it; both end the programme.
- Trust is a UX property: citations visible at the answer, abstain states designed rather than apologised for, confidence communicated honestly.
- The review queue is a workplace. Reviewers live in it daily; design it as a workbench, not an afterthought form, or the human gate becomes the bottleneck.
- Accessibility is tested, not assumed, the same as any production surface.
Stakeholder management
- Different audiences, different questions. The board wants risk posture and the value trajectory. The CEO wants a story that survives a journalist. Finance and Risk are co-owners of the frame, not approvers at the end. Managers are quietly asking about headcount; teams are quietly asking about replacement.
- Address the fears out loud, early. The week-one fears are predictable and mostly unspoken; naming them is the cheapest trust you will ever buy. The five fears, published.
- The champions network is your stakeholder engine inside functions: credible people showing real usage beats any communications plan.
The reporting cadence
- Weekly, inside the team: the ship signal. Something visible moved; the review queue is staffed; blockers named.
- Monthly, with owners and Finance and Risk: portfolio state: gate throughput, pilots reaching production, integration-tax burn-down, model performance and cost against the eval sets.
- Quarterly, to the board: value realised against hypothesis, risk posture, the roadmap across horizons, and the retired list. The retired list is the credibility item: it proves the measurement is real.
- Mechanics live in the delivery playbook: the cadence here is what each audience sees, not how the machine runs.
Talent: build builders
- Champions chosen for credibility, one per function, growing into a network that builds inside the guardrails: the enablement playbook covers the machinery.
- Role-based enablement over generic training: each role learns the piece that changes its own workflow.
- The success test at organisational scale is the same as at team scale: the organisation keeps the capability and sheds the dependencies, including its dependency on you.
The metrics that matter
| Leading (predicts value) | Lagging (proves value) | |
|---|---|---|
| Portfolio | Gate throughput, pilots reaching production, integration-tax burn-down | Realised benefit vs hypothesis, retired initiatives with written reasons |
| Organisation | Champion coverage, sanctioned-tool usage vs shadow usage | Adoption sustained at 90 days, capability retained after partner exit |
What not to claim (to your board)
- No guaranteed savings: benefit estimates agreed with Finance, revisited at value reviews.
- No transformation timeline shorter than the adoption curve: tools deploy in weeks, behaviour changes in quarters.
- No strategy that survives on one vendor's roadmap.
Questions to ask before you start
- What did the last AI initiative teach us, and did anyone write it down?
- Which single use case, shipped, would buy the portfolio another year of board patience?
- Who says the safe yes today, and how long does it take?
- If our main AI vendor doubled prices tomorrow, what is the exit path?
Facing the board AI question? Start with the Tech Advisor Review or email hello@krishnachodipilli.com.