Every enterprise I talk to is somewhere on the same journey: pilots have been run, a few demos impressed the board, and now there's pressure to "have an AI strategy" with real money attached. The temptation is to start with vendors and models. The organizations that succeed start with five questions instead — and answer them honestly before a single rupee, dollar, or euro is committed.

This is the blueprint I use in enterprise engagements. It's deliberately vendor-neutral, and none of it requires believing anyone's marketing — including mine. Use it internally, freely. If you want help running it, that's what we do.

Question 1: What altitude is the use case?

Most AI use-case lists mix three very different altitudes: assist (AI helps a human do their existing job faster), automate (AI executes a bounded task end-to-end with human checkpoints), and autonomous (AI pursues a goal across systems with minimal supervision). Each altitude has different economics, different risk, and radically different governance needs.

The most common strategic error in 2026 is funding an autonomous-altitude ambition with an assist-altitude readiness level. Gartner's warning that over 40% of agentic AI projects will be canceled by end of 2027 — for escalating costs, unclear value, and inadequate risk controls — is largely a story of altitude mismatch. Map every proposed use case to an altitude first. Fund the ones where your readiness matches the ambition.

Question 2: Does the data actually exist, and who owns it?

Not "do we have data" — every enterprise has data. The real questions: Is the data the AI needs accessible at the point of the workflow? Is it current enough to be trusted? And is there an owner who will answer for its quality next quarter? MIT's research on failed pilots found poor data quality and weak system integration were the most common root causes — not model performance. If the data answer is fuzzy, the pilot result will be fuzzy, and fuzzy results get defunded.

Question 3: Build, buy, or assemble — decided on what evidence?

MIT's data shows externally-partnered AI deployments succeed at roughly twice the rate of internal builds. That doesn't mean "always buy" — it means the burden of proof sits with building. A defensible build case needs at least one of: data or workflows so proprietary that no vendor can serve them; AI as an actual differentiator in your market (not just an efficiency); or an engineering bench with real production-ML experience, not aspirations.

Ask any vendor one question: "Show me this working at production volume in an organization shaped like mine." The quality of the silence tells you most of what you need.

Also budget for the part nobody demos: inference costs at production scale. Analyses of agentic project cancellations consistently find production token costs landing 5–10× above pilot-scale projections. Get the unit economics in writing before the contract, not after the invoice.

Question 4: What is the governance model — today, before deployment?

Governance written after an incident is always panic governance. The pre-deployment version needs to answer, on one page: Who approves new AI use cases? How are models, prompts, and data reviewed, and how often? What can the AI touch at each autonomy level — and critically, is access scoped per-agent rather than uniformly? Gartner's 2026 guidance is blunt on this: uniform governance across agents with different autonomy levels is itself a failure mode. And with Forrester predicting 60% of Fortune 100 companies will appoint a head of AI governance in 2026, the market has already voted on whether this is optional.

One more uncomfortable item: shadow AI. Your employees are already using AI tools you haven't sanctioned — and clamping down harder just drives usage underground. A workable governance model gives people a sanctioned path that's easier than the shadow one.

Question 5: Who stands up in front of the board when this works — or doesn't?

Every AI initiative that survives year one has an executive owner whose credibility is attached to a business metric — cost per ticket, cycle time, Mean Time to Resolution — not a model metric. If the answer to "who owns the number?" is a committee, you don't yet have a strategy; you have a science fair. The final step of any readiness exercise is converting the roadmap into a funded mandate with a name on it. That's an alignment workshop, not a document — documents that sit in shared drives don't create ownership.

Scoring it: give your organization one point per question you can answer in writing today. 4–5: you're ready to scale, and your constraint is probably sequencing. 2–3: fund one workflow, fix the gaps in parallel. 0–1: your first AI investment should be readiness, not licenses — it's cheaper by an order of magnitude.

This blueprint is the skeleton of AIAUTOMIC's enterprise engagement: an AI opportunity map ranked by effort and impact, a build-vs-buy assessment with the trade-offs made explicit, a governance framework sized to your actual risk, and the executive alignment workshop that turns it into a funded mandate. Grounded in five-plus years shipping Generative AI inside ServiceNow's platform — where I learned most of the above the honest way.

Book a call and bring your answers to the five questions — even partial ones. We'll tell you plainly what's solid and what isn't.