1
The Driver
Why now — and does anyone agree what success means?
—
Clarity
What is the primary force behind this?
This sets the branch. The stated driver and the real driver are often different — that gap is the first thing worth naming.
Is there a defined, agreed success criterion?
"Do AI" is anxiety with a budget. A measurable outcome is a goal.
Does the stated driver match the real one underneath?
Naming this gap in the first two weeks is the highest-leverage move you make.
Do the leaders funding this agree on why now?
Three executives can all nod at "we need AI" and mean three different things. A shared, explicit rationale is what keeps a program from fracturing in month two.
Is there a named executive sponsor?
An accountable owner, not just a mandate. No sponsor is a clarity problem, not a build problem.
Is there funded budget tied to outcomes?
A dated roadmap with money attached to a specific result — or just intent?
Are success metrics defined?
A number you'd measure the initiative by. "Do AI" with no metric is anxiety with a budget.
2
The Foundation
What they have — readiness for AI work specifically
—
Readiness
Data readiness — accessible, governed, usable?
AI ambitions die on data reality far more than on model choice. This is the usual real blocker.
Have you done a data-cleansing exercise?
Models inherit their data. Profiling, de-duping and fixing quality before AI is the unglamorous step that decides whether a pilot survives production.
Platform, identity & security posture
AI adoption without an identity/security layer is how you get the data-leak incident.
AI access — how do people reach models today?
Ungoverned access is the shadow-AI condition, even when sprawl isn't the stated driver.
Can AI reach the systems where the work actually happens?
A model that can't touch the CRM, the ERP, or the ticket queue stays a demo. Integration — not the model — is where most pilots quietly stall.
Do you validate model quality on your own data before trusting it?
Vendor benchmarks are a starting point, not evidence for your setting. A model that shined elsewhere can miss on your data — validate first, then monitor as data drifts.
3
Governance & Responsible AI
The guardrails — policy, risk, accountability
—
Readiness
Is there a documented AI / acceptable-use policy?
A rule people can actually follow — not an unwritten norm.
Can you guarantee no PCI data is mixed into the data used?
Cardholder data reaching a model is a PCI-DSS exposure. You need to know it's identified and excluded or segregated before any data feeds AI.
Can you ensure restricted PII won't violate state or local law if used?
State and county privacy rules (CCPA/CPRA and others) restrict how personal data may be used. Unmapped PII obligations are a governance gap AI exposes fast.
Is a human in the loop on consequential or irreversible AI outputs?
Speed is safe where mistakes are cheap to undo; on anything that creates an obligation or a hard-to-reverse decision, a human owns the last step — or you own the liability.
Will AI make or influence decisions about people?
Hiring, lending, eligibility, discipline — decisions about people are a regulated act. Biased history in means biased decisions out; it needs disparate-impact testing and a human check.
Do you classify AI use cases by risk?
Tiered and mapped to a framework (NIST AI RMF, EU AI Act, ISO 42001) beats ad-hoc judgment.
Is there clear accountability for AI outputs?
A named owner and a review path for when a model is wrong.
4
The Capability
What they know — and whether they can sustain it
—
Readiness
Have you completed AI projects before?
A track record changes the risk profile — first-timers need more guardrails and capability-building baked into the first move.
Where will the first use case point — internal or customer-facing?
Internal-first (augment your own people) is where wins compound safely; a raw model aimed at customers is where trust breaks on the edge cases.
Can the team critically evaluate AI output — vs. trust it?
A team that can assess AI can adopt anything. One that can't gets burned by the first confident-wrong answer.
Data & technical literacy to support the work
Can they model their own data and build, or would they depend entirely on you?
Can they sustain what gets built after you leave?
For a fractional role, this is the whole ballgame — build capability, not dependence.
Will the workforce actually adopt it?
Change management and buy-in — distinct from whether the team can build it. Tools get built and never used.
◆
The Use Case
Value vs. feasibility of the first move
How much value would the first use case create?
Business impact if it works — revenue, cost, risk, or speed.
How feasible is it with what they have today?
Data, skills, integration — can they actually ship it soon?
