One practice. Five ways to apply it.

Scope and data-security posture are covered explicitly in the session before any work begins. The short version: the architecture determines what data touches which system, and that is a design decision made with you, not imposed on you.

For retrieval-augmented systems, your documents can be indexed and queried entirely within infrastructure you control -- nothing needs to leave your environment to a third-party service. For inference, the tradeoffs between on-premises models and hosted API calls are real, and the right answer depends on your data classification and regulatory requirements. I have built both. The session's "Guardrails and reality check" segment covers this directly.

The governance approach I apply comes from building Verbatim -- a compliance-grade Medicare/Medicaid billing system where citation provenance and source integrity are non-negotiable, and every answer has to trace back to its source. That is a higher data-integrity bar than most pilots will need, and it shapes how I approach every system by default.

No -- and anyone who quotes you a specific ROI number before seeing your operation and your data is inventing it. The evidence does not support that kind of upfront claim.

What I can say: the value AI delivers in real operations is typically in capability, not in a single measurable line item. Retrieval over your own documents means your team gets accurate answers from your own material instead of searching manually or guessing. Automated document generation means repetitive, high-stakes paperwork is produced correctly and fast. Monitoring agents mean a change in a critical source gets flagged without a person watching the feed. These are real capabilities, and experienced operators recognize their value.

Every engagement is scoped with acceptance criteria agreed upfront. A system either meets them or it does not. The PMP discipline I bring to project management means a pilot has a finish line and an accountable owner -- which is the single most common reason AI pilots stall.

Two things that are hard to find together. First: I have built real AI systems, not slideware -- including Verbatim, a compliance-grade RAG system I built for Medicare/Medicaid billing with citation provenance, sha256-hashed source snapshots, and an SME evaluation loop, where every answer has to trace back to its source. I built it that way because the domain demands traceability. That is engineering I am responsible for getting right, not a demo.

Second: I have 20+ years engineering in real industrial operations -- food, pharma, hospitals, battery manufacturing -- where the details matter and a wrong number costs something. I speak the operational language of the people I work with from the inside, not from a slide deck about their industry.

Most AI consultants have one of those. Almost nobody has both.

No. This is explicit by design. The AI Opportunity Assessment snapshot and roadmap are yours. You can build with me, hand the roadmap to your internal team, or take it to another developer. No proprietary framework, no license, no dependency on me to act on it.

The reason to say this out loud: a buyer who has been burned by vendor lock-in before should not have to guess. The deliverable is a document you own. What you do with it is your decision.

Accuracy is built into the architecture, not bolted on. The approach I use -- citation provenance, source versioning, defined review gates, human checkpoints before important documents -- was developed because Verbatim required it. It is a domain where a wrong answer that cannot be traced to a source is a failure, so every answer has to trace back to its source.

The same approach applies anywhere getting the answer right matters. If you work under a formal quality system, the conversation starts with where AI outputs enter that system and what review and approval those outputs need to carry. The session's readiness scorecard includes a risk and compliance dimension for exactly this. Systems that touch controlled or high-stakes documents are built with traceability first, not retrofitted later.

It depends on the engagement type. An Assessment runs 2-3 weeks: intake, the session itself, scorecard analysis, and the written deliverable. A Build Sprint runs on a scoped project timeline with defined milestones and a clear acceptance test -- you know what done looks like before we start. A Fractional Advisor retainer has a defined monthly cadence: roadmap review, active build oversight, and evaluation runs on live systems.

What does not vary: every engagement starts with a written scope, defined acceptance criteria, and a named finish line. That is the PMP discipline applied to AI adoption, and it is why the engagements that go sideways at other firms do not here.

The work fits any operation that cares about getting answers right and being able to stand behind them -- manufacturing, food processing, pharma, professional services, and plenty of others. Wherever accuracy, traceability, and documentation matter, the governance approach is directly relevant, and the credibility from 20+ years of industrial engineering carries weight.

The method -- scope it, build it traceable, evaluate it, manage it to a finish line -- applies anywhere a pilot needs to reach production and not die in a conference room. If you are unsure whether there is a fit, the free 15-minute fit check is exactly the right place to find out.