Most generative-AI projects don't stall on the technology. They stall because the unglamorous work that makes AI usable in production gets skipped. A practitioner's framework built on three things: the use case, the data, and governance.
Generative AI is easy to start and hard to aim. New models and tools arrive every week, each with its own promise, and the pressure to “do something with GenAI” pushes teams to deploy first and ask questions later. That’s exactly how pilots end up stranded: impressive in a demo, impossible to put into production.
The projects that actually reach production share a discipline. Before any model is chosen, three things get dedicated attention: a real use case, trustworthy data, and governance. Skip any one of them and the implementation either fails to deliver value or quietly becomes a liability.
The first step isn’t picking a model, it’s defining the problem. It’s tempting to skip straight to deployment to look current, but an undefined use case is how money gets spent in the wrong place. A well-formed use case does three jobs: it points GenAI at the part of the process where it creates the most value, it lets you choose the right model for the job instead of reworking later, and it sets a measurable goal you can hold the result against and fine-tune over time.
A practical rule: organizations earlier in their data and AI maturity should start with use cases that deliver high value at low complexity. Prove the discipline on something tractable before reaching for something ambitious.
Once a use case is approved, the data underneath it determines everything. GenAI inherits the quality of what you feed it: inaccurate, stale, or poorly formatted data produces confident, wrong output. In a regulated setting, that’s not just a bad deliverable; it’s exposure, if flawed information reaches clients, stakeholders, or regulators. Training and reference data have to be deliberately selected and shaped into the format the tool can use, which varies by tool. This is foundational work, not a preliminary: getting the data right is most of what makes the rest succeed.
From defining the use case through tuning the output, governance is the through-line: the discipline that keeps the business, its people, its clients, and its data safe. It spans legal risk (copyright, discrimination, and adherence to evolving state and federal AI rules), ethical risk (including honest questions about impact on people’s work), and security. For organizations in regulated industries, this is the difference between an AI capability that passes review and one that never leaves the lab. It’s also where two decades of delivering under real compliance pressure shows up: responsible AI isn’t a slogan bolted on at the end, it’s how the implementation is designed from day one.
There’s a quieter point worth making, too: not every problem needs generative AI. We’ve seen GenAI force-fit onto use cases that solid machine learning or straightforward business-intelligence reporting would have solved more cheaply and reliably. Choosing the right tool for the problem, sometimes a smaller one, is part of the discipline.
That’s the work of getting AI from a promising idea to something your business can run on. If you’re weighing where GenAI fits, a short conversation is a good place to start.
case study
case study
event
Tell us what you're trying to do. A senior practitioner will read it.
Talk to us