People, Data, Process

Find your champions. Upskill your experts.

Every organization already has people pulling AI toward them: the ones experimenting on their own before anyone asked. Those are your champions, and the first job is to find them in every function. These are not necessarily the most technical people; they are the ones with curiosity and vision who are willing to try, fail, and learn.

The second job is upskilling. Take your best subject-matter experts, the people who know how the business really works, and teach them enough about AI and infrastructure to make a difference. This is not fast, but it is effective. Expect three to nine months; it is well worth the investment.

The reverse mostly doesn’t work. Hire AI specialists, hope they absorb twenty years of domain knowledge, and you wait a long time. A subject-matter expert who understands AI as an acceleration tool can see exactly where it belongs in their domain, because they already know where the friction is.

Governed access, not locked doors

Champions without data produce demos. The second layer is getting real data into the hands of the people who will use it, inside a governance framework instead of around one.

Governance is not red tape. Done right it is the thing that makes access possible at all, and the companies that treat it that way move faster than the ones that treat it as a gate. We set out how we handle that side of it on the security and governance page.

What matters here is speed. Scoped access paths that are permissioned, auditable, and quick to open. When approval drags, two things happen, and both are bad: your evangelists lose heart, and the data you were trying to protect walks into channels you cannot see.

Let the process change

Here is the payoff of the first two layers: upskilled people with real data access start redefining their own workflows. Nobody has to impose a new process from above. The people living inside the process are the ones who rebuild it.

The organization’s job at this layer is mostly to get out of the way, and to be flexible enough to let the process change. If every workflow adjustment needs a steering committee, you will automate the old process instead of discovering the new one. Automating a bad process just produces bad outcomes faster.

A chain, not three workstreams

It is tempting to read People, Data, Process as three parallel efforts to be resourced in some ratio: 60% here, 30% there. That reading produces three half-finished programs.

It is a chain with one driver. Champions get upskilled. Because they are upskilled, they can tell you precisely which data they need. The official sources, meaning the APIs and MCP connectors and the warehouse, and the unofficial ones, meaning the spreadsheets and work products and templates where the operating knowledge really lives. Because they have that data, they redesign the process themselves.

We tested this against the record before we sold it. The evidence behind the framework sets out what supports it, what argues against it, and why the enterprise research does not transfer to a company this size.

Each link is what makes the next one possible. Which is why the company’s job at the third layer is mostly to understand its own processes well enough to see what is changing, and to accept that they will.

Why this framework suits a hold period

A hold period is not long enough for a two-year data platform project to finish before anyone sees value, and it is not short enough to justify buying tools and hoping. PDP fits the constraint: the proof work lands in weeks, the upskilling compounds over months, and the process change arrives when the organization is capable of absorbing it.

It also produces the thing a buyer will pay for at exit. A company where AI capability is resident in the staff and instrumented in the numbers, rather than rented from a vendor whose contract the buyer would inherit.