Skip to content

News

From proof of concept to production: the distance that decides whether AI counts

November 4, 2026 · PenDraco

78 percent of organizations used AI in 2024. Over two-thirds of leaders expect 30 percent or fewer of their experiments to reach full scale within six months. One underwriting engine crossed that gap inside a product used by 22 of the top 25 P&C carriers. What it took.

Two numbers describe enterprise AI better than any product announcement. Stanford's 2025 AI Index reports that 78 percent of surveyed organizations used AI in 2024, up from 55 percent in 2023. Deloitte asked 2,773 leaders at organizations already piloting generative AI about their current experiments. Over two-thirds said 30 percent or fewer would be fully scaled in the next three to six months. Adoption is nearly universal. Production is the exception.

The distance between those numbers is not a technology gap. It is the distance between a demo that works and software that people depend on. This article is about one engagement that crossed it, and about what a crossing takes.

The demo is not the hard part

A proof of concept answers one question: can this be done? It runs on a sample of the data, in an environment the builder controls, in front of people who want it to work. Most pilots answer that question well. Deloitte's respondents are not reporting failed demos. They are reporting experiments that have not yet become production.

Production asks different questions. Does it run on all of the data, including the records nobody cleaned? Does it live inside the tool people already use, or beside it? What happens on the cases it gets wrong, and who sees them? Who owns it after the builder leaves? A pilot nobody designed to answer those questions cannot grow into answering them. Someone rebuilds it, and the rebuild is where the budget and the sponsor usually run out.

The engagement

Insurity builds cloud software for property-and-casualty insurers. Its own June 2024 release counts 22 of the top 25 P&C carriers and 7 of the top 10 MGAs in the United States among its customers, with over 400 cloud-based deployments. The ask was specific: underwriters needed the next best action on each account, with an outlier score, inside the product they already worked in. Not a dashboard beside it. Not a report sent later.

PenDraco's founder, Steve Onoja, architected that engine and the embedded analytics portal that carried it. He did the work during his decade at Qlik, embedded with the customer, and the credit for the architecture is his. The proof of concept worked in the demo, which is where most engagements end. This one went the whole way. The engine shipped inside Insurity's own product, and underwriters used it in production.

The third-party marker is public. Qlik named Insurity its North America OEM Partner of the Year for 2023. Insurity's release describes the awards as honoring partners who excel in delivering innovative and effective data solutions, and credits the embedded analytics that turn raw data into insight inside its tools. That embedded analytics layer is the portal PenDraco's founder architected.

What a crossing takes

PenDraco's embedded engineering line is built on five practices. None of them is a model.

Real data from the first week. The system meets the full population early, so the ugly records shape the design instead of breaking it at launch.

The product as the surface. The work ships inside the customer's own interface, with its permissions and its workflow. A user never has to go somewhere else to use it.

Failure cases as deliverables. A score that is wrong in a way nobody can see is worse than no score. The cases the system cannot call with confidence get routed to a person, not hidden.

Users inside the loop. The people who will depend on it see it before the finish and change it. The finish line is their adoption, not the demo.

Ownership on the customer's side. The engagement ends with the customer running the system in its own environment. Nothing stays behind as a dependency on the builder.

What this means for a leader

Fund the crossing, not the demo. A proof of concept is cheap and proves little that the Deloitte numbers do not already show. The expensive, valuable part is the distance from the demo to the day users stop noticing the thing, because it is just how the work happens now.

That distance is what PenDraco's embedded engineering line exists to cross. A senior engineer joins your team for a fixed 30 to 60 day scope, works inside your environment on your data, and ships into your production. You keep everything built. If you have a pilot that worked and has not become anything, get in touch. That is the engagement we are built for.

Sources

← Back to news