Federal Agencies Have Had an Evidence Act Mandate Since 2019, and AI Startups Like Saipion Now Pitch Tools to Meet It at the Design Stage

Since the Foundations for Evidence-Based Policymaking Act became law in January 2019, federal agencies have been required to plan how they’ll use evidence. They write learning agendas, publish evaluation plans and name chief evaluation officers and chief data officers. The law built on the Government Performance and Results Act, which already tied agency plans to measurable goals.

What the law didn’t solve is timing. Most program evaluations land after a program has run, often years later. By then the money is spent and the design is set.

A Washington startup called Saipion is betting that gap is now a software market. On November 10 it launched Civryn, an AI platform aimed at governments, development institutions, foundations and nonprofits, built to bring research and community input into a program before money is committed.

What Civryn says it does

Civryn’s first module, Program Design, walks a team from defining a problem to a full program design or funding proposal. It pulls in existing research and evaluations, adds structured input from experts and from the communities a program is meant to serve, and uses AI to tie it together.

Saipion says early testing points to 20% to 40% less staff time on design work, depending on the program. That’s the company’s own estimate from early use, not an independent result.

Two more modules are in development: Predictive Governance and Evaluation and Learning. Together they’d stretch the tool across the program lifecycle, from design to delivery to review. Saipion is also hiring engineers to build on FedRAMP-aware cloud infrastructure, a sign it’s aiming at federal buyers, since cloud services sold to agencies need FedRAMP authorization.

The problem it’s aimed at

Saipion’s Rolf Sartorius has described most program evaluations as autopsies: they explain why a program failed after it’s too late to fix. His argument is that evidence has to move upstream, into design and into the middle of delivery, and that the Evidence Act and GPRA laid the groundwork without changing how most programs are built.

That matches a long-running complaint in Washington. Agencies have built the roles and plans the Evidence Act calls for. Using findings to change programs is the harder part.

The questions buyers will ask

An AI tool that drafts program designs runs into the same issues as any AI used in government. Agencies will want to know which sources it draws on, whether it can show its reasoning, how it handles sensitive program data, and who’s accountable when its recommendations are wrong. OMB guidance already requires agencies to inventory their AI uses and apply minimum risk practices to high-impact ones.

There’s also a simpler test. A design tool earns its keep only if programs built with it perform better than programs built without it. That takes years to show, which is the same problem Civryn says it’s trying to solve.