ADEME: pre-screening aid applications to free up caseworkers

Like thousands of SMEs, we once applied for a public aid scheme — and missed out. Not because the project was weak, but because an eligibility criterion had escaped us, buried in the scheme's conditions.

What we had experienced as an applicant, ADEME experiences at scale on the reviewing side. A large share of the applications it receives are ineligible or incomplete, and every one of them takes up a caseworker's time. It is the same problem, seen from both sides of the counter.

Every application, even one bound to be rejected, follows the same path. A caseworker opens it, checks the scheme's criteria, verifies the documents, then writes a response. Multiplied by the volume of files, this manual triage consumes precious time. That time is lost to the projects that deserve real scrutiny, and to the applicants who could use support.

The principle fits in one sentence: structure the scheme's criteria, then check each application against facts rather than declarations. The agent verifies eligibility from the company's real data, notably its APE activity code retrieved through the SIRENE registry API. This replaces a declarative questionnaire. It then checks that the file is complete, document by document.

Every application comes out with an explicit status. Every conclusion stays explainable and traceable: the caseworker knows why a file was flagged, and can verify each step of the reasoning. Transparency wins over the black box: a public service must be able to justify every rejection, on the basis of an explainable score.

The agent prepares each file: it pre-screens, and the decision stays in the caseworker's hands. It is a design choice, not a technical limitation. The benefit works on both sides of the counter: caseworkers receive pre-screened files and focus on the applications that deserve their time. Applicants, in turn, can check their eligibility before submitting. "Our goal was to create a screening agent that supports applicants," as Christian Tchouaffé put it in Le7.info.

ADEME's pattern shows up everywhere a back office processes files: insurers and mutual funds handling reimbursement claims, local authorities awarding grants, banks reviewing loan applications. Each time, the same mechanics: known criteria, documents to check. And teams spending their time on files that should have been screened out much earlier.

The approach that worked for ADEME is the one we apply to every project. We start from a concrete problem and build a prototype on your real data, with success criteria defined together. We then take the solution to production, to operate it day to day. The same safeguards apply: explainable conclusions, full traceability, and the decision staying in your teams' hands.

If your teams spend time checking files, sorting requests or verifying documents, AI can often pre-screen that work. The starting point stays the same: a conversation with an engineer to scope your need. First results then arrive on your data, within a few weeks.