AI for grant applications: faster assessment, without AI making the choice
Short answer
A grant round brings many applications in a short time, all to be tested against the same scheme. AI can check applications for completeness, summarise them and set them against the scheme’s criteria. Ranking and awarding stay with the public body: a grant decision is an administrative decision, with reasoning the applicant must be able to follow and challenge.
The problem is the peak
Grants for culture, sport, welfare or sustainability often work with an application window and a ceiling. All applications arrive at once and must be tested against the same criteria within the deadline. Missing documents, budgets that do not add up and project plans of very different lengths slow the work down.
What AI can prepare
| Step | What AI does | What the assessor does |
|---|---|---|
| Completeness | Flag missing documents and fields | Request additions or set the application aside |
| Summary | Aim, activities, target group and budget on one page | Check against the application |
| Criteria | Per criterion, the relevant passages from the application | Assess and score per criterion |
| Budget | Flag arithmetic errors and ineligible items | Decide what is eligible |
| Decision | Draft text based on the approved assessment | Give reasons and decide |
Why ranking is not a task for AI
- It is a decision. Awarding a grant is an administrative decision. An applicant must be able to see why their application scored lower, and object to it. See AI in administrative objections.
- Language is not quality. A language model easily judges writing style instead of the plan. That disadvantages small associations without professional writers.
- No invented facts. A summary may add something that is not in the application. Have every statement point to its source. See AI hallucinations.
- Benefits for individuals are a different matter: there the AI Act’s high-risk regime applies. See AI in social services.
How werqly sets it up
With werqly the assessment runs on a dedicated node. Applications, budgets and contact persons’ data stay in your infrastructure, the scheme’s criteria are available as a source, and the audit log shows per application which passages belonged to which assessment. See AI for local government.
Frequently asked questions
Questions on this topic
May AI assess grant applications?
AI may prepare the assessment: check completeness, summarise and find the relevant passages per criterion. The assessment and the decision stay with the public body, because a grant decision must be reasoned and open to challenge.
Can AI rank applications under a funding ceiling?
Technically yes, but it is unwise. A language model easily rewards writing style over content, and the applicant must be able to follow why their application ended lower. Leave ranking to assessors.
Is AI in grants high-risk under the AI Act?
Grants to organisations usually fall outside Annex III. AI that decides whether individuals are entitled to benefits or essential services falls within it. When in doubt, assess the application and record your reasoning.
How do we avoid disadvantaging small applicants?
Let AI prepare rather than score. Have summaries represent content rather than judge style, and spot-check whether applications from small organisations are summarised as well as others.
Sources
- Dutch General Administrative Law Act, Title 4.2 (grants)
- AI Act — Regulation (EU) 2024/1689, Annex III
This article is general information, not legal advice. Rules and guidance change; check the sources listed or ask your lawyer or DPO if in doubt.