AI System for Comparing Grant Report Data with Source Records

The problem

When grant reports, activities, budgets and evidence are held in separate sources, their fields must be matched by hand before a list of data differences and evidence can be prepared.

The project manager must compare every suggestion in the list of data differences and evidence with the source record; missing or conflicting evidence must not replace the final decision.

If the same beneficiary's participation in several activities is not reviewed separately, duplicate counting or an unsupported report could result; the record must not progress without a human decision.

What we set out to improve

• Time taken for the draft list of data differences and evidence to reach the project manager with its supporting evidence
• Outcomes of records referred for human review because the same beneficiary participated in several activities
• Number of suggestions corrected after identifying the risk of duplicate counting or an unsupported report

How the system works

1. 01 Retrieve the grant report, activity, budget and evidence from authorised sources
2. 02 Verify field and document versions
3. 03 Prepare the list of data differences and evidence with its supporting evidence
4. 04 Exception check: the same beneficiary participated in several activities
5. 05 Have the project manager accept, correct or reject the suggestion
6. 06 Decision gate: grant report, expenditure eligibility and beneficiary eligibility decisions
7. 07 Record the decision and source separately

Methods we used

• A source-identified data schema for the grant report, activity, budget and evidence
• A structure that separates deterministic rules from AI suggestions
• A screen showing the list of data differences and evidence beside its supporting evidence
• A human review queue when the same beneficiary participated in several activities
• A record identifier that prevents duplicate processing

Where people stay involved

Data and security

Only the grant report, activity, budget and evidence fields required for the task should be processed. Role-based access should be applied, and unnecessary identity, health, financial and internal organisational data must not be sent to the model. Source retrieval, AI suggestions and the project manager's decisions should be recorded separately.

Who this suits

Frequently asked questions