This is a typical scenario. It shows how Remolda would approach AI support for application intake in a federal program, what we would build and what we would measure.
The situation
Picture a federal program that receives applications on paper and through an online portal. Before a program officer reads a file, intake staff sort documents, check each application against a checklist, key data into a case management system that has run for over a decade, and route the file to a reviewer. Weeks can pass before the substantive review starts.
Applicants cannot see where their file sits, so they call. Staff answering those calls have little to tell them.
Departments in this position have often tried before. Replacing the case management system is slow and risky. A new online form helps, but the data may still be re-keyed by hand. Staff remember those projects, and the union will ask direct questions about the next one.
The approach
The approach adds an AI processing layer in front of the existing case management system. The adjudication workflow and the system itself stay as they are.
Audit (3 weeks). We map the workflow for each main application type, time each step and interview staff at each stage. The baseline is time from receipt to first review, data-entry error rate and the volume of status calls. The audit also reviews data quality, document formats, security requirements and where data may be processed.
Strategy (4 weeks). The plan runs in three waves, starting with the highest-volume stream with the most standard documents. Success measures are agreed with leadership before the build.
The Directive on Automated Decision-Making applies to systems used to make an administrative decision or a related assessment about a client. The design keeps every eligibility decision with a program officer, and the department completes the Algorithmic Impact Assessment for the intake layer before production, with our inputs on data, model and oversight. Union representatives see the plan at this stage, with specifics on which tasks change and what redeployment looks like.
Implement (6 months, 3 waves).
- Wave 1. For the main application type: document classification, a completeness check against the program checklist and a draft deficiency notice for staff to review. For complete files, extracted fields are proposed for the case management system and confirmed by staff where confidence is low.
- Wave 2. The remaining application types, each with its own structure, plus routing to the right stream and reviewer queue based on extracted fields.
- Wave 3. English and French processing to the same standard, and connection to the correspondence system for acknowledgements and status updates to applicants.
Empower (parallel). Training for intake and processing staff: checking extracted data, handling exceptions the system flags, and sending structured feedback that improves accuracy over time.
What we would measure
Each item below is a target we would set with the department and measure against the baseline from the audit. The size of each target is set only after the baseline is measured.
- Time from receipt to first review, by application type and in peak periods.
- Data-entry error rate in the case management system.
- Time to acknowledgement and to deficiency notice for applicants.
- Status calls to the call centre.
- Parity between English and French files on time and accuracy.
Rules that frame this scenario
The Directive on Automated Decision-Making requires an Algorithmic Impact Assessment to be completed, approved and published before an automated decision system goes into production. The Treasury Board Guide on the use of generative AI adds the FASTER principles. The Official Languages Act sets duties for communications with the public in both languages. The federal government page lists sources.
Key lessons
1. Add a layer in front of the old system. Replacing a case management system is a large project with many owners. An intake layer targets the manual work before it and leaves adjudication untouched.
2. Automate the bottleneck, keep the decision. Eligibility review needs policy judgment and stays with program officers. Intake is high-volume and rule-based, and speeding it up lets every later step start sooner.
3. Staff engagement is part of the design. People who lived through failed projects adopt a new one when they know what changes, what stays and what their role becomes. Early union consultation and a written redeployment plan belong in the project plan.
For related work, see AI for federal departments, document processing and department AI training.