CASE STUDY / GOVERNMENT & PUBLIC SECTOR
Taking a state permit backlog from 11 weeks to 9 days
Digital permitting with AI-assisted document review on a FedRAMP-aligned platform — 68% faster processing, a cleared 14,000-case backlog, and 100% accessibility conformance.
68%
reduction in median permit processing time (11 weeks to 9 days)
14,000
case backlog cleared in five months
82%
drop in status-inquiry calls on migrated permit types
100%
WCAG 2.2 AA conformance, verified with assistive-technology testing
THE CLIENT
Context
The agency processes roughly 60,000 environmental permit applications a year — water discharge, air quality, land use — through a process built on PDF forms, email, and a 19-year-old case-management system. Median processing time had reached 11 weeks, a 14,000-application backlog had drawn legislative attention, and businesses waiting on permits were escalating through their representatives.
Staff were not the problem: reviewers spent under a third of their time on actual technical review. The rest went to re-keying application data, chasing missing documents, and answering status calls — 41,000 of them the previous year.
THE CHALLENGE
What was at stake
Every efficiency gain had to survive public-sector constraints: statutory process steps that cannot be skipped, Section 508/WCAG accessibility as a legal requirement, records-retention and public-transparency obligations on every document and decision, and a security review aligned to NIST 800-53 before anything touched production. AI could assist, but no determination affecting an applicant could be made by a machine — a line the agency's counsel drew clearly and we engineered to.
The legacy system could not be switched off: thousands of in-flight cases and integrations with state finance systems meant the new platform had to run alongside it, migrating permit types incrementally without ever losing a case file.
Engagement at a glance
- Client
- A US state environmental agency
- Region
- United States
- Duration
- 14 months
- Team
- 12-person team: platform engineers, document AI, service designers, security/ATO lead
Services applied
THE SOLUTION
What we built
We delivered a digital permitting platform in incremental slices, starting with the highest-volume permit type. Applicants got a plain-language, WCAG 2.2 AA online application with save-and-resume, document upload, fee payment, and real-time status tracking — the last of which eliminated most status calls on migrated permit types. Reviewers got a case workbench with complete application context, statutory-deadline tracking, and full audit trails.
Document intelligence removed the drudgery while keeping humans on every decision: AI extraction populates application data from uploaded site plans, lab reports, and legacy paper files (with per-field confidence and human verification on everything consequential), completeness checks catch missing documents at submission rather than three weeks into review, and a similarity engine surfaces comparable past permits with their determinations as reviewer reference — never as automated decisions.
The backlog got its own campaign: the extraction pipeline processed the 14,000 pending paper applications in six weeks, triaging them into review-ready queues that let the agency clear the backlog in five months using existing staff.
// ARCHITECTURE
The platform runs on AWS GovCloud within the state's authorization boundary: containerized services on EKS, PostgreSQL with full audit logging, and infrastructure entirely as code with NIST 800-53 controls mapped and continuously monitored. The public portal is a React application aligned to USWDS patterns, tested against WCAG 2.2 AA with assistive technology every release.
Document AI combines OCR/extraction services with an LLM verification layer for low-confidence fields; every extraction carries provenance to the source page, and all model activity is logged for public-records compliance. Legacy migration ran through a reconciliation pipeline with case-level checksums — zero case files lost across 300,000 migrated records.
Core stack
- AWS GovCloud
- EKS / Kubernetes
- PostgreSQL
- React (USWDS-aligned)
- AWS Textract
- Azure OpenAI (Gov)
- Terraform
- Kafka
- Splunk
HOW IT WAS DELIVERED
Implementation approach
Value delivered in phases with go/no-go evidence at each gate — never a big-bang bet.
- 01
Service research & ATO planning (months 1–3)
Field research with applicants and reviewers; security architecture and control mapping agreed with the state CISO before build.
- 02
First permit type live (months 4–7)
End-to-end digital processing for the highest-volume permit, run in parallel with legacy until reconciliation proved out.
- 03
Backlog campaign (months 6–8)
AI-assisted digitization and triage of 14,000 pending applications into review-ready queues.
- 04
Full migration (months 8–14)
Remaining permit types migrated in six releases; legacy system retired with records archived to state retention standards.
“The legislature asked how we cleared a backlog everyone said needed fifty new hires. The honest answer is we stopped making skilled reviewers do data entry. Ilmora built technology around our statutory process instead of asking us to bend the law around their software.”
Denise Okonkwo
Deputy Director, Permitting & Compliance — US state environmental agency
KEEP READING
More case studies
Want results like these?
Bring us the metric you need to move. Our architects will map how a comparable engagement would work in your environment — systems, timeline, and expected impact.