A federal civilian agency with 3,200+ annual FOIA requests ran the Aizen Event on their processing workflow. Six of eight steps were automatable. Three were strategic AI investments. Result: 62% cycle time reduction, backlog eliminated in year one.
8 steps, 8 staff, 45-day average cycle. Each request required manual intervention at nearly every stage. A senior FOIA officer estimated 60% of their team's time was spent on work that did not require legal judgment.
In the Aizen Event, we map each workflow step to an AI type: Deterministic (rules with predictable output), RAG (retrieval-grounded), Probabilistic (learned judgment), or Human Required. Classification drives build vs. defer decisions.
| Step | AI type | Recommendation | New state | Time saved | |
|---|---|---|---|---|---|
| 01 | Request receipt & intake Manual logging, case numbering |
Deterministic | Automate completely. Request fields map directly to system records. No judgment required. Parse web form / email → auto-create record → assign case number. | ● Automated | 18 min |
| 02 | Routing to program office Manual reading + email routing |
Deterministic | Build a keyword-to-office routing matrix from historical requests. 94% of requests map to known patterns. Exceptions queue for human review. | ● Automated | 19 min |
| 03 | Responsive records search Manual search across systems |
RAG | High priority. Index all document systems (SharePoint, email archive, records). RAG retrieval against request terms returns ranked candidate documents. Reduces 4.2 hrs to ~35 min for officer review. | ● Augmented | 3.5 hrs |
| 04 | Relevancy review & cull Manual document-by-document review |
Probabilistic | AI pre-screens and scores each retrieved document for relevance. High-confidence relevant and clearly non-responsive are auto-tagged. Officer reviews the middle band. Reduces review time by ~70%. | ● Augmented | 2.0 hrs |
| 05 | Legal counsel review Attorney privilege / exemption review |
Human Required | Do not automate. Legal judgment on privilege and exemption application carries statutory liability. AI can flag likely-sensitive documents to prioritize the attorney's queue but cannot replace the review itself. | ○ Unchanged | — |
| 06 | Redaction preparation Manual PDF redaction with exemption codes |
Probabilistic | Highest-value opportunity. AI detects PII, classification markers, and known exemption patterns. Suggests redactions with exemption codes. Officer approves or overrides. Confidence threshold: 92%+ before auto-apply with audit log. | ● Augmented | 3.0 hrs |
| 07 | Quality & completeness check Manual review of final package |
Deterministic | Automated rules: verify page count vs. log, check all flagged pages have exemption codes, verify redaction coverage, confirm response letter fields populated. Human reviews anomalies only. | ● Automated | 48 min |
| 08 | Package assembly & dispatch Manual compilation and transmittal |
Deterministic | Fully automatable. Auto-generate response letter from template + case data, apply watermarks, compile PDF, transmit via portal, close case in tracking system. | ● Automated | 32 min |
We plot each step by implementation complexity (X) against business impact (Y). Steps in the top-left are Quick Wins: start here. Top-right are Strategic Investments: build after wins are live. Bottom-right: skip entirely.
Deterministic rules. High time savings, very low complexity. Built in week 2.
Fully template-driven. Automated in week 3.
Form-to-record automation. Built alongside routing.
Highest ROI. Probabilistic with confidence threshold + audit log. Phase 2.
Requires corpus indexing. Phase 2 with step 4.
Human-only. Statutory liability boundary. Not touched.
The redesigned workflow after the Aizen Event implementation. Every step with a new state shows the before design, what changed, and who now handles what.
Build cost, deployment timeline, and three-year return: based on 3,200 annual requests, GS-11/12 blended labor rates, and measured time savings post-deployment.
This case study represents a composite of typical federal FOIA workflow engagements. Metrics are based on observed industry benchmarks and comparable implementations. All investment figures are illustrative ranges based on the Otonmi service tiers.
Tell us about the process your team is spending the most time on. We'll classify every step and tell you what's worth building: and what isn't.