The Challenge
The City of Seattle's Office of Police Accountability (OPA) investigates complaints and misconduct claims against Seattle Police Department employees, publishing hundreds of case reports each year as part of their commitment to transparency. However, finding specific information within these reports proved challenging for community members seeking accountability data.
The Office faced a complex challenge: With 300-350 case reports published annually and records dating back to 2015, their existing search system only allowed searches by case number and basic metadata fields. Previous attempts to manually add detailed metadata proved too complex to maintain. As a result, the Office received approximately 100 public records requests per year for information that already existed in their published PDFs but was difficult to find.
Staff spent significant time helping community members locate specific cases, creating an unsustainable burden on resources while limiting public access to important accountability information.
The Need
The OPA needed a sustainable way to make their case reports more searchable without disrupting their existing publication workflow. Community advocacy groups had been requesting better access to police accountability information, and both the Mayor and City Council showed strong interest in leveraging AI to improve government services.
The initiative aligned with the City's January 2025 deadline to define an AI project pipeline, presenting an opportunity to develop frameworks for AI viability assessment that could be applied across city departments.
Why USDR?
The City of Seattle had successfully partnered with USDR on multiple previous projects. When looking for expertise in evaluating and implementing digital tools for government services, they turned to USDR. The OPA's open dataset presented an ideal pilot project to demonstrate how AI could make government information more accessible.
"USDR kept solutions and conversations realistic within that realm. They really heard us and understood where we were, this is what we need, this how much energy we can put to this with other staffing needs. Which was really helpful." - Katie Maier, assistant director of operations, Office of Police Accountability
Our Approach
USDR assembled a team of skilled volunteers to work alongside the Seattle team over 12 weeks, ultimately extending the engagement by a month to ensure thorough delivery.
The team took a business-first approach, starting by understanding OPA's constraints and workflows before proposing solutions.
The approach evolved through discovery:
1. Tool evaluation phase The team evaluated existing tools in Seattle's toolkit, including Adobe Acrobat AI Assistant and Microsoft Power Automate, to work within budget constraints. They developed an evaluation framework distinguishing when AI solutions versus automation would be most appropriate.
2. Custom solution development After determining that available AI tools weren't suitable for the specific need, the team developed a custom application that could scan documents and extract keywords, creating a searchable database of case information.
3. Process improvement The team also addressed workflow pain points, developing a spreadsheet-based bulk upload process that dramatically reduced the time staff spent uploading documents.
Technologies used
- R programming for custom application development
- Adobe AI Assistant evaluation
- Natural language processing
- Prompt engineering
Practices Used
- Service design
- Process improvement
- Workflow documentation
- AI evaluation framework development
- Training and knowledge transfer
The impact
The partnership delivered both immediate improvements and sustainable processes for the future:
- Developed an automated solution that enables keyword searches across all case summaries, replacing a manual process that previously required knowing specific case numbers
- Created significant cost savings compared to a previous solution explored that would have cost approximately $200,000, making this project feasible within current budget constraints
- Improved internal efficiency by creating a spreadsheet upload process that reduced document upload time from 5-10 minutes per file to bulk uploads, addressing staff pain points
- Established an evaluation framework for AI vs. automation decisions that can guide future city technology projects
- Built internal capacity to search past cases for precedent and continuity, an unexpected benefit that helps case writers reference similar situations
While not fully deployed to the internal team quite yet, the solution will particularly benefit OPA's civilian staff who author case summaries. The staff will be able to search for precedent and maintain consistency across cases.
"What I appreciated is that [USDR] started from the business problem. They started from trying to understand our constraints and the way that we do our work... Then built upon that understanding to come up with a solution, as opposed to trying to fit a solution into a problem." - Mark Schmidt, program manager, Seattle IT
The team
USDR volunteers:
- Annie Rajam, project lead
- Di Dang, AI service designer
- Bhavika Tekwani, AI engineer
City of Seattle:
- Amanda DeFisher, OPA project manager
- Katie Maier, OPA assistant director of operations
- Eric Seely, OPA data analyst
- Mark Schmidt, program manager
- Ginger Armbruster, chief privacy officer
- Sarah Carrier, privacy and responsible AI program manager
- Claire Palay, open data analyst
- Chris Smith, electronic signature administrator
Thumbnail photo by Vitaly Gariev on Unsplash
