1 of 16

AI for Officer Efficiency

Turning a workflow pain point into an AI opportunity

Renee Romero • AI Systems Designer • 2025

A case study in specification precision, evaluation design, trust & security, and context architecture.

2 of 16

Project overview

My role:

AI Systems Designer, Renee Romero

The product:

A letter editor used by government officers to create and send official correspondence to applicants and petitioners.

Responsibilities:

Evaluation design, user flow analysis, stakeholder synthesis, cost & token economics, specification precision for AI concept, roadmap influence

Project duration:

8 weeks (September – October 2025)

3 of 16

Project overview

The problem:

Officers were spending more time fixing formatting than writing official letters. During discovery, I noticed officers constantly leaving the system to copy legal citations from external sites. When pasted back, formatting broke—costing 10-15 minutes per letter.

The goal:

Identify the root cause of this inefficiency and define precise specifications for a solution that could be added to the product roadmap. Quantify the cost impact to build a business case for AI-powered assistance within a trust-critical government letter editor.

4 of 16

Understanding

the system

  • Evaluation design
  • User flow analysis
  • Pain point discovery

5 of 16

Evaluation design

Conducted a systematic evaluation across 3 key pages using Nielsen's 10 Usability Heuristics to identify failure patterns and design gaps

Dashboard Page

  • Visibility of system status
  • Match with real world
  • Consistency and standards
  • Aesthetic design
  • Help and documentation

Draft Page

  • User control and freedom
  • Error prevention
  • Recognition over recall
  • Flexibility of use
  • Help users recover

Letter Page

  • Consistency and standards
  • Error prevention
  • Recognition over recall
  • Aesthetic design
  • Help and documentation

6 of 16

The insight

A two-fold pain point revealed a deeper design gap.

1. Context switching

Officers leave the letter editor to find legal citations from external sites, interrupting workflow.

2. Manual reformatting

Pasted citations break formatting, requiring manual fixes—wasting time and creating inconsistency.

The context architecture gap

Officers lacked the right context surfaced at the right time to find and apply citations efficiently within the editor.

This exposed a decision-support gap—the system had no context architecture to help officers complete their core task without leaving the application.

7 of 16

Quantifying

the impact

  • Time-on-task data
  • Cost analysis
  • ROI projection

8 of 16

Cost analysis

10 min

Time Saved

×

$35/hr

Avg Rate

×

91,256

Letters

=

$532K+

Annual Savings

10 minutes

Avg time officers spend per letter leaving the system to find citations and manually fixing formatting after pasting.

$35 per hour

Avg hourly rate across GS-5 to GS-13 pay grades, calculated from OPM federal salary tables.

91,256 letters

Total correspondence volume from 2023 fiscal year data provided by stakeholders.

Note: With 15 min savings at higher GS levels, annual impact reaches $1.5M+. This cost modeling approach reflects cost & token economics thinking--quantifying ROI before building.

9 of 16

The

opportunity

  • Current state
  • AI specification design
  • Stakeholder alignment

10 of 16

Current state: The letter editor

Officers use this interface to compose official correspondence

Current pain points:

  • No inline citation lookup
  • Officers leave system to find references
  • Copy-paste breaks formatting
  • Manual reformatting required
  • 10-15 min lost per letter

What if the system surfaced the right citations at the right time?

11 of 16

Proposed solution: AI citation assistant

A precisely specified inline assistant that surfaces relevant legal citations within the editor

Suggested citations for this letter:

8 CFR § 214.2(h)(4)(ii)

INA § 101(a)(15)(H)

Click to insert with formatting

Key benefits:

  • Eliminates context switching through better context architecture
  • Maintains consistent formatting
  • Ensures citation accuracy within trust-critical government workflows
  • Reduces time-on-task by 10-15 min
  • Improves officer focus and flow

"AI isn't magic. It's specification precision that anticipates what users need next."

12 of 16

Stakeholder alignment

From pain point to roadmap—building the case for AI exploration

1. Discovery

Synthesized insights from officer interviews and stakeholder research

2. Evaluation design

Connected pain points to measurable time-on-task inefficiencies through systematic evaluation

3. Cost & token economics

Translated time savings into dollar impact to justify building before committing resources

4. Roadmap

Influenced PM to add AI exploration to the product roadmap

Key outcome

AI-powered citation assistance was added to the product roadmap for future exploration, with analytics tracking set up via Matomo for ongoing impact measurement.

Roadmap Adopted

13 of 16

Going

forward

  • Impact
  • Takeaways

14 of 16

Impact

$1M+

Potential annual savings

10-15 min

Saved per letter

91K+

Letters impacted yearly

Key achievements

  • Identified a context architecture gap costing significant time and money
  • Applied evaluation design to translate qualitative pain points into quantitative impact
  • Defined precise specifications for AI-powered assistance in a trust-critical system
  • Secured roadmap adoption for AI exploration
  • Set up analytics tracking for ongoing evaluation and impact measurement

15 of 16

Takeaways

AI is a specification precision problem, not just a tech solution

The opportunity was not about adding AI--it was about defining precisely what officers needed and removing friction. AI happened to be the right tool.

Evaluation design speaks louder than opinions

Quantifying the cost impact through systematic evaluation transformed a "nice to have" into a business priority.

Small observations reveal big context architecture gaps

Noticing officers leaving the system to copy citations revealed a $1M+ context architecture problem--the system was not surfacing the right information at the right time.

"AI isn't magic. It's specification precision that anticipates what users need next."

— Project reflection

16 of 16

Let's connect!

If you're interested in further discussions or collaboration, I'm Renee Romero, AI Systems Designer, and I warmly welcome the opportunity to connect. Thank you for exploring this case study!

Email: reneeromero326@gmail.com

LinkedIn: linkedin.com/in/renee-romero

Portfolio: muralderomero.com