Civic AI - Sentiment Analysis for Constituent
Fabric Score 4.1
PublicPublic ServicesCivicBoth GenAI and Non-GenAI modelsExecutive / Org Leadership ×1Customer / Constituent ×2
Workflow Diagram
Fabric Score
Task
Automatically score constituent messages for sentiment on a 0.00 to 1.00 scale across all incoming communications.
Intent
Legislative offices receive high volumes of inbound messages across many channels. The sentiment score provides a consistent quantitative signal that helps staff and downstream workflows prioritise review, identify highly negative or highly positive messages, and analyse constituent mood over time, while preserving constituent data.
AI Workflow
Input
Incoming constituent messages submitted via email, web form, phone transcription, USPS scanning, or in-person notes captured as text.
Process
The message text is normalised, and individual words are analysed to estimate sentiment polarity and intensity. The model produces a single continuous score between 0.00 and 1.00, where 0.00 indicates the lowest sentiment and 1.00 the highest. The score is generated per message and stored as a message attribute for use in, batching, dashboards, and reporting.
Output
A sentiment score (0.00 to 1.00) attached to each message, available for filtering, batching logic, and analytics.
Human Oversight Level
Human-Led with AI Assistance
Institutional Oversight Examples
- •Models must be fully closed, secured, and incapable of leaking data
- •Sentiment scoring is an assistive signal only
- •Offices retain control over any decisions informed by the score, and access to message content and derived attributes is governed by role-based permissions and internal office policies
Risk
Mis-scoring sentiment, bias across writing styles or demographics, and misinterpretation of sarcasm or mixed sentiment.
Output Modification Telemetry
- Irrelevant information60%
- Missing or incomplete information23%
- Unfaithful outputs (policy, tone, or style misalignment)10%
- Internal inconsistency3%
- Factual errors / hallucinations2%
- Poor usability (unclear language)2%
Transversal Metrics
Grouped by Fabric dimension.
Efficacy
Accuracy88.3 %
Efficiency
Modification Rate16.7 %
Modification Time0.7 min/output
Verification Time1.7 min
Rejection Rate5 %
Operational Friction1 hrs/wk
Implementation Overhead16.7 hrs
Governance Overhead7.3 hrs
Time to Launch3.3 wks
Value
Effort Reduction63.3 %
Risk
Reliability2 incident/month
Autonomy70 %