From the Local Government edition of September 17, 2026
Riverside Rivy interview reports 98.7% chat resource-match and 9% 311 voice adoption
City of Riverside · Municipal 311 and resident information services · Riverside, California, United States
- Publisher
- Partnership for Public Service AI Center for Government
- Original publication
- August 13, 2026 AI Center for Government interview
- Source retrieved
- 2026-09-18
What happened
Riverside’s Rivy chat and 311 voice assistants, trained only on public website data after citywide employee testing, report 98.7% chat resource-match accuracy, 9% voice adoption when offered, and 89.9% first-call resolution among voice users, with live-agent transfer available.
Why it matters
Complements Riverside’s September council AI working-group coverage with operational chatbot/voice metrics and public-data-only training—useful for cities balancing launch claims against low voice uptake and escalation design.
Evidence and measured results
Deputy CIO describes >100,000 annual 311 calls with nearly half informational; internal pilot ~2,700 employees; guardrails redirect homework queries to library services; ongoing intent/response monitoring. Metrics are self-reported in an association ‘bright spot’ interview.
Limitations and uncertainty
Association bright-spot framing; metrics lack independent verification; 9% voice adoption limits claimed 311 impact.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-09-18; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Large volumes of routine 311/info calls leave residents waiting while answers sit scattered on city websites.
- Stakeholders
- Chief innovation / IT
- 311 operations
- Department content owners
- Mayor/council
- Residents
- Discovery questions
- How is 98.7% resource-match defined and sampled?
- Why is voice adoption only 9%?
- What is after-hours escalation policy?
- How is public-data-only training enforced with vendors?
- Value hypothesis
- Public-data chat/voice agents may improve informational routing if accuracy methods are published and live help remains available.
- Potential engagement
- Soft-launch design with department testing and dual chat/voice KPIs before public expansion.
- Unsupported claims
- 98.7% accuracy in other cities without testing
- Voice AI will absorb most 311 volume at 9% adoption
- Public-data training eliminates all privacy risk
Pre-sales engineering
Role takeaway
- Fit
- Public-content conversational assistants for web chat and 311 voice with human transfer.
- Architecture and integration
- Shared curated corpus tied to city website; IVR option into same guardrailed model; no private-case retrieval.
- Prerequisites
- Website content governance
- Employee soft-launch
- Intent/quality monitoring
- Live-agent handoff
- Constraints
- Low voice opt-in
- Business-hours human coverage
- Association-reported metrics only
- Security
- Block private data training; control logs; harden against off-topic/abuse prompts.
- Proof of value
- Published accuracy protocol
- Adoption and escalation rates
- Independent sample retest
Delivery
Role takeaway
- Work
- Run Rivy with ongoing content updates, intent review and 311 transfer quality checks.
- Dependencies
- CMS owners
- 311 supervisors
- Telephony platform
- Communications
- Ownership
- Innovation/IT owns product; 311 owns voice operations; council sets accessibility expectations.
- Skills and adoption
- Train departments for soft-launch testing; coach agents on AI-to-human handoffs.
- Governance checkpoints
- Before adding non-public data
- After major website restructures
- Quarterly independent accuracy sample
- Acceptance criteria
- Live transfer works in tested scenarios
- Public-data constraint verified
- KPIs include adoption not only accuracy
- Risks
- Bright-spot bias
- Low voice uptake undercutting ROI stories
- Residual wrong answers on high-stakes topics
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Shared knowledge base across web chat and 311 voice; constrain training to public site content; preserve live transfer during business hours.
Governance
Who approves, reviews and stays accountable for outcomes?
Pair high accuracy claims with published error definitions and independent retests; council strategy work should incorporate these KPIs and escalation SLAs.
Security and privacy
What data, permissions and controls need testing?
Public-data-only training reduces sensitive-data exposure but does not eliminate logging/privacy issues; homework/off-topic guardrails show need for abuse controls.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Voice and chat extend access but low voice adoption and business-hours live agents mean phone equity and disability accommodations still depend on human 311 capacity.
Procurement
What should contracts, pricing and exit terms secure?
If platform components are vendor-hosted, contracts must enforce public-data-only training and handoff SLAs; association case studies are not RFPs.
Operating model
Which teams own the service once it runs?
Employee soft-launch and continuous monitoring are part of the operating model, not one-time go-live tasks.
Publication history
- 2026-09-17Local Government · Issue 124 resources
Stable resource ID: riverside-rivy-chat-voice-metrics-202608