Orland Park Pauses AI Voice Agent After Routing Glitch, Expands Gemini Data Project
Orland Park, Ill. is resolving technical issues with its AI phone system after call transfer failures, while approving a $43,200 contract to optimize its Google Gemini data system.
The village of Orland Park, Illinois, is working to correct technical issues with its artificial intelligence phone system after local residents encountered difficulties reaching city staff. Launched in August by Mayor Jim Dodge, the voice agent named Humphrey—after the municipality's first mayor—was designed to maintain high service levels while managing municipal costs. However, following public feedback at an Oct. 5 Village Board meeting, officials temporarily paused the service to address integration failures between the AI system and the village's telephony infrastructure.
During the board meeting, residents Pam Stepansky and Karen Drake detailed their experiences attempting to contact municipal departments. Stepansky reported placing four calls without being connected to a human operator, suggesting the city ensure residents can reach staff directly. Drake described spending nearly two hours trying to get information regarding the Oct. 3 Great Pumpkin Party at Centennial Park, eventually visiting Village Hall in person to get her question answered.
Voice agent stumbles on phone routing
Information Technology Director Tad Spencer addressed the complaints at a subsequent board meeting, acknowledging that internal testing had not uncovered the routing failure prior to rollout. Spencer explained that the problem stemmed from the phone numbers assigned to the system rather than the AI technology itself, which prevented Humphrey from successfully transferring callers to the operator or the Parks and Recreation Department.
Spencer presented performance metrics from September, during which the village received 2,687 calls through the system. Of those calls, 71% were successfully transferred to village departments. Sentiment analysis performed on the call logs indicated that 91% of interactions were perceived as neutral, while 6% were categorized as negative. Spencer noted that while these metrics were acceptable for a malfunctioning automated voice setup, the village aims to reduce negative interactions further by continually analyzing call data. The municipality is currently working with a vendor to provision additional phone numbers so call transfers function properly before resuming Humphrey's operation. Mayor Jim Dodge reinforced that direct feedback helped technicians identify that the root issue resided within the phone network integration rather than the underlying artificial intelligence.
Data storage cost reduction with Google Gemini
While addressing the phone system, Orland Park is expanding its use of AI in other administrative areas. Last month, the village implemented Google's Gemini AI platform for internal data storage and retrieval. Spencer reported that Gemini successfully answered 97% of staff queries during its first month of operation, allowing the village to store legacy data at significantly lower costs. Previously, maintaining an unused legacy system cost the municipality $85,000 annually.
To maintain and refine the platform, the Village Board approved a $43,200 annual contract with Zion Cloud Solutions, a Deerfield-based software company. Beginning in January, the contract provides 40 hours per month of service to monitor and optimize the Gemini environment for village employees.
What it means for developers
The situation in Orland Park offers practical lessons for software engineers building voice agents and enterprise AI applications. First, it highlights how real-world failure points often exist outside the core language model. In this case, the AI model's operational metrics were stable, but a mismatch in underlying telephony routing rendered the system unresponsive for specific user flows. Developers building conversational interfaces must implement robust end-to-end integration tests that cover legacy telecom handoffs alongside prompt execution.
Second, the village's monitoring approach demonstrates the necessity of structured telemetry. By tracking metrics such as transfer rates (71%) and sentiment distributions (91% neutral, 6% negative), IT staff were able to contextualize user complaints against broader system performance. When building multi-model architectures or testing enterprise integrations, developers can try top AI models cheaply through one API at https://apixoai.online to evaluate performance and fallback strategies across different LLM providers.
Finally, Orland Park's dual-track implementation—pausing the public-facing voice bot while successfully scaling internal document search via Gemini—underscores the importance of staged deployment. Internal tools with clear data boundaries often yield immediate efficiency gains, such as replacing an $85,000 legacy system, whereas public-facing interactive voice systems require comprehensive edge-case handling before full release.
Source: Orland Park, Ill., Works to Fix AI-Powered Phone System — GovTech AI. Written by the Apixo team from that report.
One key for Claude, GPT, GLM, DeepSeek and more. Pay per token with crypto.
Get your API key
