Aligning service eligibility checks across phone and web using real time API integration

A customer finds a specific service on a website. They call the listed number to book it. The person answering the phone often lacks the context of the page the customer is viewing. This disconnect forces the customer to repeat their requirements. It leads to a support representative providing information that contradicts the latest website documentation. You can fix this by solving knowledge divergence between phone and web support.

Providing a consistent answer requires the agent to have immediate access to the same technical data and business rules across every channel. Duvi addresses this by anchoring the agent to a central knowledge base. When a business crawls its website or uploads technical PDFs into Duvi, that data becomes the source of truth. An agent on a phone call uses the exact same vector store as the one on a WhatsApp thread or the website widget. This maintains information consistency between website and phone channels.

Consistency involves more than static knowledge. Many businesses struggle with eligibility checks that depend on dynamic data like service coverage by zip code or current inventory levels. If a customer calls to ask if a technician can visit their address, the agent should not have to put them on hold to check a separate internal database. We have seen how manual data lookups stall support operations.

Duvi agents call HTTP APIs or MCP servers to pull this information during the call. In the agent builder, a business defines a tool that points to their internal eligibility API. When a customer speaks a zip code, the agent triggers the API call. It then parses the response to provide a definitive answer. This happens during the live conversation without any manual data entry from a human operator. The website chat widget uses these same tools to provide the same answers.

Setting up this automation does not require a dedicated engineering team. The process starts by describing the job of the agent in the Duvi builder. For a service eligibility agent, the system prompt is structured in six blocks to define the specific rules of the conversation. One block might instruct the agent to always verify the location before discussing pricing. Another block ensures the agent only uses facts found in the indexed knowledge. Every rule in the prompt is given a specific reason to help the model follow instructions accurately. This helps bridge the action gap crisis where teams are overwhelmed by repetitive manual tasks.

The latency of these interactions is a critical factor for voice calls. High delays between a customer asking a question and the agent responding lead to abandoned calls. Speed matters for voice. Within the Duvi dashboard, businesses choose specific models and transcribers to reach a response speed that feels natural. The interface provides a combined latency estimate. This allows the business to see the impact of their choices before the agent is deployed. This is a central part of managing lead qualification latency across different channels.

For businesses using Twilio or Meta, the integration is handled by connecting the existing account to Duvi. This allows the business to use their own phone numbers and WhatsApp Business profiles. When a customer reaches out via WhatsApp, the agent handles the messages using the same prompt and tools it uses for live calls. A reply always goes back on the channel the customer used. If the conversation becomes too complex for the AI, the system can hand the session over to a human. This approach helps with closing sales in the chat by moving beyond simple static responses.

Performance is tracked through analytics that breakdown completed conversations into insights. Instead of listening to hours of call recordings, the business can review data to see which APIs are being called or where customers are dropping off. This feedback loop allows for precise tuning of the system prompt to improve future interactions. Launching an agent requires no code, and the entire system runs on a credit based usage model.