All posts

Reducing Administrative Overhead by Mapping Voice and Message Intents to Internal APIs

Connect AI agents to internal APIs and Salesforce to reduce administrative overhead. Automate booking changes and order checks across voice and messaging.

Businesses often lose support capacity to repetitive administrative tasks. Customers call to change a booking, verify an order status, update a shipping address, or handle a return. When a staff member spends minutes on these calls, the business pays for high cost data entry. The problem is a disconnect between where information exists and where actions happen. A website FAQ explaining a return policy cannot process the return. This limitation forces the customer into a phone queue.

Duvi resolves this by allowing an AI agent to use your existing HTTP APIs or Salesforce records as tools. By mapping tool calls to specific database records, these interactions move from static answers to active resolutions. The setup starts with your own material. Duvi crawls your website and parses uploaded files into a vector store. This ensures the agent uses your actual product data instead of relying on a general model. When a customer uses the chat widget on your site, the agent references this knowledge to provide an answer. If the same customer calls your phone number later, the agent recognizes the context. It uses the same knowledge base and the same system prompt across every channel.

Action happens through tool calling. You connect your own phone number or WhatsApp business account through providers like Twilio or Meta. You then define the HTTP APIs the agent is permitted to use. If a caller asks to reschedule a service appointment, the agent queries your scheduling API for available times. It presents those times to the caller and updates the record once a selection is made. You can enable direct API execution in voice conversations to remove the delay between a customer request and a database update.

This move beyond simple text responses is possible because Duvi supports live voice on phone calls and WhatsApp. The agent can handle these transactions while the customer is driving or working. For businesses managing complex stock, bypassing contact forms by connecting inventory APIs directly to automated conversations provides customers with live pricing data. The logic remains consistent whether the interaction is a voice call or a message.

Human handover is a safety net for complex scenarios. If the agent encounters a request it cannot resolve or detects frustration, it transfers the interaction to a person. The human receiver gets the full transcript of the conversation. They do not have to ask the customer to repeat an order number or serial code. This functionality is useful for capturing service leads after hours when your primary staff is offline. It also helps prevent the voicemail abandonment leak that occurs when customers are left with passive recording systems.

The goal is to stop using humans as routers for your database. You define the rules in a system prompt and give the agent the tools to execute them. Launching does not require significant engineering work beyond the initial API connection. Usage runs on credits. This makes it easier to scale during busy periods without hiring temporary staff.