Automating Identity Verification to Reduce Support Handle Times

A human support agent spends the first 90 seconds of every call or chat asking for a name and an order ID. This repetitive data collection drains payroll. It also irritates customers who expect the business to recognize them from a logged in session or their phone number. Most support centers lose 20 percent of their productive time to these basic identity checks. An agent greets the caller, waits for them to find an invoice, and manually types characters into a CRM. This repetitive labor is one reason why manual data lookups stall support operations. If a customer moves from a website chat to a phone call, they often start this process over because the data does not follow them.

Duvi agents handle these initial steps before a human enters the conversation. When a business connects a website via a script tag or a phone number through Twilio, the agent uses the same logic for every channel. The agent asks for the account number and calls a business API or a Salesforce endpoint to check the status. This happens in the background. If the customer provides a valid ID, the agent proceeds to solve the problem using knowledge crawled from the business website and uploaded files. The agent parses this material into a vector store to ensure answers come from real company documentation. Relying on documentation alone can be a read only AI trap where staff still perform the actual paperwork. To avoid this, Duvi combines grounding with active tool calls. If the situation requires a human, the agent hands over the conversation. The human receiver sees the full transcript of what the AI already verified.

This approach addresses the problem of context loss. Because Duvi uses the same knowledge base for WhatsApp and phone calls, the verification logic remains identical across all surfaces. The business does not need to build separate workflows for text and voice. A reply always goes back on the channel the customer used. If a customer starts a verification on the phone but needs to upload a photo of an ID, the agent can move the task to an asynchronous WhatsApp thread. This ensures static knowledge bases do not create support bottlenecks when customers require transactional updates.

Businesses manage these agents without engineering work. Launching an agent involves pointing the crawler at a URL or uploading specific PDFs. Once the knowledge is ingested, the agent can be assigned tools like HTTP APIs or MCP servers to look up live customer data. This capability is fundamental for connecting support agents to live business data. This allows the agent to act on the information it gathers.

When an agent finishes a task or hits a limit it cannot resolve, it hands the interaction to a human. The human receives the data the agent already gathered from the backend. This prevents the customer from repeating their details and helps resolve the action gap crisis where teams are buried in non work tasks. Completed conversations are analyzed into analytics to show where users struggle with specific verification questions. Usage runs on a credit system, allowing companies to scale automated support capacity without increasing permanent headcount. Moving the verification loop to an AI agent changes the role of the human staff. Instead of data entry, agents focus on complex problem solving. The technical configuration requires no code, and the logic stays consistent whether the customer is typing on a website or speaking into a phone.