Voice AI agent in after-sales support: why it gets stuck when the customer needs to show

Voice AI agents have made enormous progress. They understand freely worded requests, answer without delay at any time, and know how to hand over to an adviser. On simple requests, they deliver on their promises.

In after-sales support, it is a different story. As soon as the customer has to describe what is in front of them, the voice agent starts guessing. And an agent that guesses means a customer who calls back.

What a voice AI agent does very well

• Requests that fit into one sentence: “I want to reschedule my appointment”, “Where is my order?”, “What are your opening hours?”.
• Information the customer knows by heart: their name, address, contract number.
• Triage: understanding the reason for the call and directing it to the right department.
• Overflow and off-peak hours: answering when all advisers are busy, at night, at weekends.

Where it gets stuck: everything you can see

In after-sales support, the answer often depends on what is in front of the customer. And they do not have the words to describe it.

• “There is a flashing light”: which one? What colour? At what rhythm? On an internet router, boiler or washing machine, that is the whole difference between “restart it” and “we will send a technician”.
• “It is the white model, I think”: the customer does not know the reference or serial number, printed in small type on the back of the device.
• “There is a code on the display”: E1, F28, 04… misread half the time.
• “That broken part, there”: impossible to name if you are not in the trade.

Faced with these vague descriptions, the agent has two bad options: ask ten questions and tire the customer, or choose a plausible answer and risk being wrong. Large language models have a known tendency: instead of saying “I do not know”, they produce an answer that sounds right. In after-sales support, that means an unnecessary visit, the wrong part ordered, a customer calling back even more dissatisfied.

At Vizalia, we observe that a voice-only agent resolves an average of 15% of calls of this type. Not because it is unintelligent: because it is blind.

The answer: give the voice agent eyes

The idea is simple: instead of asking the customer to describe, ask them to show.

1. During the call, the agent sends an SMS link. The customer opens it in their phone's browser, with no app or account. The conversation continues.
2. The customer films what is in front of them: the light, label, screen, part.
3. The agent sees and answers based on what it sees: it reads the serial number, recognises the product, identifies the light, then guides the customer step by step.
4. When needed, an adviser takes over, with the image and history: the customer does not have to repeat anything.

With sight, resolution rises from 15% to 70% according to our measurements, and the agent no longer needs to invent: if it cannot see clearly enough, it says so and hands over.

Two ways to do it

• You already have a voice agent (Twilio, Vapi, Retell or built in-house): it connects to Vizalia through our MCP server. It keeps its voice and gains eyes.
• You do not have one yet: Vizalia provides the voice agent and visual agent, ready to answer on your overflow and off-peak hours.

In both cases, your existing training materials form the visual agent's knowledge base. Discover the voice AI agent with vision

What about data security?

An agent that sees inside a home or a customer's screen must offer strong guarantees. Vizalia can run on its servers in France or directly on your premises (on-premise), with open-source models. No customer data trains the models and sessions are ephemeral by default. Vizalia is ISO 27001 certified. Our security guarantees

Frequently asked questions

Can a voice AI agent replace after-sales advisers? No. It handles simple requests and some common faults, especially when it can see. Complex cases return to an adviser, with the image and call history.

What is the difference between a callbot and a voice AI agent? A callbot mainly follows predefined scenarios; a voice AI agent understands open requests and leads the conversation. Neither sees unless given the means to do so.

Does the customer have to install an app? No. They receive an SMS link and film from their phone's browser.

Can we keep our current voice agent? Yes. It connects to Vizalia through our MCP server, without changing provider.