The Promise and Pitfalls of AI Voice Agents in E-commerce

AI voice agents are fast becoming the latest experiment in retail automation, promising faster resolution, cheaper operations, and smoother experiences. But as startups like Nector.io have learned, that promise comes with its fair share of noise.

For Ayush Shukla, co-founder/CEO of Nector.io, the goal wasn’t to replace humans. The goal was to test whether AI could handle simple, repetitive queries that often overwhelm support teams. “Suppose you purchase from a brand and want to know the status of your order or make minor changes,” he explained. “You just have to call that number. The AI will pick up, verify, and try to act on it.”

The idea sounds simple enough. Yet, like many early attempts at automating human interactions, reality has been unpredictable. “People just call and randomly talk to make fun of the system,” Shukla said. 

When AI Talks Too Much

The introduction of voice agents has exposed a gap between what LLMs can do in a demo and how they behave when customers are involved. Nector.io discovered that about 60% of calls received by their internal system were off-topic.

That figure mirrors a broader pattern emerging across industries experimenting with AI-powered calling systems. Most struggle with context drift, language nuance, or mischievous users testing the bot’s patience. The solution, Shukla said, lies in refinement rather than replacement.

“We try to keep iterating the prompt every time whenever there’s a bug or stuff like that because for every call there’s a post-call analysis.”

The company uses a combination of Gemini, GPT-4, and Llama models for its AI stack. Some are hosted in-house to lower the latency and cost. Yet, the technical complexity isn’t the most significant issue, user behaviour is. 

To prevent users from intentionally disrupting the AI agent with irrelevant inputs, the company is focused on strengthening the prompts. This “hardening” ensures the agent’s responses remain strictly on-brand and don’t deviate into unrelated topics.

Where AI Meets Loyalty

Nector.io works with several well-known D2C brands in India, including Boat, Minimalist, and Urban Monkey, to power loyalty and retention programmes. Boat offers around five percent cashback on purchases, while Minimalist rewards customers with certain cashback for every transaction. Urban Monkey operates a membership system that offers customers exclusive benefits comparable to Amazon Prime.

These brands, Shukla said, use Nector.io to manage everything from loyalty points to referrals and reviews. “If you buy from a D2C brand, review plays a very important role,” he explained. “You can launch a membership programme, a referral programme, or a review system using our platform.”

It’s this same network of client brands that will soon test Nector.io’s new AI-powered voice agents, which aim to automate customer support without losing the human touch. But the rollout has been intentionally cautious. 

He explained the reason behind that is the potential for misuse, which could significantly inflate a brand’s billing. Users might intentionally prolong calls with unnecessary chatter for several minutes, leading the brand to be charged. 

If this increased cost doesn’t translate into proportional value creation for the brand, and they don’t see an initial positive impact, it could lead to customer or client churn.

“To mitigate this, we are working on optimisation strategies to prevent such cases. The company plans to build ‘firewalls’—which essentially means implementing various preventative mechanisms.” 

While some misuse is perhaps inevitable, their primary goal is to ensure that value creation is evident even in the early stages of adoption.

Between Potential and Practicality

For now, Nector.io’s AI voice agent remains to be rolled out gradually. The company’s early experiments underline both the efficiency AI can bring to e-commerce support and the fragility of deploying it in the wild.

In a sense, the challenges aren’t technical at all. They are social, behavioural, and ethical — how to build systems that can converse without being exploited, assist without offending, and operate without spiralling costs.

“The goal,” Shukla said, “is to make sure there’s value creation in the initial days.” It is a measured optimism, one that reflects a new generation of startups learning that automation in customer support is not about making AI talk, it’s about teaching it when to stop.

The post The Promise and Pitfalls of AI Voice Agents in E-commerce appeared first on Analytics India Magazine.

Generated by RSStT. The copyright belongs to the original author.

Source