For years, the problem with telemarketing was not whether the caller was human. It was whether the consumer wanted to hear from them. AI voice is now adding another layer to that problem. As brands deploy automated voice agents for sales, collections, reminders and customer service, the cost and effort of making large volumes of calls have fallen. But the consumer’s tolerance has not necessarily followed. The timing is significant. TRAI has now amended its commercial communications regulations to bring application-to-person (A2P) calls, including auto-dialled, robocalls and pre-recorded or artificial voice calls, within its regulatory framework. The amendments also provide for termination charges of up to five paisa on undeclared A2P calls originating from apps or software systems.
The challenge for marketers, therefore, is no longer simply making an AI voice agent sound convincing. It is making sure that the call is answered in the first place.
When scale starts working against the brand
The scale of the spam problem gives some context. In the April-June quarter of FY2026-27, AI-based systems deployed by telecom operators flagged 22.99 billion incoming calls as suspected spam, according to government data.
“Consumers are not necessarily against AI voice; they are against irrelevant and repetitive calls. The problem begins when automation makes it economically possible for brands to call consumers far more frequently than a human team ever could. Once that happens, even a legitimate brand can start sounding like spam,” says Yasin Hamidani, Director, Media Care Brand Solutions.
“Most people have stopped answering unknown numbers altogether,” shares Ankit Shastri, Senior Group Director, Gozoop Creative. He adds, “Consumers often check caller-identification labels before answering, and the first few seconds can determine whether they stay on the call. A slight pause, or a voice that is a little too polished, and it is over for the brand.”
For Bajrang Saharan, Founder & CEO, PressMate OS, the issue is even more basic. “If a call is about something real, say a pickup that’s actually scheduled or a payment that’s genuinely due, customers don’t mind it much,” he shares. But repeat the same call without adding anything new, and the response changes. “The polish of the voice barely matters at that point.”
Context may matter more than how human the bot sounds
That puts the focus on what the AI knows before it makes the call. Saharan believes automated calls work better when they are connected to something the consumer has actually done. “They make sure every automated call connects back to something the customer did, not something a marketing calendar decided. If someone booked a pickup last night, a call referencing that pickup lands completely differently than a generic outreach call would.”
Shastri makes a similar point, arguing that consumers want to know why a brand is calling rather than simply who is calling. “The calls that survive are short, tied to something the customer actually did, and easy to exit in one step,” he notes.
For brands, that means the customer journey has to determine when an AI voice enters the conversation. A payment reminder, delivery update or service notification may have a clear trigger. A repeated collection call after an issue has already been resolved risks turning a service interaction into an irritation.
Roshan Mohan, Co-founder and CMO of FlowBlinq, mentions that the technology itself is not the core problem. “Cold calls have always been mostly intrusive. AI has simply made it possible to do them at a much bigger scale and lower cost.”
He argues that AI should be used as a guide for defined processes rather than as a mechanism to maximise call volumes. “The objective shouldn’t be to make AI sound more human; it should be to make the interaction more relevant, useful, and consistent with the experience the brand wants to deliver.”
The efficiency trap
A system that makes thousands of calls may look efficient on an operational dashboard. But if those calls are ignored, rejected or reported as spam, the efficiency comes with a customer-experience cost.
Mohan adds, “There needs to be clear safeguards around consent, frequency, timing, and the purpose of every interaction. The AI should also be able to recognise when a conversation is no longer useful or needs to be handed over to a human.”
Saharan believes this also places greater responsibility on businesses adopting automated calling. “You can’t just claim you’re being responsible anymore, you have to document consent and purpose properly,” he says, adding, “For traditional businesses adopting automation for the first time—whether in logistics, services or collections—getting consent tracking and call logging right from the outset matters far more than how sophisticated the voice technology is.”
Could permission become the new starting point?
Shastri suggests that brands could go one step further by asking for permission before placing the AI call. “Brands could maybe send a message first, stating that an AI assistant would like to call about a specific topic, and allow the call to happen only if the customer says yes. That turns an interruption into an appointment,” he notes.
Mohan points to another possibility: moving some customer acquisition away from outbound calling altogether. Voice search, AI assistants and shopping agents allow consumers to initiate the interaction themselves, creating a different starting point for brands. Voice AI may still have a role in sales, service and collections. But as the number of automated calls grows, brands will have to be more careful about which calls they make, how often they make them and whether the customer has any reason to answer them.
