Voice communication is evolving from a simple conversation channel to an intelligent business asset. What do you believe is driving this transformation, and how should enterprises prepare for it?
Voice did not suddenly become intelligent. What has actually happened is that the technology around it has finally caught up. For almost twenty years, voice was the one channel that stayed passive. A call would happen; maybe it got recorded, and after that the data simply died. Everything useful in that conversation- the intent, the objections, the commitments, and the sentiment- was lost the moment the person hung up. Today, speech-to-text has become cheap and accurate, and the models can actually understand what was said and act on it. So the call is no longer just an event. It becomes a source of data.
That is the real shift. And remember, voice is still where the highest-value conversations are happening, whether it is sales, collections, support or hiring. Text is fine for the routine things. But the moment money or trust is involved, people pick up the phone. So when you make voice intelligent, you are making your most important interactions measurable and connected for the first time.
Subhash Kalluri, Founder, FreJun
An advice to enterprises is to not begin with AI agents. Begin with the plumbing. Most organisations today cannot even get a call and its outcome into the same system. Fix that first. Get every call structured and flowing into your CRM and your systems of record. The companies that will win here are not the ones with the fanciest bot. They are the ones whose voice data actually goes somewhere.
Many organisations still rely on disconnected communication, CRM, and workflow tools. How can AI-powered voice infrastructure help businesses eliminate these silos and improve operational efficiency?
The silo problem is almost always the same story. The call is sitting in one tool, the customer record is in another, and the workflow is in some third place. So the representative finishes a call and then spends the next few minutes copying and pasting notes into the CRM, either badly or not at all. And then the manager is making decisions on data that is half fiction.
AI-powered voice infrastructure fixes this by treating the call itself as a structured event, not as a separate activity that somebody has to remember to log later. The transcript, the outcome, the next step, all of it flows into the systems already running the business. This is exactly the problem we built FreJun CP for. It is native to the CRM by design, so the call and everything it produces sit inside Salesforce, HubSpot, Zoho, or wherever the team is already working. The representative is not switching tabs to yet another dialler.
There’s a limit too; tools by themselves do not remove silos. A large part of the silo is organisational. It is about who owns the data, and whether the process was built around the customer or around the organisation chart. Software can take away the friction and the manual data entry, and that itself is a significant efficiency gain. But if the underlying process is broken, connected tools will only help you do the wrong thing faster. So the infrastructure removes the excuse. The discipline is still on you.
As AI automates more customer interactions, how can businesses ensure they maintain empathy, personalisation, and trust while delivering seamless experiences at scale?
The question assumes automation and empathy work against each other. In practice, they do not. What kills empathy is bad automation, and we have been living with bad automation for twenty years. The press one for sales maze, the bot that loops you back to the beginning, and the hold music that never ends. People did not lose trust because things got automated. They lost trust because it was done badly.
If you do it properly, an automated interaction can actually be more personal than a human one, because it remembers you. It knows your history, your last three calls, what you bought, and what is broken. A tired agent on his fortieth call of the day simply cannot do that.
So a few things matter. First, be transparent and tell the person they are speaking to an AI. They can make it out anyway, so pretending only insults them. Second, know when to hand over to a human. Anything high-stakes, emotional, or unclear should go to a person quickly. Trying to automate everything is a trap. And third, which people tend to forget, the call has to feel human at the physical level itself. If there is even half a second of lag, or the audio breaks up, no amount of model intelligence will save that call. That reliability layer is exactly what Teler is built to handle so that the agent sounds present and in the room and not like it is calling from far away. Empathy at scale is a design decision. It is not something you automatically lose the moment you automate.
FreJun has been at the forefront of building AI-powered voice solutions. Looking back on your entrepreneurial journey, what has been the biggest challenge in changing how businesses perceive and adopt modern voice infrastructure?
The most persistent challenge has been overcoming the perception that voice is a solved, commoditised problem. When people hear the word “calling”, they tend to disengage immediately, because they assume it is simply about phones. Convincing a business to view the voice layer as strategic infrastructure, rather than a cost line to be minimised, has been a genuinely difficult exercise. Telephony rarely excites anyone, whereas AI captures everybody’s attention. So a significant part of our work has been reframing voice as the layer that makes all of that AI actually usable in the real world.
This is a large part of why we built Teler as a separate product, a programmable voice API aimed squarely at developers and teams building AI agents. It lets us make the argument in the most direct way possible. Here is the infrastructure; build your voice agent on top of it and stop worrying about the layer underneath.
The second challenge is regulatory, and it is badly underappreciated. In India, if you want to run voice properly, you have to deal with DoT licensing. We hold a UL-VNO licence, and maintaining it is real, continuous work. Most people building voice agents today have no idea that this “boring” carrier and compliance layer is precisely where things quietly break the moment you try to scale. So educating the market on why licensed, compliant infrastructure matters, especially in markets like India and the UAE, is slow and very unglamorous work.
From a founder’s perspective, we have done all of this bootstrapped since 2016. Changing how a market perceives you without a large funding war chest to simply buy attention means you have to earn every bit of credibility the hard way, through customers, through being right, through showing up again and again. It is slower, yes. But the belief you build that way actually stays.
Looking ahead, what emerging trends – whether intelligent voice agents, contextual AI, or autonomous workflows – do you believe will define the next generation of enterprise communication over the next five years?
Everybody will point to intelligent voice agents, and they are correct, but that is the obvious part. The bigger shift is going from agents that only talk to agents that actually do the work. Right now most voice AI stops at the conversation. It answers your question, and the call ends. Over the next five years it will move to autonomous workflows, where the agent does not merely tell you your order is delayed but reschedules it, updates the record, and follows up, without a human in between. Voice becomes an interface to action, not just to information.
The second is contextual AI. Agents that carry memory across channels and across time, so the conversation is not being reset on every single call. That is what will finally make an automated interaction feel like a relationship and not just a transaction.
The real conviction is that the winner will not be the flashiest agent. It will be the infrastructure sitting underneath them. A power generation analogy can help explain this. Everybody is going to build the appliances, but very few will build the reliable grid that powers all of them. That utility layer, which has to be carrier grade, low latency, compliant, and available across geographies, is unexciting to talk about but completely essential. This is precisely the bet behind Teler, our programmable voice API for AI agents. We are choosing to build the grid rather than one more appliance.
And two more that people are not discussing enough. Regulation and localisation will become more important, not less, as governments turn their attention to AI voice. And in markets like ours, voice first will dominate, because for a very large population, speaking is simply easier than typing. So the next generation of enterprise communication will not be built in Silicon Valley’s image. A good part of it will be built for, and from, markets like India, the Middle East and South East Asia.
Enjoyed this interview? Now imagine yours. Write to:
jeevika@thefoundermedia.in
About The Author
