A message can tell a customer that something has gone wrong. A useful conversation can help them put it right. That difference became the central lesson of my conversation with Beerud Sheth on Dilse Omni Talks.
Beerud is the Co-Founder and CEO of Gupshup, a global conversational AI platform for marketing, commerce, and customer support. An IIT Bombay and MIT alumnus, he previously co-founded Elance, which later became Upwork. At Gupshup, he has led the journey from a messaging platform to a global conversational AI company.
I wanted to understand what that evolution really means for brands. Are we simply making business messaging more sophisticated, or are we changing the relationship between businesses and customers?
His answer brought us back to something refreshingly simple: the customer’s need has not changed nearly as much as the technology available to serve it.
Table of Contents
1. The Technology Changes. The Mission Stays the Same.
2. The Opportunity Is Bigger Than Enterprise Adoption
3. When Automation Becomes an Endless Loop
4. Why Context Makes an AI Agent Useful
5. From Answering Questions to Taking Action
The Real Shift Is From Sending to Serving
1. The Technology Changes. The Mission Stays the Same.
I began by asking Beerud about Gupshup’s progression from messaging company to platform company, and now to AI company. His response was not about abandoning an old business for a fashionable new one.
“The more things change, the more they stay the same.”
Beerud Sheth
Gupshup still serves businesses that want to engage large numbers of consumers. Those businesses still need to inform, assist, sell, and support. What has changed is how much technology can now do within those interactions.
Beerud explained this through a familiar situation: a delayed flight.
An SMS tells you the flight is delayed. A WhatsApp message can deliver the same information with an image and richer presentation. An AI-powered conversation can go further, helping you explore rebooking, rescheduling, cancellation, upgrades, or refunds, perhaps alongside a voucher.
The difference is not merely visual. It is the shift from communicating a problem to helping resolve it.
The Engagement Ladder: Notify → Enrich → Resolve
I would turn Beerud’s example into this simple framework. It is my synthesis of his explanation, not a named Gupshup model:
Notify: Tell the customer what happened. The flight-delay SMS does this.
Enrich: Make the information clearer and more engaging through a richer messaging format.
Resolve: Use conversation to understand what the customer wants to do next and help them act.
This is a useful test for any customer engagement initiative. Have we improved the message, or have we improved the customer’s ability to get something done?
Beerud compared earlier messaging improvements with the progression from plain-text email to HTML and more interactive formats. But he sees AI as a much bigger leap. It expands what the interaction can accomplish, not just how it looks.
2. The Opportunity Is Bigger Than Enterprise Adoption
Another point stayed with me: Beerud believes enterprise adoption remains well behind the capabilities already available.
That matters because the conversation around AI often focuses on what might become possible next. His emphasis was more immediate. There is already considerable capability that businesses and marketers have yet to put to work.
He described two sides to this opportunity:
Better customer interactions: AI enables businesses to do more within a conversation, across more channels.
More productive marketers: AI also creates opportunities for the people designing and managing customer engagement to work more effectively.
Gupshup’s direction, as he described it, is to bring those capabilities together, simplify them, and make adoption easier.
My takeaway was that omnichannel strategy cannot stop at channel availability. Being present on SMS, WhatsApp, and other channels is useful, but the bigger question is what help the customer can receive there. This connects closely with the broader shift toward conversational commerce and customer engagement.
3. When Automation Becomes an Endless Loop
Then Beerud turned the question around. Had I encountered a good AI agent for customer interactions?
I had both a good example and a bad one. I started with the bad one, because failed automation makes the design problem painfully clear.
When I began my consulting work, I was experimenting with Google, Amazon, and Microsoft’s cloud platforms. After losing access to my Azure account, I found myself relying on bot-led support to recover it.
What followed was about a month of circular instructions. One route pointed to another, then another, without getting me back into the account. Eventually, I gave up on using the platform.
This was my experience, not a verdict on every customer’s experience with Azure. But it illustrates a failure mode that brands should take seriously: a support system can keep responding without actually helping.
The customer does not care how elaborate the workflow is. They care whether there is a way out.
A Human Exit Is Part of the Design
My conclusion was straightforward: automated customer support needs an exit and access to a person when the automated path fails.
Over-engineering can become a problem when the system sends people through more steps instead of bringing them closer to resolution. A chatbot should not become a barrier between a customer and the business.
For me, the practical questions are:
Does the customer move closer to solving the problem?
If the suggested route fails, is there a genuinely different next step?
Can the customer reach a human rather than repeat the same loop?
These are service-design questions before they are AI questions. Adding conversational technology does not remove the responsibility to design a workable support journey.
4. Why Context Makes an AI Agent Useful
My positive example was Amazon’s shopping assistant, Rufus.
What interested me was not simply that Amazon had placed a chatbot inside its shopping app. It was that the assistant operates within an environment rich in shopping context.
Amazon has product listings, reviews, ratings, pricing information, relationships with vendors, and an advertising ecosystem. That surrounding context creates the conditions for more relevant assistance.
The point I made to Beerud was:
“The bot is not great because of its tech. I think so the bot is great because of its context.”
Saurabh Agrawal
Beerud agreed that context makes the difference.
One use case I discussed was price history. When buying online, a recurring concern is whether today’s price is a good one. Prices fluctuate, and customers want some basis for deciding when to buy.
I described asking for an iPhone 16’s price history to understand how its price had moved. The value is not another product description. It is information that addresses a specific purchase anxiety.
Another example was a parent looking for a gift for a son heading to college, with a budget below 10,000 and a preference for well-reviewed products.
A helpful assistant can suggest a small selection and then ask whether the recipient prefers technology or clothing. That follow-up is important. The customer has expressed a situation, not a perfectly specified product search.
A good conversation helps clarify the requirement instead of demanding that the customer arrive with it fully formed. That is also why I find the chatbot conversion question so relevant: the quality of assistance deserves attention before we judge the interface alone.
5. From Answering Questions to Taking Action
Beerud then suggested the next step: what if the assistant could monitor a product and buy it when the price fell below a chosen level?
We were discussing a desired capability, not demonstrating that Rufus already performs it. But the idea neatly captured the difference between an assistant that answers and an agent that acts.
I could relate immediately. While buying a smartwatch, I had set myself a daily reminder to check its changing price.
Beerud laughed at the idea of keeping a watch on a smartwatch. Behind the joke was a useful observation: the customer was still doing repetitive work that an agent might eventually handle.
My Practical Framework: Context → Conversation → Completion
Bringing the discussion together, I would use three connected stages to evaluate a conversational AI experience:
Context: Does it have the relevant information? For shopping, that could mean product details, reviews, and pricing.
Conversation: Can it clarify intent and help the customer explore suitable choices?
Completion: Can it help reach an outcome, or provide a human handoff when it cannot?
This is my working framework from the conversation. It links the flight example, the failed support loop, and the shopping assistant: each succeeds or fails on whether useful information becomes useful help.
The Real Shift Is From Sending to Serving
I came away from this conversation with a clearer way to think about business messaging. The ambition should not be to send more sophisticated messages. It should be to help customers make progress.
Beerud’s flight example showed the opportunity. My support experience showed the risk. Rufus illustrated the power of context, and the price-monitoring idea pointed toward more action-oriented agents.
The technology may change dramatically. The standard remains simple: did we make the customer’s task easier?
I am Saurabh Agrawal and we come with a new episode on Dilse omni talks every fortnight and cover different aspect of omnichannel with amazing speakers.
