Sachin Bhatia outlines how unified memory layers, AI-human orchestration, and data-sovereign infrastructure are redefining customer experience across the region.
Enterprises across the Middle East are entering a new phase of customer experience transformation, where AI is no longer confined to isolated pilots but is becoming core digital infrastructure. Against this backdrop, Exotel has introduced Harmony, its agentic AI-led CX orchestration platform, designed to unify voice, messaging, AI agents, analytics, and human interactions into a single, context-aware system. The move comes as the regional conversational AI market accelerates towards an estimated USD 2.3 billion by 2031, driven by national digital agendas such as the UAE National AI Strategy 2031 and Saudi Vision 2030.
Speaking to TahawulTech, Sachin Bhatia, Co-Founder and Chief Growth Officer at Exotel, explains that the next wave of customer experience will not be defined by automation alone, but by how effectively organisations combine AI efficiency with human judgment, empathy, and governance. He argues that fragmented CX architectures -- built on disconnected tools and point solutions -- are giving way to unified platforms built around real-time customer memory, AI-human orchestration, and continuous supervision.
Bhatia also highlights how data sovereignty, regulatory compliance, and trust are shaping enterprise AI strategies in markets such as the UAE and Saudi Arabia. As billions of AI agents are expected to manage customer interactions in the coming years, he believes CX platforms will evolve into critical digital infrastructure, with memory, context, and human oversight at their core.
Interview Excerpts
The UAE positions itself as a "digital-first" nation. Where do you see the country heading?
If you look at every major tech shift -- mobile, internet, apps -- the UAE has always adopted technology very fast, largely because of the diaspora and the pace of consumer adoption here. With AI, though, I'm seeing something different for the first time: consumers are not automatically excited. When people are frustrated, they usually want a human, not a machine.
We work with a food aggregator across the Gulf, where people call when they're hungry and angry -- "Where is my order?" What's surprising is that the success rate we're seeing in Kuwait or Saudi Arabia is higher than in the UAE. We always assumed the UAE would be more digitally receptive, but early AI experiments weren't very successful, so trust is still being built here.
"The top-down push -- from government and boardrooms -- is driving adoption beyond experimentation. The key is managing consumer behaviour: when someone is anxious or frustrated, they should be able to speak to a human; when it's transactional, bring in the machine."
And honestly, what's happening on the ground in the UAE is unprecedented compared to other emerging markets. I recently got my Emirates ID, went for a medical examination, and saw an autonomous coffee machine. There was also a robot distributing water. These are practical use cases where you don't need a human carrying a tray all the time -- automation makes sense. Overall, I think the UAE is far ahead of other markets we've seen.
In a world where billions of AI agents manage customer interactions, what will the customer experience platform of 2030 look like?
The early promise of AI was split into two narratives: AI will make humans better, and AI will replace humans. Both are true depending on the use case. In CX, I believe the future is about combining the best of both worlds.
When a task is repetitive, the platform should bring in an AI agent. When there's anxiety, fear, or judgment involved, a human should come in. The centrepiece of this future platform is memory -- customers shouldn't have to repeat themselves.
If I've already spoken to an AI agent and it has collected information, when a human joins the conversation, they shouldn't start with "How can I help you?" They should start with context -- "I see you were trying to block your card. Let me help you."
The goal is a central memory and context layer across channels -- WhatsApp, email, phone -- so the customer feels heard and not like a stranger every time. That's why we launched Harmony in the UAE -- so enterprises can have humans and AI agents working together across channels, instead of piecemeal solutions for individual use cases.
Do you see AI agents eventually handling entire customer journeys autonomously, with humans stepping in only for complex emotional or strategic moments?
It depends on the use case. In areas like medical triaging, accountability still sits with humans because of legal and regulatory frameworks. I don't see bots doing more than collecting information there -- at least for now. Technology can make recommendations today, but governance and accountability frameworks aren't ready for that shift yet. For most other use cases, I do see a blend.
"Humans will do what only humans can do -- take judgment calls, soothe customers when emotion is involved, handle anxiety. Most context and data collection will be handled by AI agents."
How will agentic AI reshape the role, skills, and value of contact centre agents over the next five years?
Historically, contact centre agents were trained to communicate clearly and follow scripted conversations. I think scripted conversations will die -- you won't need them anymore. And frankly, scripted work is a minimal use of human intelligence and creativity. Humans bring perception, judgment, and the ability to read what isn't explicitly said. In the future, contact centre agents become what I call the "last line of defense" for an enterprise. They'll handle the cases where policies don't solve the customer's problem.
A machine can block a card -- authenticate and block, done. But "I lost my job and I can't pay my EMI this month" is not a conversation I want a machine to have right now. You need a human to empathise, suggest solutions, and make judgment calls. So yes, volumes will reduce because transactions get automated, but you can't get those human conversations wrong. The accountability and impact of the human layer will become very high. Customer experience will become front and centre -- because products and offerings will look the same, and experience will be the differentiator.
Could real-time customer memory layers evolve into predictive systems that resolve issues before customers even reach out?
Think about YouTube or Spotify -- they understand your mood and preferences because they've observed what you listen to and when. The same will happen in customer conversations. Also, look at how companies measure customer experience today -- surveys. Who fills surveys? I haven't filled one in the last 30 days, but I've definitely had conversations with service providers. My view is: if you listen, you don't need to ask.
If the CEO of a company could listen to every conversation, they would have a crystal-clear view of what's wrong in the organisation. With these conversations, a perceptive memory layer will be built for each customer. When someone calls, the system can predict the reason. It can say, "I see you lost your luggage -- let me track it," or "We know your flight is delayed," without the customer explaining everything again. I think that future is coming faster than people expect, because the memory layer is built from real conversations and models.
Will future CX platforms become critical national digital infrastructure as data sovereignty and compliance requirements intensify in the Middle East?
Governments want sovereignty and governance of data to remain with the owner of the data. What you don't want is data going outside without your knowledge and being used for other purposes. Frameworks like GDPR addressed parts of this, but now we need to extend that thinking to AI. Governments and large enterprises have to be conscious of this.
That's why we invested heavily in local infrastructure -- in the UAE and Saudi Arabia -- so consumer data doesn't go out if an enterprise wants full control. For heavily regulated sectors -- government, banks, insurance -- this is non-negotiable. It might slow down speed, but it's the right way. Anyone designing for these enterprises has to be ready to invest in sovereignty platforms, because trust depends on it -- especially when financial and national data is involved.
Harmony is built around a real-time customer memory layer and AI-human orchestration. How does this change CX strategy versus layering AI on legacy systems?
Earlier, enterprises approached AI by building point solutions -- one sales bot, one collections bot, one service bot -- mostly as proof of concept. But going forward, those piecemeal approaches won't make sense because you need to understand the customer across the full journey.
Harmony is built for that. The memory layer works across channels and conversations -- email, chat, phone -- everything. And the orchestration is the key: if I already understand a customer and I suspect they're frustrated, I don't need to put a bot in front of them; I can route to a human. If it's a simple task like blocking a card, there should be no queue -- an AI agent should handle it instantly.
Large enterprises also can't replace legacy platforms overnight. So I see incremental shifts: first, remove the IVR "press 1, press 2" layer and use an AI agent to understand intent and route correctly. Once that's in place, you start automating specific intents, step-by-step. Then the memory layer grows, so the experience becomes contextual even before the customer speaks.
I'll give you a real example from a large food aggregator: "Where is my order?" sounds simple, but people started asking extra things -- change the rider, avoid plastic cutlery, and more. Unless you train for those intents, the bot fails. So we created an observer loop -- review conversations daily and reinforce learning back into the bot. Today, about 70% of those conversations are handled by the bot, but it took six months to get there. The promise is big, but it takes hard work to make it effective without losing customer trust.
Harmony enables up to 60% automation with Human-in-the-Loop supervision. How should enterprises rethink governance, accountability, and measurement?
First, the 60% is not theoretical -- we've seen it in real deployments. In our largest use case, we handle about 3.6 million conversations a year, and we've reached about 70% containment. But it's not universal -- containment varies by use case. Transactional tasks can be highly automated; emotional and anxiety-driven conversations start low and require trust-building over time.
Customers often try to bypass automation -- like pressing 9 in an IVR to reach a human. So you win trust gradually. The bot can say, "I'll transfer you to a human -- can I first understand why you're calling so I route you correctly?" Then, sometimes the bot can resolve it quickly, and the customer may not even need the human.
"From an enterprise point of view, my recommendation is: don't build standalone bots. Think governance first. You need observability. If AI is answering something wrong -- or promising something incorrect -- you must catch it, because AI is representing your company."
So you need an observer and governance layer across channels and use cases that can flag when a conversation has gone bad and bring in a human to repair it. This can't be something you discover every quarter. It has to happen in near real time -- every minute.
Will this disruption impact the BPO/contact centre industry itself?
BPO business models will change. The original reason BPOs existed was to manage non-core operations, handle seasonality, and support overflow without enterprises staffing for peak demand. That logic still matters, but the "seat-based" model -- selling human agents by volume -- won't survive as-is.
BPOs will need to move to outcome-based models. Instead of selling seats and agent hours, they'll sell outcomes priced per resolution or per business result -- and they'll augment delivery with AI. If BPOs stay protective of the old model, they will be disrupted.
I also agree that more large enterprises will bring these functions in-house because of governance and data sovereignty. BPOs will still exist, but they'll shift toward niche, specialised tasks and outcome-based delivery. And this change won't be limited to BPOs -- IT services will also move from effort-based models to outcome-based models, especially as software coding is being disrupted.
Do we see a lot of upskilling and reskilling happening in the BPO industry?
No doubt. You will have AI supervisors. Agents will do non-scripted, high-accountability conversations. And the people most impacted by this shift will have to adopt AI the most to remain relevant and add value. This won't be a stepping-stone job for freshers anymore -- it becomes a last line of defense for the enterprise, where judgment, empathy, and decision-making matter.
Related Articles