A friend who runs a foreign education advisory business asked me recently, “Everyone tells me I need a chatbot. Do I?”
My answer was the one I give for most AI questions: it depends on what you are actually trying to fix. Chatbots are neither the miracle the vendors promise nor the disaster the horror stories suggest. After watching this space closely — and having spent three decades watching earlier “must-have” technologies come and go — I can tell you the truth sits somewhere in the middle. Let me walk you through both sides, with real names and real numbers.
First, the honest good news.
When a chatbot is deployed on the right problems, the results are hard to argue with.
Zomato built an AI support assistant called Zia and reported that it doubled customer satisfaction scores while cutting average response times by 75 per cent, to under 10 seconds — while handling over 1,000 messages a minute during peak hours.
Swiggy, working with Databricks, built an enterprise-scale AI agent that now handles millions of support conversations, and notably, they track quality obsessively: conversation quality scores, factual accuracy, resolution efficiency. That last point matters, and I will come back to it.
Now, you may say, “I am not Zomato.” Fair. But the underlying pattern applies at any scale. Both companies pointed their AI at the same category of work: high-volume, repetitive, low-judgement queries. Where is my order? How do I cancel? What is the refund policy? This is exactly the work that burns out a two-person support team in a small business, and it is the work chatbots handle genuinely well.
For Indian small businesses specifically, the opportunity is even more concrete because our customers live on WhatsApp. Platforms like Wati, Haptik and Yellow.ai have made WhatsApp automation accessible to businesses with three-person teams, not just enterprises. WhatsApp messages see open rates of 70–80 per cent, against 15–20 per cent for email. A D2C brand or a local services business that answers routine queries on WhatsApp within minutes — at midnight, on Sunday, during Diwali rush — is delivering something a small team cannot do manually.
Gartner has estimated conversational AI will cut contact-centre labour costs by $80 billion by 2026. You will not save billions, but the same economics work in miniature: a chatbot that deflects even half of your routine queries frees real hours every single day.
Now, the part the vendors will not tell you.
Three stories every business owner should know before signing a chatbot contract.
Air Canada. In 2024, the airline’s chatbot confidently told a grieving passenger he could book a full-fare ticket and claim a bereavement discount afterwards. The policy said no such thing. When the customer sought his refund, Air Canada argued — I am not making this up — that the chatbot was “a separate legal entity responsible for its own actions.” A Canadian tribunal called this absurd and ordered the airline to pay.
The lesson: legally and morally, your chatbot’s words are your words. If it promises a refund, you owe a refund.
DPD. The UK delivery firm’s chatbot, after an update, was coaxed by a frustrated customer into swearing at him, calling DPD “the worst delivery firm in the world,” and composing a poem about its own company’s failures. It went viral, naturally.
The lesson: an AI chatbot connected to your brand can be provoked into saying things no employee ever would, unless it is properly constrained and tested.
Klarna. This one deserves the most attention. The Swedish fintech announced in 2024 that its AI assistant was doing the work of 700 customer service agents. It became the poster child for AI replacing humans. Then, quietly, Klarna began rehiring human agents. The CEO admitted publicly that the company had focused too much on cost and not enough on customer experience — quality had suffered.
The lesson: even a company with world-class engineering resources discovered that removing humans from customer service entirely was a mistake. If Klarna could not make “AI-only” work, a small business should not bet its reputation on it.
So what is the truth?
Here it is, as plainly as I can put it: chatbots are excellent assistants and poor replacements.
They excel at the predictable 60–70 per cent of queries — order status, timings, pricing, FAQs, appointment booking, basic troubleshooting. They fail, sometimes expensively, at the remaining 30–40 per cent: complaints, exceptions, emotional situations, anything requiring judgement or empathy. The businesses that win with chatbots are designed for both halves. The businesses that get burned pretend the second half does not exist.
Notice what Zomato and Swiggy did that Air Canada did not: they measured accuracy continuously, and they kept humans in the loop for anything the AI could not resolve cleanly. The technology was similar. The discipline was different.
A practical way to decide before you spend a rupee.
If you are considering a chatbot for your business, sit down with your team for one hour and do this simple exercise. List your last 100 customer queries — from WhatsApp, calls, emails, walk-ins. Sort them into two piles: “same question, same answer, every time” and “needed a human to think.” If the first pile is 50 or more, a chatbot will likely pay for itself.
If most of your queries are complex, consultative or emotional — as they often are in professional services — a chatbot may frustrate more customers than it helps.
If you do proceed, four rules from the case studies above will protect you.
Start with your top 10 repetitive questions only; do not try to automate everything on day one. Always give customers a clearly visible path to a human — “type AGENT to talk to a person” — because hiding this is the single most common cause of chatbot rage.
Review the bot’s actual conversations weekly in the first three months; you will be surprised, sometimes pleasantly, sometimes not. And treat every word your bot says as a commitment your business is making, because as Air Canada learned, courts will treat it exactly that way.
The summary.
We have seen this pattern across 35 years: every powerful tool gets oversold on the way up and unfairly dismissed on the way down. Chatbots are no different. They will not transform your business. They will, if deployed thoughtfully, give your small team the response speed of a much larger one — and give your customers answers at 11 pm when your office is dark. That is not magic. It is just good operations.
The businesses getting this right are not asking, “How do I replace my support staff?” They are asking, “How do I let my people spend their time on the conversations that actually need them?” That question, in my experience, is where the real return lives.
Sources: Together AI – Zomato case study · Databricks – Swiggy’s AI support agent · Forbes – Air Canada chatbot ruling · CX Today – Air Canada tribunal · DEV Community – Klarna and AI reversals · ChatMaxima – AI customer support statistics · Wati case studies
Over to you: Have you used a chatbot for your business — or been on the receiving end of a terrible one? Share your experience in the comments. Your stories often shape future editions.
If you found this useful, do subscribe to The AI Compass and share it with a business owner who is being pitched a chatbot this week.
