We Deployed the Bot First and Asked the Questions Later – An Indian View on AI Outrunning Its Own Governance

A European survey landed in front of me this week, and I can’t stop turning over one pair of numbers. CallMiner asked 200 senior leaders across Western and Central Europe about AI in customer experience. 99% said they were under pressure to scale it. Only 38% said they had a clear approach to governing it. Seven in ten admitted, in so many words, that speed of adoption usually beats compliance.

I read that and let out the tired kind of laugh. Because if you ran the same survey in India and people answered truthfully, I suspect the gap would be wider, not narrower.

Let me explain what I mean, from the ground.

The Bots Are Already Everywhere

Open any banking app here, and a bot is waiting.

HDFC has Eva, ICICI has iPal, Axis has Axis Aha.

I order dinner, and a chatbot on Swiggy assures me my rider is two minutes away while he is plainly parked outside the wrong building.

I missed a train, and IRCTC’s AskDISHA processed the refund query before a human agent could even pick up.

ServiceNow found that 80% of Indian consumers now use AI chatbots to check on complaints, get product recommendations, or find a self-help guide.

Our conversational AI market was worth about ₹38 billion in 2024 and is projected to reach ₹152 billion by 2030.

Notice what is missing from all of that. Nobody is debating whether to deploy. That decision was made for the rest of us a while ago.

India’s AI Governance Is a Law

Here is where India and Europe part ways in a way I find fascinating. Europe has GDPR and the EU AI Act, both heavy, binding, and tiered by risk. India looked at that road and deliberately took the other one.

Our official AI philosophy even has a name: “Innovation over Restraint.” There is no standalone AI law. We govern technology through the old IT Act of 2000 and the Digital Personal Data Protection Act, which Parliament passed in 2023.

That sounds reassuring until you check the calendar. The DPDP Rules were only notified in November 2025, and full compliance is not required until May 2027.

So, the personal data of roughly 800 million Indians online is currently protected by a law that exists on paper but is not fully switched on. The bots, meanwhile, went live years ago.

I want to be fair to my own country. I like that we did not strangle the industry in its crib. We have something like 420,000 AI professionals and a real shot at the economic upside everyone keeps projecting. The IndiaAI Mission put more than ₹10,000 crore behind it. We now have an AI Safety Institute. This is not a nation asleep at the wheel. But “we’ll write the safety rules once everybody is already driving” is a particular kind of bet, and I keep noticing that I am a passenger in it, not the one who placed it.

The Language Problem, Multiplied by 22

The European survey flagged multilingual support as a serious headache. Sixty-four percent called it a major challenge, and they are juggling maybe a handful of languages per market. India has 22 official languages and hundreds more that people actually speak every day.

A bot that is articulate in English turns into a flustered intern the moment my mother switches into Marathi halfway through a sentence, or a farmer in Bihar asks his question in Bhojpuri.

The government knows this is the real battleground. It built Bhashini, a national language stack, and in February it launched VoicERA, an open-source voice layer, with help from AI4Bharat.

At this year’s AI summit, NPCI put the stakes in plain terms: around 500 million Indians use UPI, and the next 300 million will not arrive in English. They will arrive by voice, in their own tongues.

Whoever’s bot understands them well and treats them fairly gets to serve the largest banking expansion left anywhere on earth. Whoever’s bot fumbles, it does real harm to the people who can least afford a misunderstanding about their own money.

Trust is the Real Ceiling

This is the part that keeps me up at night. The CallMiner survey concluded that trust, not technology, now decides how far AI can scale, and that accuracy is the single biggest driver of that trust. I believe it, because I have watched trust break in public.

Remember the Air Canada case. A grieving passenger asked the airline’s chatbot about bereavement fares. The bot invented a refund policy that did not exist, and a tribunal ordered the airline to honor it. The machine’s hallucination became a legally binding promise. Now transplant that liability into an Indian bank’s Hindi voice agent, running across a billion conversations, and sit with the size of it.

Then there is the darker edge. India’s financial sector is in the middle of a deepfake fraud wave: cloned voices, faked video calls, and the “your son has had an accident, transfer the money now” scam that sounds exactly like the person it impersonates.

The same voice AI meant to welcome the next 300 million is the same technology now being aimed at the last generation, the ones who learned to trust a familiar voice on the phone. I do not have a tidy answer for that. I only know that my own parents trust a voice far more readily than a text message, and that frightens me more than any survey statistic.

Even when nothing dramatic goes wrong, the everyday experience is often just bad. A Qualtrics study this year found that nearly one in five people who used AI for customer service received no benefit at all, a failure rate much higher than AI’s overall rate.

I have been stuck in those loops myself, typing “talk to agent, TALK TO AGENT” while a relentlessly cheerful bot offers me FAQ links. Too many companies here have bolted AI onto their call centers to shave costs, not to solve customers’ problems, and customers can feel the difference immediately.

Where I Land

So, what do I want? Not a moratorium. Not Europe’s full rulebook copied over either. What I want is the boring thing tucked inside that survey: visibility. The ability to know what the bot actually said to my mother in Marathi, whether it was accurate, and whether it treated her the same as it treats an English speaker in Mumbai. The companies that will earn trust here are not the ones with the slickest bot. They are the ones who can answer the simple question, “What is our AI telling people, and is it true?” Most of us, right now, cannot.

India placed a bet that we can scale first and govern as we go. In May 2027, when DPDP finally bites, we will learn whether that was confidence or just momentum. I am rooting for it to work.

I am also keeping a human being’s phone number saved, for the day it doesn’t.