In India’s booming e‑commerce scene, one founder made a brutal bet on artificial intelligence that shocked his own workforce.
In the summer of 2023, Suumit Shah, boss of online retail platform Dukaan, laid off almost his entire support team and replaced them with chatbots. A year later, he is proudly sharing the first results, while critics warn his “experiment” could be a glimpse of where millions of jobs are heading.
An online chief who swapped humans for chatbots
Dukaan is a Bangalore-based startup that helps small merchants set up digital storefronts and sell online. Like most e‑commerce platforms, it relied heavily on human customer support agents to answer questions, fix orders and calm angry buyers.
That changed dramatically in mid‑2023. Shah decided that human agents had become too slow, too costly and too hard to scale. He chose to cut 90% of the customer support workforce and deploy an AI-powered chatbot system in their place.
The move triggered a wave of outrage on social media. Critics accused him of treating people as disposable and turning a real company into a lab test for AI. Supporters, on the other hand, praised him for moving fast and proving that generative AI could run a core business function at industrial scale.
Shah’s decision to automate nine out of ten roles with AI has become one of the starkest real-life examples of large-scale white-collar replacement.
At the time, even some tech optimists expected a messy transition: botched replies, furious customers, and a quiet return to human teams after a few quarters. Instead, Shah insists things went the opposite way.
A “positive” report card, at least for the company
One year after the layoffs, the Dukaan chief describes the outcome in strikingly upbeat terms. According to his figures, customer service metrics have improved sharply since chatbots took over.
Response times cut from minutes to seconds
Before the switch, human agents needed close to two minutes on average to respond to a customer query. That is not terrible in call-centre terms, but far from instant.
With the chatbot in charge, the platform now replies almost immediately. Customers type a question and see an answer pop up in seconds, at any time of day or night, without waiting in a queue.
➡️ This haircut quietly improves hair appearance even when you do nothing
➡️ Major breakthrough: the world’s longest venomous snake is actually four distinct species
➡️ Goodbye air fryer as a new zero-oil device delivering even crispier results wins over consumers
➡️ The brilliant salt trick in dishwashing liquid will solve your biggest kitchen headache
➡️ If your bank card PIN is on this list, change it fast: your bank account could be emptied
➡️ Rushed to emergency surgery, she healed in just 24 hours thanks to a Coca-Cola treatment
➡️ An AI detector questions the human origin of one of history’s most important texts
Shah also claims the time needed to actually resolve an issue has plummeted. Instead of going back and forth with an overwhelmed agent for more than two hours, users now see problems sorted within a few minutes in many cases.
The company reports that AI has slashed support response times from nearly two minutes to near-instant replies, and cut resolution times from hours to minutes.
On the company’s side, costs have also changed. Dozens of salaries, training programmes, office space, and management layers have been replaced by cloud bills, AI model licences, and a small technical team overseeing the system. The firm can scale support to busy shopping periods without hiring temporary staff.
What the company says it has gained
- Near-instant responses for most customer queries
- Faster resolution of common delivery and payment problems
- Lower staffing and training costs for the support function
- 24/7 availability across time zones without shift work
For Shah, this confirms his belief that AI can handle routine service work more efficiently than humans, at least in a structured environment like an e‑commerce helpdesk.
Behind the numbers, a brutal shock for workers
The positive spin from the boardroom contrasts with the experience of employees who lost their livelihoods overnight. Many had specialised in customer support, a role already under wage pressure in India’s tech hubs.
The replacement of 90% of staff in a single team is rare even in Silicon Valley. For unions and labour advocates, the Dukaan case is a warning: once an AI system proves itself, employers may be tempted to move much faster on layoffs than in earlier waves of automation.
Tech companies usually frame AI as a way to “augment” workers. In this case, the stated aim was to remove them. That difference fuels anger among critics who argue that the change was not about new opportunities, but about shifting value from salaries to shareholders.
For thousands of workers watching from the sidelines, Dukaan’s shift to AI looks less like future-of-work hype and more like a spreadsheet-driven decision with human consequences.
A polarised debate over replacing humans with AI
The Dukaan story feeds directly into a global argument that has been building for years: should AI assist people, or replace them?
The case made by AI supporters
Pro‑AI voices argue that such tools can take over repetitive and highly predictable tasks. In customer service, that could mean answering standard questions about deliveries, refunds, account access and product details.
They claim this frees people to handle complex, sensitive cases that need empathy, negotiation or judgment. In theory, that leads to more satisfying roles, less burnout, and better service for customers with unusual problems.
Fans also point to macro benefits. If companies cut costs, they may lower prices, invest in new offerings, or hire in areas where humans still excel, such as product design, marketing or merchant onboarding.
The concerns of AI sceptics
On the other side, sceptics see Dukaan as proof that when AI works well, executives will be tempted to push humans out entirely, not just “augment” them. Customer support roles, data entry positions and back-office jobs could be first in line.
They also worry about quality and fairness. An AI system trained on past data can misinterpret slang, miss cultural nuance, or handle complaints bluntly. Customers with unusual needs or accessibility requirements may struggle to get real help.
There is also the question of accountability. When a chatbot offers a wrong solution or denies a refund unfairly, who carries responsibility: the machine, the engineers, or the executive who chose the system?
How AI support actually works in practice
Behind the scenes, Dukaan’s chatbot most likely combines several AI techniques: natural language processing to understand messages, a knowledge base with company policies, and integration with order and payment systems.
A typical interaction might look like this:
| Step | What the chatbot does |
|---|---|
| 1. Greeting | Welcomes the user and asks them to describe the issue in their own words. |
| 2. Understanding | Uses language models to classify whether the problem is about delivery, payment, returns or something else. |
| 3. Lookup | Pulls data from internal systems: order status, payment confirmation, delivery partner updates. |
| 4. Action | Suggests a solution, such as resending an item, starting a refund or updating an address, based on company rules. |
| 5. Escalation | If the case falls outside known patterns or triggers risk flags, the system forwards it to a human supervisor. |
In theory, this blend of pattern recognition and system access lets the AI handle thousands of similar tickets faster than any human team. The risk is that rare or sensitive cases end up mishandled if escalation rules are too strict or supervisors are overwhelmed.
What this means for other companies and workers
For start‑ups and mid‑sized platforms, Dukaan’s numbers will be tempting. If a small firm can cut support staff by 90% and keep customers happy, investors will ask why others are not doing the same.
Larger brands may move more slowly due to reputational risk. A global retailer has more to lose from a viral story about a chatbot mishandling a harassment complaint or denying a legitimate refund. Still, many are already testing AI assistants in limited channels such as FAQ chats or internal helpdesks.
For workers in customer support, data processing or basic admin roles, this raises hard questions. Skills that once felt safe can now be simulated by large language models at a fraction of the cost. People in those roles may need to pivot towards tasks that are harder to automate: negotiation, sales, content creation, or managing AI systems themselves.
Key concepts and practical scenarios
Two ideas sit at the heart of this story and often cause confusion:
- Automation: handing a task entirely to software or machines, with minimal human involvement once things are set up.
- Augmentation: using tools to assist human workers, speeding them up while leaving final decisions to people.
Dukaan’s case leans decisively towards automation. A different company could choose augmentation instead. For example, a bank might let AI draft replies to customers, while human agents review and personalise them before sending. That kind of setup changes the nature of the job without removing the worker.
Another scenario involves hybrid teams. An insurer could use bots to gather initial details for a claim, run basic checks, and then pass a structured file to a human claims officer. The officer spends less time asking routine questions and more time judging grey areas and preventing fraud.
Each path comes with trade-offs. Full automation promises speed and savings but carries reputational and social risk. Augmentation keeps people in the loop but may deliver smaller cost reductions and still demands new training.
As AI tools mature, more business leaders will face the same choice that confronted Suumit Shah. The outcome at Dukaan shows what can happen when a company picks the most aggressive option, and sticks with it long enough to gather real data.
Originally posted 2026-02-05 13:38:39.