Dukaan and Suumit Shah: One Year After Replacing 90% of Staff With AI

A year later, his radical automation trial is beginning to alter how technology leaders view staff, profitability and the boundaries of artificial intelligence in the workplace.

An Indian startup that wagered almost everything on AI

The choice was made by Suumit Shah, the founder and CEO of Dukaan, a Bangalore-based platform enabling small retailers to operate online stores. With narrow margins and intense competition, Shah took a severe course of action: he let go of roughly 90% of the workforce, largely from customer support, and substituted AI-powered chatbots.

The decision triggered anger on social media and among labour campaigners at the time. Many regarded it as an early warning of what might unfold throughout the worldwide customer service sector. Others described it as an unsentimental yet rational move by a lean, venture-backed startup facing pressure to expand.

Shah’s stated goal was simple: cut costs sharply while making support faster and more efficient for customers.

Customer support departments are commonly among the first areas exposed to automation. Their staff deal with recurring questions, work from scripted responses and face strong demands to remain low-cost and available 24 hours a day. That combination is particularly appealing for an AI system.

A “positive” first-year report, according to the CEO

A year after the redundancies, Shah has offered his first detailed evaluation. In his view, the bet has paid off.

Based on the figures he has highlighted, the average time taken to answer customer questions has fallen from almost two minutes to near-immediate responses. Complaints and order-related problems that previously needed more than two hours to resolve can now frequently be completed in minutes.

The company reports near-instant answers and much faster problem resolution, at a fraction of the former staffing cost.

For an online retailer, this level of speed may mean fewer abandoned orders, fewer frustrated customers and possibly more repeat purchases. Dukaan says its users value the quicker support, although independently verified customer satisfaction data has not been widely released.

Financially, removing salary and office expenditure can substantially improve a startup’s short-term position. Support teams carry significant costs, including training, management, HR assistance and physical or digital infrastructure. In theory, a chatbot requires cloud computing, updates and periodic fine-tuning alone.

What changed within Dukaan?

Dukaan’s shift involved more than merely replacing people with software. The company had to rebuild its processes around AI systems and reconsider the purpose of its remaining human roles.

From human-first to AI-first support

Before the change, a standard customer exchange at Dukaan involved several stages: a user submitted a ticket, a human representative replied, the matter was escalated where necessary, and managers or technical employees were occasionally brought in. Following the restructuring, the chatbot became the first point of contact and, in most instances, the sole one.

• The bot welcomes customers and determines the issue.
• It retrieves details from order databases and help documents.
• It recommends solutions or handles refunds and amendments where authorised.
• Only difficult or exceptional cases are passed on internally.

Shah has stated that the overwhelming majority of enquiries can be resolved without a human being involved. If accurate, this would indicate that much of the earlier support workload was repetitive and governed by rules.

How the metrics compare

Dukaan has not published a complete collection of audited data, but it focuses on three principal measures:

Metric	Before AI rollout	After AI rollout
Average first response time	Just under 2 minutes	Almost instant
Average resolution time	Over 2 hours	Several minutes
Support staffing level	100% of original team	~10% of original team retained

The figures reflect a wider trend in early AI roll-outs: striking improvements in speed and obvious savings, alongside limited visibility into long-term customer retention and brand sentiment.

A divisive case study of AI replacing people

The Dukaan story brings into focus a question affecting numerous sectors: at what point does AI support people, and when does it simply take their place?

Advocates for business AI contend that tools such as chatbots eliminate tedious and repetitive tasks. They argue that this allows workers to move into creative, strategic or relationship-led positions. They also cite cheaper prices for customers and improved service access, particularly beyond conventional office hours.

For AI enthusiasts, Dukaan is proof that software can handle high-volume customer work faster, cheaper and at scale.

Critics interpret the case very differently. They fear that if one chief executive can remove 90% of a team and publicly praise the outcome, other businesses may copy the approach. This could threaten tens of thousands of support roles in markets where call centres and help desks are significant employers.

Ethical concerns also emerged when Shah first revealed the redundancies. Observers questioned whether Dukaan had made sufficient efforts to retrain or reassign the affected employees. Others challenged the social consequences of presenting such a severe reduction solely as a “productivity win”.

What this means for workers and businesses

Dukaan’s experience is likely to be watched closely by both startups and major companies. Generative AI tools are already being applied to customer service, content moderation and routine back-office tasks.

For employees, particularly in countries such as India and the Philippines, which host vast support centres, the episode underlines the need to acquire new capabilities. Workers able to oversee AI systems, manage complicated customer cases or create workflows are more likely to remain sought after.

For businesses, however, the lesson is less straightforward. Software can replace people and sharply reduce costs, but it can also reshape a brand’s connection with its customers. A chatbot may be quick but inflexible, leaving customers dissatisfied when they have unusual requests, emotionally charged complaints or sensitive concerns.

When every competitor offers instant AI replies, companies may compete again on something deeply human: empathy and trust.

Key terms and concepts behind the Dukaan experiment

Several ideas laden with jargon underpin Shah’s attention-grabbing decision. Some merit clarification.

Chatbot vs. human agent

A chatbot is software designed to imitate a conversation through written text or voice. Modern chatbots use large language models to produce answers, rather than selecting them from a rigid script. A human agent, by comparison, applies training and judgement and can adapt rules in ways that AI still finds difficult to replicate.

When organisations compare the two options, they generally consider:

• Cost per interaction
• Response speed
• Information accuracy
• Customer satisfaction and loyalty
• Regulatory and ethical risks

Automation risk scenarios

When a medium-sized e-commerce company follows Dukaan’s route and substitutes AI for most of its support workforce, several outcomes are possible:

• Short-term boost: Spending falls rapidly, profit margins rise and investors respond favourably.
• Mixed customer reaction: Straightforward enquiries are handled more smoothly, while edge cases and emotionally charged complaints create difficulties.
• Reputational shift: The brand develops a technology-forward identity but could be viewed as less human or less concerned.
• Policy pushback: Should large-scale redundancies become widespread, regulators and unions may press for fresh rules governing AI-led restructuring.

Risks, benefits and what may happen next

The Dukaan case demonstrates both the distinct advantages and substantial risks of forceful AI adoption.

On the benefits side, businesses can extend support availability, manage surges in demand during sales periods and lower operating costs. For rapidly moving startups, these savings could determine whether they survive or close down.

On the risk side, depending too heavily on AI creates concerns about bias, data privacy and the handling of errors. A poorly configured bot could repeat an identical mistake thousands of times before it is detected. A dissatisfied customer who cannot reach a human representative may leave permanently and then widely publicise the experience on social platforms.

Some analysts anticipate that a hybrid model will become standard. In this arrangement, AI would handle uncomplicated enquiries, while smaller teams of highly trained people would address sensitive, valuable or complex cases. Dukaan’s approach is far more extreme, but its results will contribute to boardroom discussions across the technology and retail industries.

At present, Suumit Shah argues that the figures validate his choice. For the staff made redundant, and for millions of people performing comparable jobs around the world, the meaningful outcome will be judged not by response times but by whether new, decent employment becomes available in an economy learning to live with - and at times replace - human labour through AI.