Who captures the gains

Three pockets the surplus can land in

When AI is dropped into a job and the job gets faster, the gain has to go somewhere. There are essentially three pockets.

The first pocket is the worker. The hour you no longer spend on the easy emails could become an hour you spend doing nothing — a shorter day, with the same pay. Or it could become an hour you spend on harder work that you are now compensated for at a higher rate. Either way, the worker captures the gain. This is what most of the early twentieth-century productivity revolutions eventually did in countries with strong unions and tight labour markets: shorter weeks, higher wages, better conditions.

The second pocket is the owner — the proprietor of the BPO, the founder of the agency, the shareholder of the firm. The hour you no longer spend on the easy emails becomes an hour the firm bills the client for at the same rate, while paying you the same wage. The margin widens. The owner captures the gain. This is what the early decades of most productivity revolutions actually did before workers organised, and it is the default outcome when bargaining power is weak.

The third pocket is new, and it is the one that matters most for Nepal: the platform rentier. A foreign vendor sells the AI tool that made the work faster, and charges a per-seat or per-task fee that captures a large slice of the productivity gain before either worker or owner sees it. The hour you saved is the hour the platform priced. The platform sits in California or Bangalore; the surplus sits in their P&L. Neither labour nor capital in Nepal touches it.

A Nepal-shaped example: two BPOs, same tool

Consider two BPO firms in Bhainsepati, on the same street, both serving foreign clients with first-line customer-support work. Both adopt the same AI-assistant tool that lets a single agent handle roughly twice as many tickets per shift as before.

The first firm is Nepali-owned, runs its own back-office, and pays a flat monthly licence for the AI tool. Its foreign client still pays on a per-ticket basis at the old rate. Suddenly the firm has nearly double the revenue per agent-hour. The owner has a choice: raise wages to retain staff in a tight Lalitpur labour market, or pocket the difference. Suppose the local labour market is competitive enough that the firm raises wages by twenty per cent and still keeps most of the margin. The surplus is split between worker and owner, and most of it stays in Nepal.

The second firm is a subsidiary of a foreign company, whose foreign client now requires every ticket to flow through the client’s own AI platform. The client renegotiates the contract to pay per-ticket at half the old rate — “because the AI is doing most of the work now” — and charges the BPO a per-seat fee for the platform on top. The agent is now handling twice the tickets for the same wage, the BPO’s margin has thinned, and a meaningful slice of the surplus is being booked as platform revenue in another country. Same technology. Same productivity gain. Almost none of the surplus reaches the worker, and not much of it stays in Nepal.

The technology is identical. The outcomes are opposite. The difference is bargaining power, contract structure, and ownership — the entirely political and economic furniture around the model.

The framing trap, and why it matters

The most common mistake in this conversation is to describe outcomes as if the technology caused them. AI is reducing wages in the BPO sector. That sentence sounds like a description of physics. It is not. The accurate sentence is: a particular contract renegotiation, made possible by a productivity gain that AI enabled and made visible to the buyer, is reducing wages in the BPO sector. The renegotiation is the cause; the AI is the enabling condition.

This distinction matters because it changes who you talk to and what you do. If the technology is the cause, the only response is to slow the technology, which no Nepali policymaker can actually do. If the renegotiation is the cause, the responses are recognisable labour-policy responses: collective bargaining for sectoral workers, minimum per-task floors, contract-review requirements for foreign clients of Nepali firms, tax treatment of foreign platform fees, and competition policy that prevents a single foreign vendor from becoming the only choice.

What “policy” actually means here

When the surplus question is on the table, “policy” is not an abstract category. It is a small set of specific levers that the Government of Nepal, sectoral associations, and worker organisations can actually pull.

Tax treatment of foreign-platform fees decides whether per-seat payments to California vendors are a deductible expense at face value, or whether a withholding tax claws a portion back into Nepal’s revenue. Competition policy decides whether a single foreign AI platform can become the only one BPO firms can buy, or whether multiple vendors are encouraged in. Sectoral wage floors — explicit or negotiated — decide whether productivity gains can legally be captured entirely by the owner without any portion reaching the worker. Disclosure requirements for foreign client contracts decide whether a BPO worker can know that the per-ticket rate has been halved because the client now uses AI for the first draft. Each of these is a normal policy tool. None of them are about AI specifically. All of them shape who captures the gain.

The point is not that any particular policy is correct. The point is that the question who captures the gain? is answerable, and the answer is decided in rooms that include ministries, sectoral bodies, and worker representatives — not in rooms that include only engineers.

Check your understanding

Quick check

—

An AI tool is introduced into a Lalitpur BPO and doubles the number of tickets each agent can handle per shift. Six months later, agent wages have not risen, the foreign client has cut the per-ticket rate, and the foreign AI vendor is collecting a per-seat fee. What is the most accurate way to describe this outcome?

1. AThe technology caused the wage stagnation — this is what AI does to labour, by its nature.
2. BIt was bad luck — the firm happened to be unlucky with its client.
3. CThe productivity gain was real, but who captured it was decided by the bargaining power of the worker, the contract terms with the foreign client, and the absence of regulation on platform fees — not by the technology. The technology enabled the outcome; politics and contracts decided it.
4. DThe agents must be underperforming and that is why wages did not rise.

Check answer→

What comes next

The surplus question gets sharper when we widen the lens beyond the BPO floor. Nepal’s labour market does not end at the Ring Road — it extends to the Gulf, Malaysia, and Korea through migration, and it extends down into the gig economy through Pathao, inDriver, and Foodmandu. Both of those channels carry their own AI exposure. The next section, Remittances, gig work, and AI, looks at what happens when automation reshapes destination labour markets and when algorithmic management quietly reshapes earnings on the rider’s phone at home.