Systems

The Price of Saved Time

October 2026
9 min

Faster work does not automatically produce greater profit or lower prices. Before anyone benefits, organisations must decide what the saved time is worth and to whom it belongs.

India’s largest technology companies are preparing to report what analysts expect will be their weakest sequential performance in three years.

The industry is not small or peripheral. It produces around $315 billion in annual revenue and employs nearly six million people. Its companies maintain banks, operate corporate systems, build software, manage infrastructure and perform much of the ordinary technical work on which organisations around the world depend. Yet the six largest providers are expected to report quarterly revenue growth of only 0.7% to 3.5%, while India’s main technology index has fallen by about 27% this year.

Part of the explanation is familiar. Clients are cautious. Energy and borrowing costs are high. Large projects are being delayed. But another pressure is now appearing in contracts: customers expect artificial intelligence to make the work cheaper.

This is usually described as AI-led deflation. The phrase makes the process sound almost automatic, as if better software simply causes prices to fall. What is actually happening is a negotiation over who owns the value of work that no longer takes as long as it once did.

For decades, much of the IT-services industry operated through a straightforward commercial unit: human time. A customer bought the effort of a team. Ten engineers working for six months generated more billable capacity than five engineers working for three. Large providers benefited from being able to recruit, train and deploy thousands of people across standardised projects.

AI disturbs that relationship. If a task that required ten people can now be completed by six, the provider has become more productive. But if the customer is paying for people or hours, the provider has also reduced the quantity of its own product.

Efficiency becomes a commercial problem.

Productivity has no natural owner

It is tempting to assume that whoever introduces a useful technology captures its benefits. Economic life is rarely that tidy.

A company can invest in an AI coding tool, train its staff and redesign its delivery process, only for its customer to demand the same project for 25% less. Persistent Systems’ chief executive told Reuters that clients were already requesting the same work for 25% to 30% less while also expecting faster delivery. Some work is disappearing altogether as customers bring AI-assisted tasks back inside their own organisations.

The technology provider bears the cost of adopting the tool. The client may capture the saving.

This is not necessarily unfair. If a supplier can produce the same result with substantially less effort, a competitive market should eventually reflect that. Nor should customers continue paying for labour that is no longer required simply to preserve a provider’s old revenue model.

But there is no rule inside an AI system determining how the benefit should be divided. That decision is made through bargaining power, competition and contractual design.

The same issue is beginning to appear in legal work. Thomson Reuters reported this year that 41% of law firms and 47% of corporate legal departments were using generative AI, up substantially from 2025. Its earlier research estimated that AI tools could save lawyers almost 240 hours a year through faster document review, research, summarisation and drafting.

Those hours could become lower bills for clients. They could allow lawyers to handle more matters. They could increase the firm’s profit. They could produce shorter working days, although commercial history gives us some reason to be cautious about that possibility.

The tool does not choose among these outcomes. The organisation’s business model does.

A company selling a fixed-price service has an incentive to become more efficient because it keeps some of the saving. A company billing by the hour can be penalised for the same improvement. A salaried employee may save four hours a week only to receive four hours of additional work. A consumer may see no price reduction at all if competition is weak.

Productivity is often discussed as if it were a quantity that spreads naturally through an economy. In practice, it travels through institutions. Contracts, wages, prices and market power determine where it stops.

When the hour no longer works

IT-services companies are responding by changing what they sell.

Rather than charging for a team’s time, more contracts are tying payment to measurable results: the number of transactions completed, incidents resolved, systems migrated, costs removed or service levels achieved. TCS’s chief executive told Reuters that around 80% of contracts in its finance, human-resources and other business-services segment now use outcome-performance measures. In one reported arrangement between Cognizant and Daimler Truck, AI-related cost savings were to be divided between the supplier and the customer.

This seems more rational than charging for effort. Customers do not ultimately want 20,000 engineering hours. They want a functioning payment system, fewer service failures or a process that costs less to operate.

But selling an outcome is harder than selling time.

An hour can be counted. An outcome must be defined.

Suppose a supplier is paid partly according to how many service-desk incidents it prevents. A fall in incidents might mean the supplier improved the underlying system. It might mean employees stopped reporting problems. It might reflect a quieter business period, a change made by another vendor or the removal of a troublesome product. A rise in incidents might indicate deteriorating service, or it might mean the organisation has grown.

The measure becomes the place where commercial interests meet operational reality.

TCS itself identifies some of the difficulties surrounding outcome-based service-desk pricing: missing historical baselines, dependencies on third parties, uncertain demand and variations in the complexity of individual cases. It proposes using AI to forecast volumes, classify complexity, monitor service levels and support real-time billing.

That is revealing. AI helps create the productivity gain, but it may also be needed to administer the new commercial system created by that gain.

The customer and supplier must agree what the world would have looked like without the intervention. They must decide which changes the supplier controls, how quality will be protected and what happens when an external event makes the promised outcome more expensive. They also need evidence strong enough to support payment.

A contract based on hours purchases labour. A contract based on outcomes asks the supplier to underwrite part of reality.

This transfers risk. If automation performs better than expected, the provider may earn a healthy return. If volumes rise, chip costs increase or a client’s legacy systems prove harder to change than assumed, the provider can lose money. Tech Mahindra’s chief executive recently criticised competitors for guaranteeing prices based on predicted productivity improvements of 70% to 80% over periods as long as seven years. Infosys has said it walked away from contracts that were no longer economically viable.

The new model therefore aligns incentives only when the outcome can be measured fairly and the relevant risks can be allocated sensibly. Otherwise, it replaces one imperfect proxy, time with another.

The measure becomes part of the service

Once payment depends on an outcome, the definition of success stops being an internal management detail.

Consider a call centre paid according to average handling time. Calls may become shorter without customers receiving better answers. A cybersecurity provider rewarded for closing vulnerabilities may prioritise easily resolved findings over the obscure dependency capable of causing the greatest damage. A hospital contractor measured by processing speed may move people through a system faster while leaving more work for the next part of the service.

None of this is an argument for retaining hourly billing. Hours can reward delay, unnecessary complexity and oversized teams. But an outcome measure can be gamed too, especially when it captures only the visible part of the service.

This is why the change in pricing is also a change in governance. Clients need to know not merely whether a target was reached, but how it was reached and what happened outside the target’s field of view. Quality measures, audit rights and safeguards against deferred work become part of the commercial architecture.

The consequences will reach employees as well.

The Indian technology-services model developed around a workforce pyramid: large cohorts of graduates performed more standardised work while smaller numbers of experienced staff handled architecture, client relationships and complex decisions. Automation is strongest precisely where work is repetitive enough to be described and checked.

That may reduce the demand for some entry-level work. It can also damage the way expertise is created. Senior engineers do not arrive fully formed; they become senior by spending years doing simpler work, encountering failures and learning how systems behave.

There is not yet evidence for a simple story of employment collapse. NASSCOM estimated that direct technology employment would reach approximately six million in the 2026 financial year, an increase of about 135,000 people.But companies have warned that their traditional role as very large recruiters of new graduates may be diminishing, and Reuters reports that TCS cut more than 12,000 positions last year.

The likely change is more specific than “AI takes the jobs.” Providers will need fewer people for some units of delivery, while placing greater value on those who can define outcomes, supervise automated work, understand client operations and intervene when the model encounters an exception.

That creates opportunity, but it also narrows the first rung of the ladder. An industry can need sophisticated people while weakening the ordinary route through which people become sophisticated.

The negotiation behind the numbers

The earnings reports due from India’s technology companies will be read as evidence about demand, margins and investor confidence. They are also evidence of a deeper renegotiation.

AI can save time, but saved time is not yet an economic benefit. Someone must convert it into a lower price, additional work, higher profit, better service or shorter working hours. Each option favours different people.

Clients currently possess considerable leverage. They know providers are competing fiercely and are using the promise of AI efficiency to demand discounts before the long-term economics are fully understood. Providers are trying to move toward performance pricing, build reusable platforms and find new work quickly enough to replace the revenue that automation removes.

Neither side has discovered a settled model. Outcome-based pricing may produce healthier incentives where success is genuinely measurable. Elsewhere, hybrids will probably persist because the outcome depends on too many participants, too much uncertainty or too much professional judgement to assign cleanly to one supplier.

The broader lesson is not confined to outsourcing. Many organisations are about to discover that adopting AI and capturing value from AI are different acts.

A faster employee does not automatically receive a higher wage. A more productive supplier does not automatically earn a higher margin. A cheaper process does not automatically produce a cheaper service. Between the technical improvement and the economic benefit sits a system of prices, contracts and power.

That is what India’s subdued technology results are beginning to show. The industry is not simply being asked to do its existing work with better tools. It is being forced to renegotiate what the work consists of, how its value is demonstrated and who can claim the hours that no longer need to be worked.

The time may have been saved.

Its destination remains unsettled.

References