HCLTech and AI: the claim that's still waiting for its numbers

Late 2022: the moment HCLTech saw the model break

The September-October 2026 issue of Harvard Business Review carries a first-person piece by C. Vijayakumar, chief executive of HCLTech. The subject is the AI pivot at the Indian IT services group. The story starts on a specific date: the end of 2022.

In those months generative AI became accessible to anyone. HCLTech, its chief writes, grasped quickly what the change meant for the company and for the entire sector (Harvard Business Review, September-October 2026^[1]).

The central point concerns the economics of the trade. For decades IT services grew in a straight line: more revenue meant more people. Every technology wave, from the internet to digitalisation to cloud, called for more engineers, more consultants, wider teams.

AI, Vijayakumar argues, breaks that link: large volumes of knowledge work now get finished far faster and with far less human effort.

The original insight: reading AI as an economic fact

The interesting decision comes before the technology. In 2022 almost everyone looked at generative models as a productivity tool: help with writing code, copy, replies.

HCLTech chose a different and more uncomfortable reading. Its chief puts it in writing: the technology goes beyond improvement at the margins, because it changes the economics of the business itself.

The difference carries weight. Those who treat AI as a tool buy licences, train their people and measure hours saved. Those who treat it as an economic fact reopen pricing, contracts, the skills mix and the shape of the income statement.

This is the position this desk repeats case after case: the original architectural decision counts for more than the technology chosen. For an IT services provider, the pricing model is pure architecture.

What the text states, and what it leaves out

Here comes the friction. It has to be said plainly. The publicly visible portion of the piece carries the thesis, the date and the logic; then it stops.

The before and after is missing. The part readable in the clear has no revenue per employee, no headcount trend, no share of outcome-based contracts, no margin per project. The horizon is missing too: how many quarters it will take for the new model to show up in the accounts.

That changes the label to use. Testimony from the top on a strategic choice is worth a great deal to anyone making decisions. It nonetheless remains a claim, a long way from a measured case study.

The distinction serves the reader in a practical way. A case study offers a number to compare against your own. A claim offers a direction to verify in the next set of results.

Why the missing number really matters

ROI is measured with a before and an after on a metric that has a denominator. It holds for everyone. A ten-person web agency and a group with tens of billions in revenue use the same yardstick.

In IT services the right denominator already exists. Anyone who reads a sector balance sheet knows it: revenue per employee. When AI breaks the linear link, that ratio climbs visibly, quarter after quarter.

That is why asking for the number is an act of respect towards the story, rather than scepticism. Vijayakumar's thesis becomes verifiable in public, in the financial documents his company files every three months.

A statement like this, put in writing in a publication such as Harvard Business Review, creates a commitment. Whoever signs it accepts that the market will measure it.

The vendor register: a comparison that clarifies

Alongside the CEO's testimony sits another kind of document, useful and to be read with the same care: the customer page published by the technology vendor.

Google Cloud, for instance, devotes a page to the German private bank Berenberg (cloud.google.com/customers/berenberg^[2]). Material of this kind describes real architectures and named organisations: it remains partisan material, because whoever publishes it has a stake in the outcome.

The working rule turns out to be simple. A story becomes solid when the number comes from a source independent of both seller and buyer. A regulatory filing counts, so does an earnings call transcript, an auditor, a control group.

The rest should be treated for what it is: an indication of direction, sometimes valuable, still to be confirmed.

In the most solid stories the number passes through an auditor, an A/B test or a comparison against a control group.

How to read a strategic statement from the top

A text of this kind still has plenty to offer, provided you know what to look for in it. The value sits in the logic laid out, rather than in the absent figures.

Four questions help squeeze it dry:

• Which economic link does the company declare broken, and in what year did it work that out
• Which internal choice did it make as a result: pricing, skills, contracts, structure
• Which public metric will make the result visible, and over what horizon
• Which correction has it already admitted along the way

On the fourth question the piece in the clear is silent. This remains the most serious limitation. Every verified success story contains a correction: Klarna brought human agents back, Morgan Stanley redesigned the workflow.

Anyone showing a line that only ever rises is doing communications. The moment of friction, by contrast, is the most informative thing an organisation can share, because it shows where the plan met reality.

A willingness to adjust course signals operational maturity, rather than failure.

One revision admitted in public is worth more than ten slides with a perfect curve.

What we can take away

The replicable part of the HCLTech story lives in the timing of the reading, rather than in the size of the company. At the end of 2022 millions of managers had the same technology in front of them. Few rewrote their revenue model.

For an SME founder the playbook stays light: ask which part of your price is tied to hours sold, and how well that part holds up with AI in the building.

For a CTO the work consists of measuring output, continuously. The candidates: throughput per person, delivery time, quality after release. Numbers collected from day one turn an opinion into a time series.

For a board the bar moves here: suppliers' revenue per employee becomes a question to put on the record, quarter by quarter. For a team lead what counts is the local translation of the thesis. Which activity in my workflow still earns the time I give it?

The open question

HCLTech's thesis sounds clear and can be checked in public: revenue growth detached from headcount growth, starting from the reading made in late 2022. The next set of results will say how much of the claim has become practice.

The question that remains applies to any organisation, of any size. Which linear link still holds up your income statement, and which public metric would show it broken within twelve months?