Last week, Infosys brought clients and analysts to Geneva for its 25th Confluence event in EMEA. We expected the usual AI promises. Instead, Salil Parekh opened the event with a remarkably down-to-earth message: Infosys is a delivery organization, its foundation is client trust, and it intends to remain a stable partner in a world that is changing quickly. There was no grand talk of the autonomous enterprise being around the corner. In the current AI climate, we found that refreshing. It also says something about how Infosys is choosing to navigate a difficult moment for the whole IT services industry.

The industry is repricing, and not everyone is pricing responsibly

IT services firms are going through a repricing that we think has another 12–24 months to run and may be close to its peak. Clients at every stage of AI maturity are asking their providers for productivity gains, and those gains are working their way into contracts as they come up for renewal. Some providers appear to be responding by bidding below what they can realistically deliver, pricing in productivity that their own delivery has not yet achieved. That may win accounts in the short term, but it is much less clear what it does to service quality and to client relationships two or three years into those contracts.

Infosys seems to have chosen a different path. On its last earnings call, it said that it had declined contracts whose economics did not work. That discipline comes amid subdued growth. Constant-currency revenue grew 2.4% year on year in the quarter to June, and Infosys narrowed its full-year guidance to between 1.5–3% while holding its operating margin at around 21%. We think this is a sensible bet, though not a certain one. The repricing will pass. When it does, and when fast growing AI-led work becomes a much larger share of what clients buy, a provider that kept its clients’ trust through this period is likely to be better placed than one that won share with promises it struggled to keep.

There is also a values dimension here that we don’t want to lose. The fact that a technology allows you to do something is not, on its own, a reason to do it. Staying grounded in respect and trust is a choice, and it has consequences for clients, for employees, and for society.

A major European bank shows both sides of the repricing in one account

A major European bank, a long-standing Infosys client, offered a clear illustration of what is at stake. The bank announced its strategy in June 2023, just as generative AI was emerging, and soon afterward chose Infosys as a strategic partner, with Infosys agreeing to take over the bank’s IT center in India. On stage, the COO explained that the bank first raised its expected productivity gain in technology development to 25% by 2028. It now expects to reach 40% by the end of this year. That is the repricing seen from the client’s side. Under an effort-based contract, these productivity gains put downward pressure on the cost of delivering the work. Whether the total contract value grows or falls depends on how much the bank reinvests the savings in new work.

The same conversation showed how AI can also create demand for new work. For instance, the COO said that he had never believed in a mainframe or core banking transformation. Today, working with Infosys and AWS, the bank is translating its mainframe applications into natural language and, from there, into new platforms, application by application. The COO now sees it through a time period of perhaps over 3–5 years. That is substantial new work that AI has made viable and feasible. The bank also runs an AI innovation lab with Infosys, built around “proof of value” rather than proof of concept so that ideas prove their worth at a small scale before they move into normal processes.

Which side proves larger on this account is not yet clear, and it may be a few years before we know. What is clear is where the bank sees its constraints: completing risk assessments for AI solutions, getting the right data in place, and bringing tens of thousands of people along the AI voyage. AI can accelerate the technical work, but it does not remove the organizational complexity. A trusted partner adds value by helping the bank navigate that complexity and carry the transformation forward.

Infosys is making delivery fit for AI

IT service delivery used to depend almost entirely on people. Infosys is redesigning it around people, proprietary software (IP), and tokens, the units of AI model usage it pays for to get work done. Its workforce is shifting toward specialists who combine domain knowledge with engineering skills, with frontier engineers at the core of the new model. Infosys has trained people at scale before. But scaling this model requires more than training. We see three constraints on how quickly it can deliver:

  • The first is securing expertise that training alone cannot supply. The experts at the center of frontier teams (aka PODs, for product-oriented delivery teams) need AI skills, industry knowledge, and judgment, a combination that cannot be developed in a few months. Infosys must compete for these people with other providers, AI labs, and large enterprises.
  • The second is making the right technology bets. Value is moving quickly from the models to the harnesses — the tooling that makes a model work in a specific business scenario. Infosys’ answer is Infosys Topaz Fabric, an open stack on which clients can build and run their own agents, using whichever agent framework they choose. But what AI researchers call the “bitter lesson” applies here, too: A harness designed around the limits of today’s models may be overtaken by the next one. That puts a premium on agility (or the ability to rebuild harnesses quickly) and on a steady focus on the value they create inside the client’s business rather than on the harness itself.
  • The third is the client’s own pace of change. A provider can redesign its delivery faster than a client can change its organization. As the conversation with the bank showed, the hard constraints sit inside the client.

What CIOs should take from this

It is too early to know whether Infosys’ stance will pay off. A few signals would suggest that it is working: The share of revenue it counts as AI-first is moving toward 25–30% (its most recent quarter put the number at 8%) while its margin holds; accounts it walked away from are returning after lower-priced winners struggle; and organic growth is recovering faster than its peers once the repricing has run its course.

For CIOs, the more immediate lesson is about how to read the bids in front of them. A bid far below what the incumbent believes the work requires may deserve a second look, because someone will eventually pay for the gap between the price and the work, often through service quality and difficult renegotiations. When a provider walks away, its reasons can say something useful about what the work really costs. Ask each bidder to show the productivity gains that it has already demonstrated, which remain assumptions, and how service quality will be protected if those assumptions fail.

More broadly, as argued after Accenture’s analyst summit earlier this year, the AI productivity paradox is an organizational design problem. The enterprises that get through this period well will be the ones that pair their AI ambitions with partners they can trust to tell them what is realistic — and to stay for the harder organizational reinvention work that follows.

Share