The roles AI is quietly creating inside GCCs

Every quarter the coverage lands on the same question: which jobs is AI taking. Inside the GCCs we recruit for, the more consequential movement runs the other way. As a function automates the routine layer of its work, it discovers it now needs someone to govern that layer. And that someone sits higher than the roles being displaced.

We’ve watched a handful of these seats appear across a dozen centres in the last year. A data-quality owner who exists because a model now depends on the input. An evaluation lead who owns whether an assisted output is good enough to ship. A governance seat that reported to no one for months, because the org chart hadn’t caught up to the fact that it needed to exist.

The pattern holds because the demand behind it is real and measured. LinkedIn’s Economic Graph reports the global economy added roughly 1.3 million AI-related jobs in two years, with India among the fastest-growing AI-hiring markets at about +40%. Most coverage reads that as engineers. A rising share of it, in the centres we see, is the layer that sits above the engineers, the people who decide what the engineers are allowed to ship.

The titles vary by centre, but the shape of the work has settled into a few recognisable seats. Naming them matters, because a role you can’t name is a role you can’t budget, scope, or search for.

The AI product owner comes first. Once a team has models in a workflow, someone has to decide what the AI is for: which decisions it assists, where a human stays in the loop, what “good” looks like for the business rather than the benchmark. This is a product judgment about consequence, not a modelling judgment about accuracy, and it’s usually the first seat a maturing AI programme finds it’s missing.

The evaluation and quality lead follows close behind. When output is generated rather than written, whether it’s fit to ship becomes a discipline of its own: test sets, drift monitoring, the cost of a wrong answer at scale. Someone owns that or no one does, and in the centres where no one does, the failure surfaces in production.

The ML-platform owner is the seat that makes the rest repeatable. As models multiply across a centre, the pipelines, tooling, deployment and monitoring that support them become a platform, and platforms need a dedicated owner accountable for their reliability. And the model-risk and governance seat is the one that decides, across all of it, what carries acceptable risk and what doesn’t. Four seats, one underlying shift: the function has moved from producing outputs to governing a system that produces them.

When a team moves from executing a task to overseeing a system that executes it, the skill that matters changes underneath the title. The producer optimises for output. The person governing the system optimises for where the output can be trusted, where it must be checked, and who carries the consequence when it’s wrong. Those are rarely the same person, and they rarely come from the same shortlist.

This is why so many of these seats stay open. Centres try to promote the strongest producer into the governance role and find the instinct doesn’t transfer, the best engineer often makes an uneasy governor, because the job is to say no to work they’d have been proud to ship. The candidate who fits has usually held accountability for a system at scale before, somewhere the failure was theirs to own. Knowing where to find that person is most of the search.

The market compounds the problem. Deloitte and Nasscom estimate India’s AI-talent demand will exceed 1.25 million by 2027, against a shortfall of roughly 50% in 2024, with only around 16% of IT professionals AI-skilled. The scarce layer within that scarce pool, people senior enough to govern, fluent enough to be credible, is a fraction of a fraction, and it’s the fraction these seats compete for.

The clearest instance we see is inside BFSI centres, where the governance seat has a name the industry is only now settling on: an ‘AI risk’ layer. As these GCCs move AI into workflows that touch decisions a regulator cares about, a gap opens that the existing structure doesn’t cover. The model-risk team understands models, but not always the way this generation of them behaves. Compliance understands the regulator, but not the internals of the system. Between them sits a decision no one currently owns: is this AI use defensible, and who says so.

That gap is becoming a seat, a person accountable for how AI is used against the standard a regulator will hold the firm to. It appears before there’s a budget line for it, because the need shows up faster than the annual plan can. And it’s hard to fill for the reason all these seats are hard to fill, only sharper here. It demands model fluency deep enough to challenge the system on its own terms, and regulatory scar tissue deep enough to know what will survive scrutiny. Hire for one half and the seat fails predictably: a pure compliance profile can’t interrogate the model, and a pure technical profile carries no weight with the regulator or the board.

Where it reports is as much of the design as who fills it. Place it under the team building the models and its independence is gone the moment it’s judging its colleagues’ work. Bury it deep in a control function and it’s out of the loop until the decisions are already made. The centres that get this right give the role real proximity and real independence at once, close enough to see the work early, structurally separate enough to say no. That reporting line is the part that’s easiest to get wrong under time pressure, and the part that’s most expensive to discover you got wrong in an audit.

The centres getting ahead of this read where their own automation is heading, name the governance seat before the gap becomes painful, and hire into it while there’s still room to onboard properly rather than under fire. A governance hire made in calm is a different, better hire than the same one made 3 weeks after an incident.

The diagnostic is simple to run. If your AI programme has matured over the last 18 months and your senior structure looks identical to where it started, that flat line is the signal, the work has changed and the org chart hasn’t. Ask which of the four seats your current function is already leaning on someone to informally cover, and you’ll usually find one person holding a governance responsibility that was never scoped, budgeted, or given the standing to do the job.

That’s the seat to name and fill next. The org chart lags the work by design; the centres that close the lag early are the ones that don’t stall at the point where governance was supposed to already exist. If you want help mapping where your function is heading and which seat it needs first, that mapping is exactly what we do, start a conversation with Recruise.