Twenty years ago, building a research centre in India was a leap of faith. The executives who sponsored the first captives – the India centres now often called Global Capability Centres, or centres of excellence – were signing off on operations in cities most had never visited, against scepticism from desks that doubted the work could be done well at distance, on the strength of a cost case and a handful of believers.
They were right. What began as an experiment is now a core strategy for many of the world's largest markets businesses, and the heads who built those centres run operations of a scale and quality the original business case never dared promise.
The models and infrastructure that analytical captives grew out of – software and middle office – will need to be left even further behind in future. For the next phase of the captive's growth, talent will need to be leveraged even further; and onshore colleagues will need support in how to integrate higher-end talent. The point underneath both is the same, and worth saying plainly: deeper penetration into onshore – more desks served at headquarters, on harder questions, and a centre that matters more to the people there as a result.
The next phase: captive growth in a world of AI
A captive grows by taking on work for onshore desks – first more of it, then more kinds of it. The next phase of that growth is about which work comes next: new desks served at headquarters, new products covered, more of the work senior stakeholders actually fight over – in a world where AI is steadily absorbing the work the centre grew up on. Two paths lead there.
Path one: higher judgment work
The first path continues the progress of the captive's centre of gravity from process work to judgment work.
Judgment is what happens when the facts do not settle the question. Two intelligent analysts can look at the same numbers, the same filings, the same price action – and reach different conclusions, both defensible. The credit view that disagrees with the market, the call on what a client needs to hear today, the recommendation that follows from the evidence but is not compelled by it – none of these are read off the inputs. Someone forms them, and owns them.
That is precisely the work AI cannot do. A model produces the conclusion most likely to follow from what it has seen – the consensus continuation. Where the facts genuinely underdetermine the answer, there is no consensus to retrieve; there is only a view. And here is the interesting consequence for a captive: judgment work asks more of headquarters, not less. A desk has to extend faith to an analyst whose conclusion cannot be checked against a template. That faith is harder to win than a service-level agreement – and it is exactly what makes the work valuable, because it is the difference between output that gets consumed and a view that gets relied on.
A captive that builds a defensible mass of this work – not as a side activity but as the visible centre of what it does – is serving desks it never served before, on work head office cannot easily ask a machine or a smaller team to do. That means a different headcount mix: more senior analysts, fewer juniors; more writing, less data plumbing; more cross-product roles, fewer single-product seats.
Path two: deep onshore integration
The second path is the one captives most often underrate – partly because it looks like soft work, and partly because almost nobody has seen it done properly. Done properly, it is earned practice: cultural understanding running in both directions, framing, permissions, and the patient conversion of sceptics.
The mistake is treating integration as one-sided: a programme that coaches the offshore team to communicate better and presents the result to headquarters as a finished article. Integration that works is run on both sides, under one coherent strategy. Onshore desks need framing, permissions and reasons to engage, just as offshore analysts need rotations and named relationships. And some onshore stakeholders are simply not comfortable working with an offshore centre at all. That is normal, and it is not a moral failing on either side; a portion of those people can be brought around – but only if both ends of the relationship are being managed deliberately, by someone with the standing to coach a managing director in London and an analyst in Mumbai in the same week.
Integration is also not a function of headcount. A 200-person centre with deep onshore embedding will serve more desks, and matter more to them, than a 1,000-person centre without it.
Integration is what gets judgment work commissioned in the first place – no desk hands its hardest questions to a centre it half-trusts. And integration multiplies the value of everything that travels across it, including well-run routine work: a properly integrated centre doing disciplined process work, with the right direct relationships behind it, is already hard to dislodge.
The how: bringing in the senior analyst layer
Building the structures that nurture judgement and the talent that delivers judgement may require a change in culture at the captive, including across different management layers. Integration around this certainly requires peer-to-peer coaching, explanation, context, and experience sharing. Few centres have that coaching bandwidth spare internally, which is exactly where an outside partner earns its place. The models for doing it – and how to choose between them – are set out in a companion piece: Accelerating India Captives.
When a captive should not move this way
Not every centre should pursue both paths, and not every workflow should be climbed out of. Centres whose mandate is deliberately transactional may rationally choose efficiency over expansion, and some workflows will remain process-heavy by design – there is no virtue in forcing judgment into work where judgment adds little. The argument here is for centres whose stakeholders are already asking for more than the current model gives them – which, in our experience, is most of them.
Why now
None of this is abstract. As of the second quarter of 2026: HSBC is weighing around 20,000 job cuts over three to five years, with non-client-facing roles in global service centres explicitly in scope. Standard Chartered has announced 7,800 reductions by 2030 and named Bengaluru, Shenzhen and Warsaw among the operational centres affected. Citi has pledged 20,000 reductions and is paying several of the major AI labs to automate swathes of middle- and back-office work, and Bloomberg Intelligence has forecast that global banks may shed up to 200,000 roles over the same window. And the speed of the shift is the real message: when one CEO could describe redundant offshore roles as "lower-value human capital" – roles in centres his own institution had spent two decades building – and the line had to be walked back internally within days, every operating head learned the same two things. Head office is now willing to say the quiet part publicly, and to act on it.
Yet the more important fact for anyone planning a centre's next five years is what has not happened. Across large organisations everywhere – not captives specifically – there is a gap between the AI adoption shown on consumption dashboards and the reality on the floor: a few people who have genuinely absorbed AI into their work, a larger group who use the sanctioned tools when prompted, and a substantial group who log the mandated hours and otherwise work the way they did in 2022. That is not a failing of any one centre; it is what every large rollout measured by consumption looks like. But it means the real transition has not happened yet – and the gap is the time a centre still has to evolve on its own terms: to choose which work to cede, which to climb into, and how to tell that story to head office before head office writes it.
What the centre looks like in five years
A captive that has taken both paths seriously will, by 2031, look something like this.
Its headcount mix will have shifted measurably away from process roles – some displaced by AI, some migrated to lower-cost locations, some absorbed back onshore. It may well be smaller in seats; it will certainly be more senior, and it will matter more. Crucially, that evolution will have been led and narrated by the centre itself, not imposed on it.
A meaningful proportion of analyst time – perhaps 10 to 20 per cent – will be spent in onshore offices on rotation or structured visits, funded, scheduled and reviewed at senior level rather than treated as a perk. Named analysts will be tagged to named senior stakeholders.
Stakeholder maps will exist as living documents, reviewed when an analyst moves or leaves, the way client coverage is reviewed when a banker departs. Integration will be auditable, not anecdotal.
AI usage will be measured by output transformation rather than licence consumption. The centre will hold its own honest view of where AI has genuinely changed the workflow and where it has not – and will tell head office both. The centre that is straight about adoption is the centre head office trusts on everything else.
And it will have an external partner or two providing senior-judgment capacity and seeded capabilities, contracted in daylight rather than quietly – because by then the partnership reads as evidence of the centre's reach: desks served at headquarters in 2031 that were not served in 2026.
None of this is radical. All of it is hard. The difference between the centres that get there and those that do not is whether they started while the window was still open.
The second act
The captives were one of the great operational bets of the last twenty years, and the people who made it – in Bengaluru and in head office alike – were early and right.
What that bet found, and what no review or restructuring takes away, is the talent. India holds one of the deepest pools of financial-analytical talent in the world, and global markets need it now as much as they ever have – arguably more, as the judgment layer becomes the part of the work the machines reach last. That is the constant. What changes – what has always changed – is how that talent is found, developed, deployed and integrated. The first act answered those questions for a world of cost and scale. The second act answers them for a world where the routine is absorbed by machines and the value sits in judgment, and in the relationships that carry it.
This is not disruption arriving from outside to be survived. It is the ordinary evolution of an industry doing what industries have always done – and the only real choice is whether to meet that history with your eyes open, and decide what the second act looks like, or let the next operating model decide it for you.
Frontline has spent twenty-one years running senior offshore analyst teams from India for global banks and asset managers – through their centres, not around them. If you're running a centre heading into its next planning cycle, or sponsoring one from half a world away, and want a senior conversation about the model – what the judgment layer might look like, what onshore integration would actually require, how external capacity might or might not fit – that's a conversation we're well placed to have.
Read next: Accelerating India Captives – how Frontline helps build the senior layer
Part of: Offshore Analyst Teams – what works and what breaks
Talk to us: a senior conversation about your centre's next phase