
By Kapil Sharma, Head, Presales & Solutions, NewVision Software
India has won the delivery argument. More than 2,100 Global Capability Centres now employ over 2.36 million people and generated an estimated $98.4 billion in FY26. Nearly half of GCCs in India run frontier work that matches or exceeds their headquarters.
So here is the uncomfortable part. According to a recent Dell-Zinnov study, nearly 70% of India’s GCCs are still stuck at the AI pilot stage. The same study estimates that 55% of routine GCC work is exposed to AI-driven automation.
Put those two numbers together and a hard truth appears. The scoreboard that built India’s GCCs, measured in headcount, cost per FTE and SLAs met, is the scoreboard that will hold them back.
The headcount trap
When HQ asks a GCC to innovate, the most common answer is still a hiring plan. More engineers, a new CoE, a bigger AI team.
But in a world where over half of routine work can be automated, a bigger centre isn’t a safer one. By 2030, HQ won’t ask how many people your centre has. It will ask what your centre changed.
The GCCs that will lead are already answering that question with a different metric: outcome per engineer, not engineers.
For US enterprises running global capability centers in India, this shift will decide whether the centre is seen as a cost line or a growth engine.
Why AI pilots stall inside GCCs
From what we see in GCC engagements, pilots rarely stall because the technology fails. They stall for three structural reasons:
- They run beside delivery, not inside it. A pilot in a sandbox impresses in a demo, but nobody owns taking it into the release pipeline.
- Nobody agreed on the baseline. Without a “before” number agreed with HQ, every “after” is a claim, and claims don’t earn bigger mandates.
- The incentives point the wrong way. If a centre or its partners are measured by headcount, automating work quietly reduces their own scoreboard.
In regulated industries there’s a fourth: AI output that can’t be defended doesn’t get deployed. If you can’t show how an agent’s work was checked, your risk function will, rightly, stop it.
Three shifts driving GCC transformation
- From headcount to measurable outcomes
Pick one product line. Before any AI touches it, agree the baseline with HQ: cycle time, defect escape, cost per feature. Then report the change in those numbers, not in hours or people. One proven result on a product HQ cares about does more for a centre’s mandate than ten pilots.
- AI built into how engineering is done, not run in a lab
The centres moving fastest put agents inside the delivery loop, writing tests, reviewing code, keeping documentation current, with people signing off on the decisions that matter: priorities, architecture, risk and production release. AI becomes part of how work gets done, not a separate initiative competing for attention.
- Assurance as the accelerator, not the brake
In regulated environments, the whole chain has to hold. The GCCs that scale AI fastest treat assurance as part of engineering: they test how models and agents behave, not just whether code runs, and they produce the evidence a risk team needs as a by-product of delivery. That is what turns “interesting pilot” into “approved for production.”
The question every GCC leader should ask their partners
If your GCC services or consulting partner is paid by headcount, who in that relationship is motivated to automate?
It’s a fair question, and it’s one we ask ourselves. Our view is simple: a GCC partner should accelerate the centre’s evolution, not staff it. That means agreeing the baseline before we start, letting the client verify the results, and recommending automation even when it reduces our own footprint on the account.
With vendor counts under pressure across most GCCs, the right partner should also mean fewer relationships to manage, not one more.
Why we can say this
We’ve learned this inside one of the most demanding environments there is.
For more than five years, NewVision has been an engineering partner to a Big Four GCC, where every output must stand up to audit review. We started with a handful of people on one product. Today, more than 350 of our specialists work alongside that centre, and the scope grew because results did: 30% lower release cycle time, 25% lower defect leakage and 20% lower cost per feature.
Along the way, we’ve built 100+ AI agents for audit work and 25+ in-house AI frameworks that we now bring to every GCC engagement. And as part of SoftServe, with 10,000+ engineers worldwide, we offer GCCs global depth through a single partner already working in India.
Where to start
Don’t start with a transformation programme. Start with one product line, one agreed baseline and 90 days.
If the numbers move, HQ will notice, and the next mandate will follow. If they don’t, you’ll have learned that cheaply too.
The next decade won’t belong to the GCCs that grow the fastest. It will belong to the ones that can prove what they changed.
If you’re leading a GCC and thinking about what your centre should prove next, I’d welcome the conversation.
