EASE 9.0 · Innovation · Global Capability Centres & Talent

Built Once, or Bought Four Times

Building in-house AI capability through GCC hubs.
Agent components built once and reused across every function, instead of restarting with each vendor.

Most PSU banks buy AI capability from external vendors, one function at a time: a fraud-scoring vendor for fraud, a chatbot vendor for customer service, a credit-scoring vendor for underwriting. Each engagement rebuilds foundational infrastructure the bank already paid for in the last one.

Why In-House Capability Compounds

A Global Capability Centre (GCC) approach inverts vendor-by-vendor buying: the bank builds AI agent components once, in-house, and reuses them across functions. A policy-encoding framework built for credit decisioning is the same framework a fraud team can adapt for transaction monitoring.

This is not a hypothetical for the PSU peer set. In January 2026, a leading PSU bank inaugurated its own Global Capability Centre in Bengaluru, the first PSU bank to do so, and signed an MoU with the Karnataka Digital Economy Mission to also provide banking and relationship-management services to the wider GCC ecosystem in the state. Industry expectation is that other public-sector banks will follow with their own GCC strategy by FY27. The precedent now exists inside the PSU peer set; the open question for every other PSU bank is not whether to build one, but what kind — specifically, what an AI-era GCC needs that a 2015-era GCC did not.

The Path to Progress

Capability that lives with a vendor rebuilds foundational infrastructure the bank already paid for, every single engagement.

The Business Case

The macro backdrop is not in question: Zinnov and Nasscom's FY2026 report counts 2,117 GCCs operating in India, employing 2.36 million people and generating $98.4 billion in annual revenue, up 32% since FY2021. EY's 2025 GCC Pulse survey finds 58% of GCCs already investing specifically in agentic AI and 83% scaling generative AI more broadly, and the operating model EY flags as winning pairs humans and autonomous agents in one coordinated layer, rather than treating automation as a separate department bolted onto the existing org chart.

Against that backdrop, the number that matters is not a cost-arbitrage multiplier but a maturity one. Zinnov and Nasscom's FY2026 landscape report, covering all 2,117 GCCs in India, finds only 5% have reached “Transformation Hub” status: AI-led operations with real CXO mandates run from India. The largest single cohort, 43%, sit at “Satellite”, meaning capability at scale, and the tier most GCCs never grow out of. That gap compounds visibly over three vendor engagements. The first contract, for fraud scoring, pays for a data pipeline into core banking, an integration layer, and a compliance review of that pipeline. The second, for a chatbot vendor, pays for a data pipeline into core banking again, an integration layer again, and a compliance review again: different vendor, same underlying infrastructure, purchased a second time because no one owns the reusable version. By the third or fourth point solution, a bank has paid for the same foundational plumbing multiple times over, under different line items, without ever owning a version of it the next vendor engagement could simply plug into. An in-house component built once and reused across credit, fraud, and collections is what actually moves a GCC from Satellite toward Portfolio Hub. Another point solution does not.

Source: Zinnov & Nasscom, “GCC Value Orbit” / India GCC Landscape Report, FY2026.

Where AI Agents Fit

Role: Enabler

A GCC's output is other categories' infrastructure: the reusable components a credit or fraud team builds on. The componentised layers a GCC builds and owns are the ones that repeat across every EASE 9.0 category: a Sense layer for real-time data unification, a Decide layer for policy-bound scoring, and an Inform layer for audit logging. Build each once and a Credit Decisioning Agent and a Fraud Detection Agent both run on the same underlying Decide infrastructure, configured with different policy. No customer interacts with a GCC directly; its value is entirely in what it makes cheaper and faster to build everywhere else.

The Talent Question, With Real Numbers Attached

Building a GCC is as much a talent strategy as a technology one, and AI is inverting the economics of that strategy in a way most GCC playbooks have not caught up to. The old model was volume hiring: bring in many junior engineers at standard PSU-adjacent pay bands, accept steady attrition, and backfill from a deep campus pipeline. That model assumed the work itself stayed junior. It does not anymore — AI coding assistants are already absorbing the L1/L2 work a large junior bench used to do at both JPMorgan and Goldman Sachs, which means the AI-era GCC needs fewer, more senior engineers. The catch: industry compensation data shows that exact senior AI/ML tier carries the highest voluntary attrition in the market, 25–30% against 16–22% overall, with a 20–40% pay premium for platform and ML-infrastructure skills specifically. A GCC built on standard PSU compensation bands and a junior-heavy pyramid will be structurally unable to retain the people the AI-first model most depends on.

There is a second, less obvious signal in how the leading banks are actually using their India capability once AI absorbs the routine work: not to shrink it, but to move it up the value chain. Goldman Sachs's roughly 9,000-person Bengaluru and Hyderabad centre now runs live global functions (investment banking, risk, engineering) rather than back-office support, and the centre produced 49 India-based Managing Director promotions in 2025 alone. That is the strategic choice a PSU bank's GCC charter has to make explicitly: is the centre being built to do cheaper versions of headquarters' work, or to own a slice of it outright, once AI removes the volume-based excuse for keeping ownership elsewhere?

Sources: Retail Banker International, coverage of Indian PSBs' Global Capability Centre plans (2026); Deccan Herald, PSU bank GCC inauguration coverage (2026); India GCC salary and attrition benchmarks (2025–26); reporting on JPMorgan and Goldman Sachs AI coding-productivity initiatives (2025–26). Figures as publicly reported, not independently audited.

The Skeptic's Answer

Senior AI/ML talent already churns at 25–30% a year across the industry, so building a GCC around exactly that tier can look like training people at real expense so a competitor can poach them at a 20–40% pay premium the bank cannot always match. A CHRO who has watched a strong hire walk out after eighteen months has every reason to question whether the investment pays off.

The vendor alternative is not actually lower-turnover; it just hides the churn inside someone else's payroll. A vendor's account team rotates too, and every rotation means re-explaining the bank's specific policy logic to a new set of people who were never accountable for getting it right the first time. Goldman Sachs's answer to the retention question was not a bigger pay band alone — it was moving real global-function ownership and 49 India-based MD promotions into its Bengaluru and Hyderabad centre in a single year. People stay for mandate as much as money; a GCC chartered to own outcomes rather than merely execute tickets competes on more than salary.

Infrastructure a GCC Extends Instead of Reinventing

What a GCC actually owns, if it is chartered right, is the layer above the plumbing. The decisioning infrastructure, the policy-engine framework, and the audit-logging layer are parts every bank needs and none needs to reinvent, so Appice provides them ready-built. What the bank's own engineers configure, extend, and own is the policy logic specific to their institution — the fraud thresholds, credit criteria, and collections triggers that only make sense inside one bank's risk appetite. That ownership is the layer worth a senior engineer's career, the one a GCC chartered merely to integrate someone else's product can never hand anyone.