Nearly every Indian leader in Pearson's 2026 Value of IT Certification Employer Report sees an AI and machine learning skills gap: 96 per cent, with cybersecurity a distant second. The response is just as lopsided. Some 91 per cent of Indian companies are training and upskilling IT staff, and 94 per cent fully pay for certifications, either for any course employees want or for those their roles require.

The survey reached 505 IT and HR leaders at firms with 1,000 or more employees across seven countries, including 64 in India, in mid-2025.

What sits behind those numbers is a shortage that has moved from knowing AI to running it in production, and a hiring market that increasingly wants proof.

The gap is about application

"The biggest gap today is not awareness of AI. It is the ability to apply it effectively in a business environment," says Vinay Kumar Swamy, Country Head of Pearson India. Employers, he adds, want people who can check whether outputs are accurate and know what to do when AI gets it wrong.

Sudarshan Mogasale, CEO of Dassault Systèmes Solutions Lab (India), says that the shortage lies in "production-grade engineering and operationalisation," not in basic model training. A year ago, companies wanted prompt designers. Now they want people who can build autonomous agent workflows, optimise token use and work with small language models.

For Nakul Jain, Co-Founder and CEO of APLYD and Partner at Athena Infonomics, the shift is from "knowing AI" to operationalising it responsibly. That, he argues, takes people who understand "users, workflows, incentives, inclusion, privacy, bias."

Why companies build instead of hire

Globally, the report finds that 83 per cent of companies plan to close skill gaps by upskilling existing staff. Hiring (55 per cent) ranks well behind, and so does buying AI solutions (65 per cent). Among those upskilling, 78 per cent choose certification as their main vehicle.

"Organisations cannot recruit their way through a transformation of this scale," says Swamy. Hiring fills individual roles, he explains, whereas "learning builds capability across the organisation." Mogasale adds that outside hires rarely bring "deep enterprise context" or familiarity with legacy architecture, and that costly specialists and slow searches tilt the economics towards insiders.

He splits upskilling into two tiers. Everyone needs to use AI in daily work, and he believes that live workshops with real tools do that best. A smaller group needs depth in data preparation, model training and agentic AI. Jain and Swamy both say that the right people are found by mapping tasks within roles, not by job title.

Certification as a licence to operate

Once companies train their own people, they need to show that it worked. Pearson's report says that for regulated and government work, credentials function as a "license to operate." In India, 69 per cent of leaders rely on certification to qualify for projects and client engagements. Some 39 per cent formally require it for certain roles, while 61 per cent strongly encourage it. Across the survey, organisations that require certification were likely to call their IT workforce "fully equipped": 70 per cent against 48 per cent.

The practitioners are more measured. Mogasale calls certificates "foundational hygiene filters," proof of vocabulary and willingness to learn. For senior hires, his engineers look for live production deployments, open-source contributions and reproducible benchmarks. Jain cautions that a certification "on its own should not be treated as proof" of real-world ability, and that its value "will increasingly depend on how current and relevant the underlying skills are."

Swamy, naturally, makes the strongest case for credentials. "They should be both," he says of signal and substance, provided they come backed by assessments and project experience.

Hiring is moving the same way. Pedigree still screens candidates at scale, Mogasale says, but "ten years of legacy software engineering experience does not automatically translate into AI architecture capability." Dassault tests candidates with live pair-programming alongside a tool like GitHub Copilot or Cursor, watching how they review, debug and audit the code it writes. Swamy sums up the broader shift as a move from asking "What did you study?" to also asking "What can you do today?"

A higher bar, set by GCCs

Experts point that Global Capability Centres are lifting expectations. Mogasale says that they are "no longer viewed as low-cost operational back offices," and now work as innovation hubs, so engineers must go "beyond writing code to defining product vision." Jain describes a story "moving beyond cost arbitrage towards value creation," with Indian teams expected to shape global AI work.

Because tools turn over every few months, all three say that training tied to one platform ages quickly. They favour durable skills: data literacy, evaluating outputs, and knowing when a human must stay in the loop. Jain puts the test plainly. The measure, he says, is whether employees can use AI "effectively and responsibly as the technology evolves, rather than simply whether they have completed a course."

Verification as routine

The report's leaders expect certification to behave more like a professional licence. Some 88 per cent say that it will matter more within three to five years, and 62 per cent say "much more."

Swamy expects assessment to shift from "a one-time gate" to "an ongoing talent journey," since a qualification earned years ago "may not fully reflect an individual's current capabilities." Jain foresees more practical tests, portfolios and micro-credentials, but warns that judgement and responsible AI use need assessing too, not only technical skill.

For workers, that means continuous learning becomes "an ongoing requirement for remaining employable," in Jain's words. For employers, it means measuring talent in ways that don't expire at the point of hiring. The advantage, he says, will go to those who "identify emerging skills early, build them quickly and demonstrate them credibly."