A useful AI tool and a dependable business process are not the same achievement. Saving time on a task is one test. Knowing who checks the output, handles an exception and takes responsibility when something goes wrong is another.
Fresh survey evidence from India makes that distinction worth examining.
The Breakdown
In Kyndryl’s 2026–27 Readiness Report, 60% of Indian respondents investing in AI and machine learning reported positive returns, against 57% in the previous edition. Yet the average share reporting complete preparedness for external risks stood at 27%, down eight percentage points.
Those are different measures. The figures do not establish that successful AI investments caused preparedness to fall, or that the same respondents reporting gains were also reporting weaker readiness.
There is a more operational contrast: 85% of Indian respondents agreed that they were deploying AI in workflows beyond pilots, while 31% completely agreed that they had formally redesigned roles around AI. The response thresholds differ, so this is not a precise measure of an implementation gap. It does suggest that deployment and organisational redesign need separate checks.
What the survey can tell us
The report, released globally on November 10, surveyed 3,700 senior leaders across 21 countries, including 200 in India. Edelman Intelligence conducted online surveys and telephone interviews for Kyndryl between July 10 and September 2, 2026.
Half the global sample represented companies with annual revenue of at least $1 billion. These are executives’ assessments, not audited returns or a representative verdict on every Indian business. Kyndryl also sells enterprise technology services. Its research is useful evidence about respondents’ views, not an independent audit of corporate AI performance.
Why This Matters
India already has a concrete example of how the operational question can be framed. The RBI’s FREE-AI committee report, published in August 2025, recommended model inventories, validation, continuing monitoring and defined fallback mechanisms for financial institutions. It also recommended separating product approval evaluation from the teams developing and deploying models.
Those were committee recommendations, not a new rule announced this week. Nor do they automatically apply to every industry. Their relevance here is practical: a business case and an operating-control case answer different questions.
For example, an internal document-summary tool might reduce reading time. Letting its output trigger a customer-facing financial decision changes what must be checked. The saving does not establish the accuracy of the decision, the appropriateness of access to data, or who can halt the process. This is an illustrative distinction, not a claim about a surveyed company.
The Crunch
Before expanding an AI workflow, ask two questions separately: what value has it demonstrated, and what makes it safe and dependable at the proposed scale? A positive answer to the first should not stand in for an answer to the second.
Cover: original conceptual illustration.
Related categories: People & Work; Business & Brands
Leave your thoughts