India's AI ROI Gap: A Measurement Framework for Leaders

India has one of the highest AI adoption rates in the world, and one of the widest gaps between using AI and seeing a return from it. If you're leading a team or a company right now, that gap is probably closer to home than a global statistic — it's the reason you can see people using AI everywhere, without being able to say what it's actually worth.

Salesforce's July 2026 research on workplace AI, covered in its press release on India's AI adoption, found that 80% of Indian employees use AI multiple times a week and 41% use it daily — ahead of the global pace. But only 35% say AI's return on investment has met or exceeded their expectations, against a global average of 22%. India is doing better than most, and still, most leaders can't confidently say what their AI spend is buying them.

Why the gap exists

Part of the answer is structural. Through 2022–2024, most Indian organizations treated AI as a series of discrete experiments: pick a use case, run a pilot, measure the result, decide whether to scale. That model worked when AI use was contained to a handful of pilots. By 2026, AI is woven into daily work across writing, coding, customer support and analysis — and the old one-pilot-at-a-time measurement approach has broken down. Usage is everywhere; visibility into what it's producing is not.

The other part is that most tools report the wrong number. Token counts, seat licenses and query volume tell you how much AI is being used, not what it's worth. Which is a large part of why adoption keeps climbing while confidence in ROI doesn't.

A tool worth noting: what better visibility looks like

Anthropic's Smart Reports for Claude Enterprise, which went into beta on September 10, 2026, is a useful example of the shift leaders should be pushing their own AI vendors toward. Instead of reporting how many tokens a team burned, it reports the kind of work getting done, what it cost, and — notably — where sessions ran into friction: connector failures, rework loops and repeated tool errors. Enterprise admins get up to ten reports a month free during the beta.

You don't need to be a Claude Enterprise customer to take the underlying lesson. Whatever AI tools your team uses — Copilot, Gemini, ChatGPT Enterprise or others — ask the same three questions of your usage data: what work is actually getting done, what is it costing per unit of that work, and where are people getting stuck. If your current tooling can't answer those questions, that's the gap to close first, before you invest in more licenses.

A practical framework for closing the ROI gap

1. Pick one workflow, not the whole company

Don't try to measure AI's ROI across the organization at once — that's exactly the fragmented, unmeasurable pattern most Indian companies are stuck in. Pick one frequent, painful workflow: drafting proposals, first-line customer replies, expense reconciliation, code review. Measure its baseline cost and time before AI touches it.

2. Track usage, outcomes and friction — not just adoption

"80% of our team uses AI weekly" is an adoption number, not a value number. For your chosen workflow, track:

• Outcome: did the task get done faster, cheaper, or better, measured against your baseline.
• Friction: where people had to redo work, escalate, or abandon the AI output and start over.
• Verification cost: how much time went into checking the AI's work before it was safe to use. (This is the same "botsitting" cost we've written about separately — it belongs in any honest ROI calculation.)

3. Tie the number to a business outcome, not a productivity feeling

"It feels faster" is not a metric. Convert the baseline-to-AI comparison into something finance recognizes: hours saved per week at a loaded cost, error rate before and after, cycle time from request to delivery. This is what turns a pilot into a business case you can defend at budget time.

4. Review monthly, and feed friction back into training

A workflow that looked good in week one often develops friction by week four, as people hit edge cases the pilot didn't cover. Put a monthly review on the calendar: look at what's working, what's causing rework, and what needs a training fix rather than a tooling fix. Most friction is a skills gap, not a model limitation — which is why role-specific training closes more of the ROI gap than switching tools.

5. Decide to scale, pause or kill — on purpose

At the end of each review cycle, make an explicit call: scale the workflow to more teams, pause and fix a specific problem, or kill it if the numbers don't support it. Organizations that struggle with AI ROI are rarely the ones running bad pilots — they're the ones that never formally closed the loop on the pilots they ran, so nothing was learned and nothing scaled.

Measurement without governance is incomplete

A measurement framework tells you if AI is working. It doesn't tell you if it's being used safely — that's a separate, related problem covered in our practical checklist for India's AI governance gap. The two go together: know what your AI spend is producing, and know that it's being produced within the guardrails your business needs.

India's lead in AI adoption is a real advantage — most companies elsewhere are still trying to get people to use these tools at all. The leaders who turn that lead into an actual return will be the ones who stop asking "how much are we using AI" and start asking "what is it worth." If you're building that measurement discipline into how your team works with AI, our AI for Leaders program is built around exactly this: narrow pilots, real measurement, and a path to scale.