The skills index: what GCCs actually paid for AI talent this quarter

Most compensation commentary on AI talent is built on what companies say they intend to pay. This index is built on what they did pay, the offers that closed, for real roles, at real numbers, across the IT and AI & Data placements we completed in Q1. Intention drifts; a closed offer is a fact. That distinction is the whole reason to read a placement-based index rather than a survey.

It also means the picture is narrower and sharper. We’re reporting the bands we actually moved talent through, and where within them the money went this quarter. Where we cite a specific rupee figure it’s wrapped and awaiting sign-off; the direction of movement is drawn straight from the desk. For the full band tables by role, level and city, this index sits alongside Recruise’s Compensation Index, which reads the same closed-offer data at report depth.

The headline story people expected, runaway premiums at the very top, wasn’t where the real movement sat this quarter. The sharper action was in the mid-senior bands, where AI & Data roles that carry judgment and ownership were bidding against each other. That’s consistent with what we’re seeing structurally: as routine work is assisted, the premium concentrates on the people who own the assisted system rather than operate it. James Gleick’s Chaos gave that behaviour a name, sensitive dependence, where a small input produces an outsized, system-wide effect. A pay band works the same way. A handful of scarce judgment roles bidding against one another doesn’t stay a local event; it re-prices the band around them, because every offer that follows is benchmarked against those closes. The scarcity underneath it is well documented: India’s AI-talent demand is projected to exceed 1.25 million by 2027 against roughly a 50% shortfall in 2024, according to Deloitte and Nasscom, and India was among the fastest-growing AI-hiring markets at roughly +40% per the LinkedIn Economic Graph. What those national numbers can’t show is where inside the bands the money actually moves, which is what a placement-based index reads.

Within IT, the split was visible too. The roles that read as production softened relative to the roles that read as direction and review. It’s the same pattern the rest of our AI coverage keeps surfacing, now showing up in the numbers on the offers as well as the shape of the reqs. Read the band as a single average and it looks calm; read it by what the role actually does and it has pulled into two.

The split is not a rupee quirk; it follows the supply. When the routine build is assisted, the value moves to the people who specify the problem, review what the tool produces, and own whether the result survives scrutiny. That is a scarcer person than the one who produces the output, and the market is pricing the scarcity. The Compensation Index puts the skill premium on the scarce end at 25–50% over a generalist at the same level, against 6–8% for commoditised roles, the widest split we have read inside a single “senior IT” band.

The bench underneath it is genuinely thin. Only about 16% of India’s IT professionals are AI-skilled, according to Nasscom, even as India ranks first globally for AI-skill penetration on Nasscom’s own measure. Thin supply meeting concentrated demand is exactly the condition under which a scarce-skill premium widens while the generalist band sits still. The production roles held because the tool made their output cheaper to produce; the judgment roles ran because the tool made their judgment more valuable to have.

The index is most useful held against your own structure. If your bands were set before this movement, the bigger risk is mis-pricing the mid-senior judgment roles that are quietly where the competition is now. Those judgment roles clear against a shallow bench, so they are the offers most likely to be declined for a reason that never makes it into your data.

For context, India pay overall is projected to rise about 9% in 2026, Aon puts it at 9.1%, with Mercer converging near the same figure, and the mid-senior AI bands here moved faster than that general floor. Use the movement to pressure-test the bands you’ll be hiring into next quarter, rather than to benchmark last quarter. Treat the specific figures as provisional until they’re signed off; the direction is reliable, and the exact numbers are held to the same proof standard as everything we publish.

If you read one signal off this index, read the accept rate on your mid-senior AI & Data offers. A band can look defensible on a benchmark sheet and still lose people at the table, and the judgment seat is where that gap opens first. When a well-benchmarked segment starts posting a low accept rate, that is the survey band and the closed offer telling you they have already drifted apart. The candidate you wanted is being quoted a number your grid hasn’t seen.

The trap is that the miss is invisible in aggregate. The generalist roles you are filling easily hold the average up, so the band looks healthy while the 2 or 3 seats that actually matter quietly fail to close. Watch the scarce band on its own, not blended into the group, and the mis-pricing shows up a quarter before your next budget cycle would surface it.