August 26, 2026
· 7 min readThe Billable Hour Is Dead — AI Unbundled India's $315B IT Industry
TCS runs 80% of its business-services contracts on outcomes, five-year deal values shrank up to 35%, and a 500-person migration now prices at 50 people. The hourly billing model isn't being discounted — it's being taken apart layer by layer.
AI didn't shave a percentage off Indian IT's billing rates. It took the billable hour apart and replaced it with three different pricing machines. Here's exactly what changed, who's already living in the new model, and what it means if you sell anything priced by the hour.
The invoice your clients stopped believing in
Picture the average enterprise procurement review in 2026. A five-year managed-services proposal lands on the desk, priced the way Indian IT has priced everything since the 1990s: heads × hours × rate. Except the buyer has already watched agents clear 40% of their support tickets, and their CFO wants to know why line two still charges for humans.
That gap — between what AI actually costs to deliver and what the old invoice claims — is the whole story. India's IT services industry earns about $315 billion a year, built on billable effort. In 2025-26, both ends of that equation got attacked at once: buyers started demanding 25–30% less for identical scope (Persistent Systems' CEO told Reuters this directly), while agents collapsed the effort base underneath the price.
Everyone's instinct is that this ends with cheaper hours. Wrong. Hours aren't being discounted. They're being unbundled.
One invoice becomes three
💡 The reframe: there is no single replacement model. The old invoice is splitting into three layers, and each layer prices differently.
Layer 1 — Outcomes where results are measurable. Payment ties to business metrics instead of effort. TCS says ~80% of its finance/HR/business-services contracts are now measured on outcomes — roughly double the share since late 2023. Real structures already signed: Cognizant and Daimler Truck split AI cost savings between vendor and client; HCLTech agreed to a cloud deal for E.ON with zero payment in year one, then outcome-tied payments; Tech Mahindra runs a healthcare engagement paid on 40% fewer tickets, 20% faster resolution, 30–35% less technical debt.
Layer 2 — Tokens where agents do the work. When software delivers the service, usage gets metered like cloud compute. HfS Research describes the emerging deal shape: a subscription for the platform, a consumption charge where "AI time is measured in tokens," and a performance kicker on top. Your agent fleet is now a COGS line item, not a salary line.
Layer 3 — Shrinking T&M residue. Whatever can't be measured and isn't fully agent-driven still bills by the hour — just smaller every cycle.
How fast each firm moved there varies wildly:
| Firm | Outcome-based share | Source |
|---|---|---|
| TCS — business services segment | ~80% (2x since late 2023) | Reuters via ET |
| Cognizant — new BPO contracts | 45% | Business Standard |
| Wipro | Officially declared the shift: "effort-based pricing to consumption, transaction, and business outcome" | Wipro AGM via Moneycontrol |
| Coforge — total revenue | 6–7% | Business Standard |
| Infosys | ~None yet — CEO: "not so much that it has become a large part of our activity" | Business Standard |
⚠️ That spread is the tell. An ~80× difference between the leader and the laggard means this isn't marketing sync — it's genuine strategic divergence, and the segment gradient matters more than company size. People-heavy BPO/BPS work repriced first because its outcomes were always countable.
The deflation nobody's headline shows
Rate cards are the smallest part of the story. The pressure stacks in three layers, each bigger than the last:
| Layer | Magnitude | Evidence |
|---|---|---|
| Blended rate-card drift | Declining ~24 straight months (e.g., US web/mobile dev: $77–94 → $75–91/hr) | Everest Group data via ET |
| Scope-adjusted discounting | Typical 5-year deal TCV down 20–35% in 2025; further 32–44% fall forecast; 30–70% productivity-discount demands | Everest Group via ET |
| Effort-base collapse | A 500-person, 18-month T&M migration reprices to ~50 people | Motilal Oswal via Financial Express |
Moneycontrol puts it in revenue terms: firms are seeing 5–15% compression where AI gains are visible. And the market already voted — the Nifty IT index dropped roughly a fifth in 2026, erasing about $73 billion in market value (Reuters). Everest Group's CEO summarized the bargaining table in one line: "It's a desperate market for the service providers. The odds are very much in favour of clients."
Where do the savings go? Motilal Oswal's answer: they're "extracted almost entirely by enterprise buyers" — at least until vendors learn to reprice first.
The "$2.6B AI revenue" mirage
Every major now waves a big AI number. None of them mean the same thing:
| Firm | Label | Unit |
|---|---|---|
| TCS | "Annualized AI Revenue" — $2.6B | dollars, annualized, scope undefined |
| HCLTech | "Advanced AI" — $620M | dollars, annualized, scope undefined |
| Infosys | "AI revenues at 8.2%" | % of quarterly revenue, undefined |
| LTIMindtree | "$60M incremental via agentic AI" | net-new dollars — the only precise one |
| Wipro | nothing disclosed | — |
Notice the pattern: the biggest number has the vaguest definition, and the only honestly-scoped figure ($60M, ~1.3% of LTIM's revenue) is the smallest. Four labels, three units, one silence. Don't sum them; don't compare them. They're narrative cover for what is currently a cost-takeout business — Kotak's analyst note calls GenAI-led revenue deflation "a real drag" that AI opportunities haven't offset yet.
There's one genuinely interesting signal inside LTIMindtree's disclosure though: its top five accounts have finished harvesting AI productivity savings and started adding net-new spend. Defense first, offense second — account by account, not industry-wide.
Who actually wins right now
- Clients/buyers — capturing most of the savings today (Motilal Oswal).
- Vendors who redesign delivery toward productized, outcome-linked services — HfS: they "can protect, and in some cases expand, margins". Those who absorb AI without repricing bleed.
- Consolidating mid-tiers — Persistent posted record $1.15B TCV, +16.5% constant-currency organic growth while large-cap peers grew 0.9–2.4%. But check the headlines: Coforge's +49% was Encora-acquisition math, ~1.1% organic.
- Losing first: entry-level, ticket-shaped work. Wipro's own BPS chief puts 15–20% of entry-level services revenue structurally at risk — while noting over 70% of that same business is already AI-embedded.
And the industry body? NASSCOM's Strategic Review 2026 now describes providers as having "re-engineered revenue models, moving away from FTE delivery" — past tense. The experiment phase is officially over.
Where this model still hurts
Trade-off honesty section, because the new pricing isn't free money for anyone:
- Gain-share deals transfer risk to the vendor. The public typology (via Wipro's BPS head): recover your investment first, then split gains 50-50 or 60-40. If the AI savings don't materialize, the vendor eats it. Actual ratios on named deals? Private — that's the one place the industry won't show its cards.
- Token pricing is unproven as a margin pool. If agent delivery commoditizes into pass-through compute costs, Layer 2 becomes a race to zero too.
- Outcome measurement is a contract-negotiation tax. Someone has to define, instrument, and audit those 40%-fewer-tickets SLAs — that's real engineering work before revenue moves.
- The laggard position isn't crazy. Infosys holding back looks bad until a gain-share deal goes sideways somewhere and reprices everyone's risk appetite overnight.
If you price by the hour, read this twice
You may not run an IT giant, but the mechanism doesn't care about company size:
- Audit your invoices the way buyers now do. Which line items could a client replace with agents this quarter? Price that risk before they do.
- Attach your fee to a metric the client already tracks. Tickets closed, uptime, conversion — measurability is what moves you from Layer 3 into Layer 1, where pricing power survives.
- If agents are your delivery, meter like infrastructure. Subscription + usage + performance beat hourly every time the work is machine-shaped.
- Never discount hours to defend a deal. Re-scope against outcomes instead — a smaller guarantee beats a bigger hourly discount in every renewal negotiation from here on.
- Watch who signs gain-share terms. When vendors start publishing real split ratios, the repricing window closes for late movers.
Final Thoughts
- The billable hour isn't dying by discount — it's being unbundled: outcomes where measurable, tokens where agents deliver, shrinking T&M residue elsewhere.
- Penetration is wildly uneven (~80% at TCS business services vs ~0% at Infosys) — segment and measurability decide speed, not company size.
- The headline "AI revenue" numbers are unit-incompatible narratives; the only clean datapoint is also the smallest ($60M net-new).
- Deflation stacks in three layers, and the biggest (effort-base collapse, 500→50) barely shows up in "rates."
- Savings currently flow to buyers; vendor margins now depend on repricing architecture, not labor pyramids.
👉 Pull up your current rate card tonight and mark every line a client could reasonably ask to replace with agents by next quarter. Then decide which of those lines you'll convert to an outcome fee before the renewal call — because that conversation is already scheduled on their side.
FAQ
Is outcome-based pricing actually replacing time-and-materials in Indian IT?
It is replacing it segment by segment, not everywhere at once.
TCS reports roughly 80% of its business-services segment contracts are measured on outcomes, and Cognizant signs 45% of new BPO deals that way. But mid-tier firms sit at single-digit percentages and Infosys has publicly said there has been no material uptick yet. The direction is one-way; the speed depends on how measurable the work is.
What is token-based pricing in IT services?
When AI agents do the delivery work, vendors meter that work like cloud compute: you pay for tokens consumed or agent-hours executed instead of human hours billed.
Analyst firm HfS Research describes new deals as hybrids — a platform subscription, plus consumption priced in tokens, plus a performance component tied to results.
Are the 'AI revenue' numbers from TCS, Infosys and HCLTech comparable?
No. Each firm self-defines the label: TCS cites $2.6 billion annualized AI revenue, HCLTech counts $620 million of Advanced AI revenue, Infosys reports AI as 8.2% of quarterly revenue, and LTIMindtree only discloses net-new agentic revenue.
Different units, undefined scope, zero standardization — treat them as marketing narratives, not an industry metric.
Who keeps the savings when AI makes delivery cheaper?
Right now, mostly the client. Analysts at Motilal Oswal conclude the productivity savings are extracted almost entirely by enterprise buyers.
On gain-share deals, vendors typically recover their investment first and then split remaining gains — commonly described as 50-50 or 60-40 — which means the vendor underwrites the model risk upfront.