The Practical Technologist · Industry Analysis
A friend of mine runs a large account at an Indian IT services company. Last week he was staring at a renewal spreadsheet that refused to add up. The client wanted the same scope, the same SLAs, the same team strength for 30% less.
I thought it was one difficult client. So I asked around. Two more mid-size firms, same story. Different clients, different geographies, same number.
30% has quietly become the default ask.
And here is what bothers me about it. That number arrives with no reference to the complexity of what we deliver, no reference to the 6–8% wage revision we just paid out, no reference to whether AI can actually touch the work in that particular engagement. It is a round number borrowed from a boardroom slide, not calculated from our estate.
But I want to be careful here, because there are two separate arguments getting mixed up. One is "is the number fair?" The other is "is the number real?"
The number is real. Denying it is the fastest way to lose the account.
The evidence: this is not your account, it's the market
Sandeep Kalra, CEO of Persistent Systems, said it publicly: clients are asking for "the same work for 25% to 30% less" — faster, with higher productivity. (Reuters, Aug 2026)
Everest Group's CEO Jimit Arora put the power balance plainly: "It's a desperate market for the service providers. The odds are very much in favour of clients."
The structure of contracts is shifting underneath us, not just the price:
- At TCS, roughly 80% of finance and HR business services contracts are now outcome-linked double the share since late 2023.
- Cognizant–Daimler Truck signed an engagement where AI-driven cost savings are split between vendor and client.
- HCLTech–E.ON: no payment at all in year one, then payments tied to efficiency gains.
- HFS Research reports contracts being reopened within 24 months of signing, not at renewal. Deals above $50M ACV face the most pressure.
Meanwhile the market has already priced this in. The Nifty IT index is down about 20% this year roughly $73 billion of value across ten constituents.
And Nasscom still projects the industry at $315 billion in FY26, growing 6.1%, with headcount up only ~2.3%. Read those two numbers together: revenue is growing three times faster than headcount. The link between people and revenue is breaking. That is the AI dividend — and the client has noticed it before we finished measuring it.
👉 So the 30% is not an insult. It is the market telling us the unit of sale has changed.
The sugarcane juice stall
There's a sugarcane juice (ऊसाचा रस) stall near our office. For years the man crushed cane by hand with a wheel and two helpers. Then he bought an electric crusher.
A regular customer saw the machine and said, half-joking: "Ab machine kaam kar rahi hai — daam kam karo." The machine does the work now, so cut the price.
Fair-sounding. But look at what the machine actually changed. The cane still costs what the mandi charges. Ice, lemon, electricity all up. The crusher itself cost money and needs servicing. What the machine really did was let him serve three glasses in the time he used to serve one, and let go of one helper.
His mistake would have been to cut the price 30% on the same glass and keep everything else identical. He'd have gone under in a season.
What he actually did: served more customers per hour, added a lemon-mint variant at a higher price, and stopped paying for the labour the machine replaced.
Volume and mix, not price. That's the whole argument of this article.
The arithmetic nobody puts on the slide
Here's why a flat 30% cut on unchanged scope is dangerous, and worth showing your client.
Take an engagement at ₹100 revenue, ₹80 cost, ₹20 profit a healthy 20% margin.
- Cut price 30% → revenue ₹70.
- To hold a 20% margin, cost must fall to ₹56. That's a 30% cost reduction.
- To hold the same ₹20 of absolute profit, cost must fall to ₹50. That's a 37.5% reduction.
- Now add the 5–8% wage revision you just gave. Year-one cost reduction needed: roughly 33–40%.
Now the other side of the ledger. What does AI actually give you today?
The 2025 DORA report:
90% AI adoption, over 80% of developers believing they're faster found a positive relationship with throughput but a negative one with delivery stability. AI is, in their words, an amplifier:
strong teams get faster, weak teams get faster at producing defects.
METR's randomised trial is more sobering. Sixteen experienced developers on codebases they knew well were 19% slower with AI tools, while estimating afterwards that they'd been 20% faster. That 39-point perception gap is exactly the gap that walks into renewal negotiations.
And remember what coding actually is inside a services engagement. In the accounts I've worked on, hands-on-keyboard development is rarely more than 20–30% of the total effort the rest is analysis, testing, environments, support, coordination and documentation. Even a generous 30% speedup on the coding slice is 6–9% off the total. Check your own numbers; the point holds at any reasonable split.
Tech Mahindra's CEO Mohit Joshi called competitors baking in 70–80% productivity gains over five to seven years while guaranteeing prices "irrational." He's right. The gap between 8% and 30% is not closed in the IDE. It's closed everywhere else.
Where the other 20 points actually come from
If AI-in-coding gives you 6–9%, the rest has to be engineered. This is the honest list:
1. The 70% that isn't coding. Test design and regression, environment provisioning, L1/L2 tickets, incident triage, release documentation, knowledge transfer, onboarding. This is where agentic workflows pay properly, because the work is repetitive and the ground truth is well defined. Most firms have deployed Copilot to developers and done nothing for the support tower. That's leaving the bulk of the money on the table.
2. The pyramid, redrawn deliberately. Former Infosys CFO V. Balakrishnan is blunt: "The pyramid model is gone. With coding agents, we no longer need basic coding." Fine but the answer isn't simply "hire fewer freshers." It's redefining what a fresher does in year one: agent supervision, eval design, test curation, data quality. Cheaper to the client, and a better apprenticeship than the ticket queues we used to hand them.
3. Reuse you actually charge for. Every accelerator, every prompt library, every eval harness you build on one account should be amortised across five. If your 30% is funded by reuse, you can give it away once and keep it forever.
4. Complexity segmentation. This is the part I feel strongest about. A flat 30% across a portfolio is intellectually lazy on the client's side and negligent on ours to accept. Split the estate: highly repeatable, well-documented, high-volume work can absorb 25–35%. Regulated, legacy, poorly documented, high-blast-radius work absorbs 5–10% at best and pushing AI into it is where the DORA stability finding will bite you. Bring that segmentation to the table as your analysis, before they bring their flat number.
What to do on Monday
✅ For service providers:
- Know your cost-to-serve per outcome, not per FTE. If you can only quote rate cards, you will only ever be squeezed on rate cards.
- Never guarantee a productivity number you haven't measured on your own estate. Run a 90-day instrumented baseline, cycle time, defect escape rate, rework — before you sign to anything.
- Split the benefit explicitly. Gain-share beats give-away. Cognizant and HCLTech have now set the precedent; use it as market comparable.
- Trade price for something. 30% off is a different conversation attached to a longer term, a wider scope, a volume commitment, or a shift from T&M to outcome pricing. Never concede price on unchanged terms that's the only real error here.
- Attack the support and test towers first. Faster payback, lower risk, and it's where the client's own measurement is weakest.
- Re-price the top of the pyramid up. The architecture, the eval design, the risk ownership = as commodity work deflates, scarce work should inflate. Don't let the average hide both.
✅ For clients reading this (and some of you are):
Ask your vendor for the segmentation, not the discount. A partner who agrees to 30% across the board without a delivery redesign is either lying to you or is about to under-invest in your account. You will pay for it in year two, in defects.
Final thought
We spent two decades selling effort. Predictable, measurable, billable effort. AI has made effort cheap and made judgement expensive.
So when a client asks for 30%, they are not being unreasonable. They are being early. They are telling us clumsily, with a borrowed round number that they no longer want to buy our hours.
Fight the number and you lose the account. Accept the number without changing the model and you lose the company.
30% is not a discount demand. It's a notice to change the unit of sale.
The sugarcane vendor understood it faster than our industry has.
💬 A question for those of you in the middle of a renewal right now: when your client asked for the flat percentage, did anyone at the table put the complexity segmentation on screen — or did the negotiation stay on the rate card? I'd genuinely like to know how often we're bringing our own analysis versus reacting to theirs.
I write The Practical Technologist every week practical takes on AI, business and building things, from 26 years in IT. Subscribe so you don't miss the next one.
Sources
- AI reshapes India's IT services sector contracts as clients demand more for less — Reuters/The Star, Aug 2026
- AI drives IT contract renegotiations within 24 months as pricing models shift — HFS Research
- Announcing the 2025 DORA Report — Google Cloud
- Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity — METR
- Nasscom sees Indian tech industry at $315 billion in FY26
- India IT salary hike 2026 — broad-base ranges across TCS, Infosys, Wipro, Accenture, Cognizant
First published on LinkedIn.