पाठशाला Pathshala · हिसाब Hisāb, Unit economics · Lesson 25 · Scale
Gross margin expansion: the five levers
Gross margin rarely rises on its own. Map cost of revenue line by line, assign each line to one of five levers and plan a ten-point improvement with an owner and a date for each.
Pathshala, The Founder Library · 11 October 2026 · 8 min read
A company that grows from ₹10 crore to ₹50 crore of revenue usually discovers that its gross margin did not grow with it. The cloud bill scaled with usage, the support team scaled with customers and the large accounts negotiated their price down. Margin does not expand because a company gets bigger. It expands because someone pulls specific levers, in a specific order, and measures each one in points.
This lesson maps cost of revenue to the five levers that move it, sizes each in points of margin, works through a ten-point plan for one company and ends with the quarterly review that keeps the plan honest.
Map the cost of revenue before touching it
Take the last three months of cost of revenue, ex-GST, and split it into lines a person can own. For most technology-led Indian companies the lines are the same: third-party vendors and licences embedded in the product, cloud and infrastructure, payment gateway and transaction costs, support and customer success, implementation and delivery staff, and for physical goods the product cost, packaging and fulfilment. If the line between cost of revenue and operating expense has not been drawn carefully, draw it first; the [lesson on COGS](/library/what-belongs-in-cogs) sets out what belongs above the line, and a margin plan built on a flattering line will deliver points on paper only.
Express every line as a share of revenue, not in rupees. A cloud bill of ₹12 lakh a month means nothing alone; at 12 per cent of revenue it is a lever worth several points. Then compute each line for each product or segment, because the blended picture hides the work. A company at 52 per cent overall may be at 75 per cent on its software line and 35 per cent on its implementation line, and the plan for each is different.
Finally, note which lines grow faster than revenue. A line that was 10 per cent of revenue a year ago and is 12 per cent now is where the margin is leaking, and it is usually the first place to work.
The five levers
Vendor and third-party cost. Everything bought from someone else and resold inside the product: SMS and WhatsApp messaging, KYC checks, maps, data feeds, licensed software, outsourced logistics. The lever is negotiation, consolidation and substitution: committed volumes for a lower rate, one supplier instead of three, a cheaper equivalent where the customer cannot tell the difference. The [vendors lesson](/library/vendors-procurement-contracts-you-keep-renewing) covers the contracts.
Infrastructure. Cloud compute, storage, data transfer and, for AI products, inference. The lever is engineering: commitments and reserved capacity, right-sizing, caching, cheaper models for easy requests, and in the extreme, moving steady workloads off the public cloud.
Support and delivery. The people who answer tickets, onboard customers and implement the product. The lever is productivity: fewer tickets per customer through better product and documentation, self-serve onboarding, deflection to help content, and moving implementation work into the product.
Mix. The share of revenue that comes from higher-margin products, segments, channels or cities. The lever is commercial: what sales is paid to sell, which plan is the default, which segment marketing spends on. Mix can move margin faster than any cost programme, and in either direction.
Price. A price rise with no change in cost flows almost entirely to gross profit. It is the most powerful lever and the most dangerous, because it can cost customers. The [pricing-power lesson](/library/pricing-power-raising-prices-without-losing-base) sets out how to raise prices without losing the base.
Infrastructure: the lever software companies forget
Software companies are told their margins should sit near 80 per cent, and many assume infrastructure takes care of itself. It does not. Sarah Wang and Martin Casado of Andreessen Horowitz, in The Cost of Cloud, a Trillion Dollar Paradox, found committed cloud spend averaging 50 per cent of cost of revenue among the companies they examined, with one large private software company reporting 75 to 80 per cent as common. Their cleanest example is Dropbox, whose gross margin rose from 33 per cent to 67 per cent between 2015 and 2017, a rise the company attributed mainly to infrastructure optimisation alongside revenue growth, with about $75 million of cumulative savings over the two years before its listing.
The lesson is not that every company should leave the cloud. It is that infrastructure is a cost line with an owner, a target and a review, like any other. For AI products it matters more. Martin Casado and Matt Bornstein, in The New Business of AI, found AI companies often at 50 to 60 per cent gross margin against 60 to 80 per cent-plus for comparable SaaS, with 25 per cent or more of revenue going to cloud resources and up to 10 to 15 per cent to human review and data work. Their advice reads as a lever list: reduce model complexity, share models across customers, narrow the problem, and measure the real variable cost rather than letting it sit in R&D.
One rule applies with particular force in India. Cloud bills are often in dollars while revenue is in rupees, so a weakening rupee lowers gross margin with no change in usage. Track the line in both currencies, keep it ex-GST like every other line, and when the rupee line moves ask first whether usage moved or the exchange rate did.
Support, delivery and mix
Support and delivery are the levers founders resist most, because they feel like quality. They need not be. The question is not how many people support costs but how many tickets each customer generates and why. A product that sends every new customer to a human for setup has a delivery cost that scales one for one with growth; a product that onboards most customers itself has one that scales with the exceptions. Measure tickets per hundred customers and implementation hours per new account every month, and give the product team a target for both. The [support lesson](/library/customer-support-as-operation-not-cost-centre) covers the operating model.
Mix is where a strategy decision shows up in the margin. A company selling software at 75 per cent margin and implementation at 35 per cent has a blended margin that depends on the split. If implementation is 30 per cent of revenue the blend is 63 per cent; at 20 per cent it is 67. Ten points of revenue shifted from one line to the other is four points of margin, with no cost cut at all. The levers are sales compensation on gross profit rather than revenue, partners who take on implementation, and pricing that makes the high-margin product the obvious choice. The same logic applies to channels and cities, which the lesson on [unit economics by business line and city](/library/unit-economics-across-business-lines-and-cities) works through.
Mix also explains most unexplained margin declines. A new large customer on a discounted price, a fast-growing low-margin city or a shift to marketplace sales with higher commissions can each pull the blend down while every individual line looks fine.
A ten-point plan, worked
A Pune company sells workflow software to manufacturers with an implementation service attached. It makes ₹1 crore of revenue a month at a 52 per cent gross margin. Its ₹48 lakh of cost of revenue splits into ₹18 lakh of vendors and third-party services, ₹12 lakh of cloud, ₹14 lakh of support and delivery and ₹4 lakh of payment and other costs. The board asks for 62 per cent in four quarters.
The plan pulls each lever modestly. Consolidating messaging and data vendors on a committed volume saves 10 per cent of vendor cost, worth 1.8 points. Reserved capacity, right-sizing and caching save a quarter of the cloud bill, worth 3 points. Self-serve onboarding for smaller customers saves 15 per cent of support and delivery cost, worth 2.1 points. Paying sales on gross profit shifts five points of revenue from implementation to software, worth 2 points. A 4 per cent price rise on renewals, applied last, adds 1.5 points. Together: 62.4 per cent, and monthly gross profit from ₹48 lakh to ₹64.9 lakh.
The figure starts on that plan. Set every lever to zero and pull them one at a time. Notice that no single lever reaches ten points on a plausible setting; the plan works because several do two or three each. Notice too that price applied last is worth less in points than it would be first, because it multiplies a smaller cost base. That is the reason to order the levers this way: the cost work is permanent, and a price rise on a cleaner base is easier to defend.
Margin is not a result. It is five levers, each with an owner, a number in points and a date.
Sizing every lever in points also makes the trade-offs visible. A cost cut that saves ₹5 lakh a month sounds large until it is written as half a point; a pricing change that adds four points sounds small until it is written as ₹40 lakh a month. Investors read the result the same way. Bill Gurley’s All Revenue is Not Created Equal argues that you cannot generate much cash from revenue saddled with large variable costs, which is why lower-margin companies earn lower multiples of revenue. Bessemer’s Scaling to $100 Million puts the average cloud company at 65 to 70 per cent, with the middle half between roughly 60 and 80; a plan that moves a company from below that band into it changes how its revenue is valued.
The quarterly margin bridge
Once a quarter, after the books close, the finance lead puts one chart in front of the founders: last quarter’s gross margin, the gain or loss from each of the five levers, and this quarter’s margin. Every bar has an owner’s name beside it. Lines that moved for reasons nobody planned, a currency swing, a large discounted customer, a usage spike, go in a sixth bar marked unplanned, and the meeting spends most of its time there.
Each month in between, check three numbers: each cost-of-revenue line as a share of revenue, margin by product line, and the share of revenue from each line. Set the targets for the next four quarters in points, write them where the team sees them, and expect two or three points a quarter rather than ten at once. A margin plan that delivers steadily is worth more to a company and to its investors than one that promises everything in the next quarter.
Nothing here is legal, tax or investment advice. The Pune company is illustrative; its figures are the figure’s defaults so that every number in the text can be reproduced.
Sources
- Sarah Wang and Martin Casado, The Cost of Cloud, a Trillion Dollar Paradox, Andreessen Horowitz, May 2021 — Committed cloud spend averaged 50 per cent of cost of revenue; Dropbox gross margin rose from 33 to 67 per cent between 2015 and 2017; about $75 million of cumulative savings.
- Martin Casado and Matt Bornstein, The New Business of AI (and How It’s Different From Traditional Software), Andreessen Horowitz, February 2020 — AI gross margins often 50–60 per cent; 25 per cent or more of revenue on cloud; up to 10–15 per cent on human data work.
- Mary D’Onofrio, Ethan Ding and Atlas editors, Scaling to $100 Million, Bessemer Venture Partners, September 2021 — Average cloud gross margin 65–70 per cent; middle half roughly 60–80 per cent.
- Bill Gurley, All Revenue is Not Created Equal: The Keys to the 10X Revenue Club, Above the Crowd, May 2011 — Revenue saddled with large variable costs generates little cash; lower-margin companies trade at lower revenue multiples.