How to Benchmark a Business, and Why Most Benchmarking Fails
Accurate analysis, a reasonable peer set, and a report nobody opens twice. The failure happens before a single number is pulled.
Summary
- A benchmark that is not attached to a decision produces a percentile table nobody acts on. Ask what the company will do differently depending on what this shows, and reframe or decline the work if there is no answer.
- The right analytical frame comes from the economic engine, not the industry. Seven engines, each with a binding constraint that tells you which gap is worth spending money to close.
- A gap is a question, not a finding. The output worth paying for is a short list of gaps worth closing, what drives each, and which ones are structural and should be left alone.
Most benchmarking fails quietly.
The analysis is accurate. The peer set is reasonable. The report lands, the management team reads that they sit at the 40th percentile on gross margin, someone says "interesting", and the file is never opened again. Nothing was wrong with the work. It simply answered a question nobody had to act on.
This is the default outcome rather than the exception, and the cause is almost always the same. The exercise was framed as where do we stand, rather than what should we do differently. Those sound like the same question. They are not, and they produce very different analysis.
A benchmark is worth building when it is attached to a decision. That might be a target being set, capital being allocated, a valuation being argued on a multiple, or a cost programme being justified. If there is no decision behind it, the percentile table is trivia.
The test we apply before starting any benchmarking work is a single question put to the client: what will you do differently depending on what this shows? If there is no answer, the engagement is a reporting exercise, and it should be reframed or declined.
What follows is how we think benchmarking should be done. The domains worth measuring. Why the same metrics mean different things in different business models. What changes for startups and smaller companies, and what separates a benchmark that survives scrutiny from one that falls apart the first time someone checks it.
Start with why, not with the data you happen to have
Every benchmarking exercise resolves four questions. Most go wrong by answering them in the wrong sequence, starting with the data that happens to be available rather than the decision that needs making.
Why comes first, and it is the decision test above. Everything downstream is scoped by it.
What you benchmark falls into six domains. Very few engagements need all six. The decision determines which two or three carry the weight.
The six domains a benchmarking exercise can cover, and what sits inside each.
| Domain | Covers |
|---|---|
| Performance | Growth, profitability, cash flow, capital efficiency |
| Operational | Productivity, cost, capacity, cycle times, service levels |
| Commercial | Sales, marketing, customers, pricing, unit economics |
| Functional | Finance, HR, technology, procurement, operations |
| Strategic | Business models, products, markets, competitive positioning |
| Capability and maturity | Finance, digital, data, governance, management systems |
What you benchmark against is a separate choice, and an underrated one. The same metric answers a different question under each comparison base: internal unit against unit, peer, industry, top performer, cohort, lifecycle stage, historical, budget, or maturity level.
This choice does more work than people expect. A gross margin at the 40th percentile against peers but improving three years running is a different business from the same number in decline. Position without direction is half a finding, and we show both as a matter of course.
How runs from straightforward ratio comparison through percentiles, indexing, gap analysis, driver trees, productivity curves and maturity models. Method should follow from the decision, and not, as commonly happens, from whatever the available data permits.
The practical discipline here is resisting scope.
A benchmarking pack covering all six domains at equal depth is a catalogue, and catalogues do not get acted on.
The same metric means different things in different businesses
This is where most benchmarking quietly goes wrong, and it is rarely a data problem.
A distribution business assessed on margin looks weak against almost any comparison you can construct. It is supposed to. The model earns its return through turnover, not margin. A 4% margin at six turns is a perfectly good business, and ROCE is the only measure that reads it correctly. Benchmark it on margin and you will produce an accurate report recommending the wrong thing.
The same trap appears everywhere. A marketplace benchmarked on GMV multiples flatters itself by an order of magnitude. A project business judged on a single year's revenue growth misses that the order book already determined the next two years. A SaaS company assessed on EBITDA at ₹30 Cr ARR is being measured against a standard its model does not yet owe.
What determines the right frame is the economic engine, not the industry. Two companies in unrelated sectors that both convert billable hours into revenue have far more in common analytically than two software companies with different revenue models. We work with seven engines.
The seven economic engines, the metrics that carry a diagnosis in each, and the constraint that binds.
| Archetype | Metrics that carry the diagnosis | Binding constraint |
|---|---|---|
| Asset-heavy manufacturing | Capacity utilisation, cost per unit, yield, fixed asset turnover, ROCE | Capacity |
| People-based services | Utilisation, realised bill rate, pyramid ratio, revenue per FTE, bench cost | Billable capacity |
| Subscription and recurring | NRR, retention, CAC payback, gross margin, ARR per FTE | Retention |
| Transactional consumer and D2C | Contribution after marketing, repeat rate, AOV, throughput per store or order | Unit contribution |
| Distribution and trading | Inventory turns, GMROI, cash conversion cycle, ROCE via turnover | Working capital |
| Marketplace and platform | Take rate, net revenue against GMV, liquidity, contribution per transaction | Two-sided balance |
| Project and contract | Order book, book-to-bill, WIP, billing against collection, margin at completion | Execution cycle |
The binding constraint column is the useful one. It tells you which metric to look at first and which gap is worth spending money to close. In a manufacturing business, a working capital gap is a finance problem. In a distribution business, it is the problem.
Some companies genuinely span two engines: a manufacturer building a services annuity, or a product company adding subscription. Benchmark the segments separately and consolidate only at the capital returns layer. Blending them produces a company that resembles nothing in any peer set.
Startups and MSMEs are not archetypes
Both get treated as categories of business. Neither is. A startup can run any of the seven engines, and so can an MSME. What they change is not which metrics are relevant but what is achievable: the quality of the company's own data, and whether comparable external data exists at all.
For startups, the shift is from performance to trajectory and efficiency.
Peer comparison largely stops working, because genuine peers are private, young and thinly filed. Cohort and lifecycle comparison replaces it: how companies at this stage, in this model, at this scale typically perform. It is a weaker claim than peer comparison, and it should be framed as one.
replaces profitability as the headline. , and cohort behaviour carry the diagnosis, while EBITDA margin at this stage says almost nothing. Capital efficiency becomes central: multiple, revenue per rupee raised, growth delivered against consumed.
And the cohort view is not optional. Blended, company-level unit economics routinely conceal that recent cohorts are being acquired at materially worse economics than earlier ones. That is exactly the thing an investor will find in diligence if the company has not found it first.
For MSMEs, the problem is harder, and more honest treatment is required.
Filed accounts are frequently tax-optimised rather than economically representative. Promoter remuneration, family expenses, rent to related entities and cash transactions distort the P&L in both directions. Abridged filings mean peer data is often too thin to build a credible set at all. And MIS is usually absent: there is no segment margin, no cost to serve, no customer profitability, so these have to be constructed before anything can be compared.
The sequence is therefore different. Normalise to real economics first: strip or add back promoter-related items, restate related-party transactions at arm's length, establish what the business actually earns. Then benchmark internally and historically, where the data is at least consistently prepared. Then use maturity benchmarking, which needs no external comparables and usually surfaces more actionable gaps than any ratio comparison would. Our own version of that ladder is the five levels of information maturity.
Peer benchmarking for an MSME is worth attempting only where a credible comparable set genuinely exists. Where it does not, the right answer is to say so, rather than to publish a distribution built on eight unreliable filings.
What makes a benchmark survive scrutiny
A benchmark gets forwarded. To a board, a lender, an investor's analyst: someone who will check three numbers and question the peer set. Three things determine whether it holds.
The peer set. A peer group built on industry classification alone is unusable in India. NIC codes are self-declared at incorporation, rarely updated and routinely wrong. A services business and a trading business can sit under the same code, and diversified groups file under whatever the holding entity was originally registered as. Any comparison built on that filter produces a wide, meaningless distribution.
We screen on five filters instead: business model, size band, sector definition built from what the company actually does, ownership and capital structure, and growth stage. Then we validate every survivor manually against its filing and its website. It is slow, it is where a third of the effort in our competitor analysis and benchmarking work goes, and there is no shortcut that survives being questioned.
Normalisation. Raw filing data is not comparable across companies, and this is where published benchmarking most often breaks without anyone noticing. Some peers have adopted Ind AS 116 and some have not, which moves rent between operating cost and depreciation. That creates EBITDA differences of several hundred basis points with no economic meaning whatsoever. Reported EBITDA needs rebuilding from the schedules. Other income has to come out of operating performance. Related-party revenue needs identifying from the RPT schedule. Standalone and consolidated figures cannot be mixed within one set. The fourth piece in this series sets out the work that turns MCA filings into comparable numbers.
None of this is visible in the output, which is precisely why it gets skipped.
Stated limits. Sample size governs what can honestly be published. Percentiles need twenty or more comparable entities; ten to nineteen supports ranges with the count disclosed; below ten, only individual comparisons are defensible. Private filings run nine to eighteen months behind, so client data and peer data frequently describe different periods, and the report should say so. Indian private transaction values are often undisclosed or loosely reported, which makes transaction multiples indicative ranges rather than central estimates.
A visible limitations section reads as confidence, not weakness. It is also the cheapest insurance available against the first sophisticated reader who checks the numbers.
A gap is a question, not a finding
Sitting at the 30th percentile on EBITDA margin tells you where you are. It does not tell you why, and the why is rarely one thing. A margin gap resolves into pricing, mix, input cost or how costs scale with revenue, and each implies an entirely different response. A benchmark that stops at the percentile hands management a number they cannot act on, which is how most of them end up unread.
The decomposition is the work, and it runs on a small number of measures rather than forty. Which ones those are is the subject of the next piece, on the metrics that carry a diagnosis.
So is the next distinction, which matters more than it gets credit for: not every gap is closeable.
Top-quartile performance often rests on structural advantages the company does not have: scale, capital access, vintage, a different customer mix, a promoter who funded the business without debt. Establishing that a gap is structural is a legitimate and genuinely useful finding. It stops a management team chasing a target that was never available to them, and redirects the effort to gaps that are.
Which returns to where this started. The useful output of benchmarking is not a position. It is a short list of gaps worth closing, an explanation of what drives each, and a statement of what the company should do differently.
If the exercise cannot produce that, it is worth asking at the outset whether it should be run at all.
How useful was this article?
One tap. It tells us what to write more of.
The next one
Get what we publish next, by email.
Working notes on raising, borrowing, protecting, growing and structuring capital in India. One email a week at most, and you can leave any time.
We use your address only to send this. See our privacy policy.
We store your address to send you these emails and nothing else. See our privacy policy.
Related reading
