Valuation Multiple Benchmarking: Building a Comparable-Company Model for Pre-IPO Companies

Comparable-company analysis (often shortened to "comps") anchors a private company's valuation to what the market is currently paying for similar public companies. It's one of the core methods used to set a pre-IPO valuation range, alongside discounted cash flow models and precedent transactions. This piece walks through how that model actually gets built, where the real data comes from, and why the final number is always a range, never a single precise figure.

One thing worth saying upfront, since it matters for how you read everything that follows: comps set a valuation range; they don't set the final IPO price on their own. The final offer price also depends on book-building demand, market conditions at the time of listing, and how much the issuer wants to raise.  

Why Comps Matter, and What They Actually Measure

Comps work by looking at what public markets are currently paying for similar businesses, using metrics like EV/Revenue (Enterprise Value divided by Revenue, a common way to price companies that aren't yet consistently profitable). It's worth being precise here: this isn't purely "current" data.

Most comps use NTM (Next Twelve Months) estimates, meaning the multiple is applied against a forecast, not a trailing number. So comps replace one kind of forecasting uncertainty (a full DCF model) with a narrower one (analyst consensus estimates for the next year), rather than eliminating forecasting from the process entirely.

Comps do four specific jobs in a pre-IPO context:

• They anchor a valuation to observable market pricing rather than internal projections alone
• Underwriters use them to build the initial indicative price range that goes on the cover of the S-1
• They give both the company and investors a shared, data-backed starting point heading into book building.
• They highlight where a company is stronger or weaker than its public peers, which feeds directly into whether it deserves a premium or a discount to the peer median

Choosing a Peer Group That Actually Holds Up

The hardest and most consequential part of a comps analysis is picking the right peer set. A peer group that looks similar on the surface but differs in business model will produce a valuation that's confidently wrong.

The standard filters analysts use go well beyond industry labels:

• Business model and revenue mix. Subscription revenue, transaction fees, and advertising-driven revenue all get priced differently, even within the same broad sector
• Revenue scale. A $50 million ARR company and a $2 billion ARR company in the same niche rarely trade on comparable multiples
• Growth rate. Companies growing 40% a year are not comparable to companies growing 8% a year, even in the same vertical
• Gross margin and net revenue retention (NRR, the percentage of recurring revenue retained and expanded from existing customers year over year). These two metrics are now treated as the primary driver of SaaS multiple dispersion, with 120%+ NRR companies commanding a structural premium over slower-retention peers at the same growth rate
• Profitability profile and Rule of 40 score (growth rate plus profit margin, explained below)
• Capital intensity, since a capex-heavy AI infrastructure business and an asset-light application layer business shouldn't share a multiple even if both are labeled "AI."
• Public float and liquidity, since thinly traded public comps can distort a median

For a data infrastructure company specifically, the right comparison set looks like Snowflake, Datadog, MongoDB, Confluent, Elastic, and Cloudflare, not a mix of legacy enterprise software names like Adobe or Salesforce, whose growth profiles, revenue mix, and customer base sit in a different category entirely.

Why the Sector Multiples Table Needs a Methodology Label

This is the part of any comps article that most often goes wrong, and it's worth being direct about why. Three respected public-market data providers, tracking overlapping but not identical sets of companies, published noticeably different median SaaS multiples within weeks of each other in 2026:

1. Aventis Advisors reported a median of 3.4x EV/Revenue as of March 2026.
2. While Multiples.vc's public comps dashboard, dated July 4, 2026, groups horizontal SaaS, vertical SaaS, and infrastructure SaaS into separate categories rather than a single blended number.
3. SaaS Capital's index put the public median closer to 5.5x to 6-7x ARR entering 2026.

None of these sources are wrong. They disagree because they measure different things: different constituent companies, mean versus median, ARR versus NTM revenue, and different index rebalancing dates. This is precisely why any sector multiple you use in a valuation model needs its source and date attached and why a single "SaaS multiple is 3.4x" statement, without that context, should be treated with caution regardless of where you read it.

Pulling Multiples.vc's dashboard and Aventis Advisors' report on the same day shows just how wide that gap actually gets in practice, which is exactly why every multiple in a real model needs its source and date attached rather than dropped in as a bare number.

A few directional patterns do hold up across sources. Public SaaS infrastructure names, the category CloudSync in the worked example below belongs to, tend to trade at the high end of the software universe, driven by the AI data buildout. Growth remains the single strongest driver of multiple dispersions within any SaaS cohort, thoughthe market has shifted meaningfully toward rewarding profitability alongside growth compared to the 2020 to 2021 period.

Adjusting the Raw Multiple for Growth and Margin

The growth-adjusted multiple answers a simple question: Is a fast grower actually a good deal, or just expensive in a different way? The formula is straightforward:

If a company trades at 6.5x revenue while growing 10%, that's 0.65x paid per percentage point of growth. A company trading at 12x while growing 22% works out to roughly 0.55x per point, cheaper on a growth-adjusted basis despite the higher headline multiple.

The Rule of 40 (growth rate percentage plus profit margin percentage should sum to 40 or higher) is widely used as a quick health check, but the premium attached to clearing it isn't a fixed number you can apply universally. The specific 20% to 30% valuation premium sometimes cited for Rule of 40 outperformers could not be independently confirmed against a single reliable source and should be treated as directional rather than precise. What's better supported is the pattern itself: companies scoring above 50 on the Rule of 40 with strong retention are the ones consistently commanding the top end of any given multiple range.

EV/EBITDA-R&D is worth flagging clearly for what it is: an analyst adjustment used to look past heavy research spending in fields like biotech or deep tech, not a standardized GAAP metric. Different analysts apply it differently, so any model using it should say so explicitly rather than presenting it as a settled industry standard.

The Liquidity Discount and What the Research Actually Shows

This is one area where the underlying research is more nuanced than a single flat percentage suggests. Two separate bodies of empirical work exist here, and they point to meaningfully different numbers. Restricted stock studies, which compare otherwise identical shares with and without trading restrictions, tend to show average discounts in the 20% to 35% range.

Separately, pre-IPO transaction studies, which compare actual private sale prices against the subsequent IPO price, have historically shown considerably larger discounts, often in the 40% to 60% range, reflecting both the cost of illiquidity and the fact that IPO pricing itself often represents a step up in value, not purely a liquidity premium.

That distinction matters. Applying a flat 20% haircut because it "sounds reasonable" understates what the actual pre-IPO discount literature shows in many cases. Any discount applied in a real model should specify which body of research it's drawing from and why.

A Worked Example: Valuing CloudSync, Properly Scoped

This is an illustrative, hypothetical example built to show the mechanics of the model, not a real company or real market data.

CloudSync is a hypothetical data infrastructure software company preparing for a late-2026 IPO, growing revenue 28% year over year with a 10% EBITDA margin.

Step 1: Build the right peer set. For a data infrastructure business, the relevant peer set is Snowflake, Datadog, MongoDB, Confluent, Elastic, and Cloudflare, companies that share CloudSync's revenue model and customer base, not general enterprise software names with a different growth and margin profile.

Step 2: Use the median, not a single average, and show a range. Because multiples vary widely even within a tight peer set, a defensible model reports a median EV/Revenue multiple alongside the 25th and 75th percentile ranges for that peer set, rather than a single blended average that can be skewed by one outlier.

Step 3: Apply the growth adjustment, using the peer group's median price paid per percentage point of growth, applied to CloudSync's 28% growth rate, to arrive at a base multiple range rather than one number.

Step 4: Apply an appropriately sourced discount, choosing between the restricted stock range (20% to 35%) or the pre-IPO transaction range (40% to 60%) based on which body of research best matches CloudSync's actual situation, disclosed explicitly rather than picked arbitrarily.

Step 5: Convert Enterprise Value to Equity Value correctly. Enterprise Value represents the value of the core business operations. To get to what the stock itself is worth:

where net debt equals total debt minus cash. If a company holds more cash than debt, net debt is negative, and equity value ends up higher than enterprise value.

Step 6: Divide by fully diluted shares, not just shares being sold in the IPO. This is a common and consequential error. The per-share price calculation must use the total fully diluted share count, meaning common shares plus the shares represented by outstanding options, RSUs (restricted stock units), warrants, and any convertible instruments like SAFEs, not just the shares being newly issued in the offering. Using only the IPO share count instead of the fully diluted count overstates the implied per-share price.

The output of this process is a valuation range, not a single confident number. That range then becomes the starting point for the S-1's indicative price range, which book-building either confirms, narrows, or moves entirely once real institutional demand comes in.

Comps Are One Input, Not the Whole Model

Comparable-company analysis is a widely used, well-established method, but it's one input among several that serious valuation work triangulates together: a discounted cash flow model, precedent transaction analysis, recent pre-IPO secondary market trades, and the company's own last private funding round. Relying on comps alone, especially a single unsourced sector multiple, produces a number that looks precise while resting on far less certainty than the decimal point suggests.

Frequently Asked Questions

If public comps trade at 6x revenue, does that mean a pre-IPO company in the same sector should also be modeled at 6x before any adjustment?

No. The public multiple is a starting reference point, not a direct assignment. It needs to be adjusted for the private company's specific growth rate, margin profile, and peer-relative positioning before it's a usable number, and then a separate liquidity discount gets applied on top of that adjusted figure.

Why do different valuation sources report such different sector multiples for what looks like the same industry?

Because "SaaS" or "AI" isn't a single measurable index. Different providers track different company sets, use mean versus median, apply different revenue bases (ARR versus NTM revenue), and rebalance on different schedules. The disagreement is a methodology difference, not an error, which is exactly why source and date matter more than the number itself.

Does a company's fully diluted share count change meaningfully between the S-1 filing and the actual IPO pricing?

It can. Additional options or RSUs granted between filing and pricing, along with any last-minute convertible note conversions, can shift the fully diluted count. That's part of why the final share count used in an actual pricing model gets confirmed again right before the offer price is set, not locked in from an earlier draft of the filing.

Is a growth-adjusted multiple a standardized metric analysts always calculate the same way?

Not entirely. The core formula, multiple divided by growth rate, is standard, but some analysts use trailing growth while others use forward growth, and some normalize the growth rate as a whole number while others use a decimal. Any model using this metric should state which convention it's using, since the two approaches produce very different-looking numbers for the same underlying data.

Does a company with a high Rule of 40 score automatically get priced above its peer median?

Not automatically. The rule of 40 is a useful screening tool, but the actual premium it commands depends heavily on which combination of growth and margin produced the score. A company hitting 45 through high growth and thin margin is viewed differently by public market investors than one hitting the same 45 through modest growth and strong margin, even though the headline score is identical.