How to Read a Tech S-1 Like an Operator

If you want to know how to read a tech S-1 like an operator, the answer is to ignore the story and go straight to the parts management cannot spin. The prospectus narrative is written by bankers to sell an offering. The numbers underneath it are written under the threat of securities liability. Those are not the same document, and the skill is knowing which one to read first.

The banker narrative leads with total addressable market, a hockey-stick chart, and a mission statement. The operator’s read skips all of it. It goes to revenue mix, net revenue retention, stock-based compensation, and customer concentration, in that order, because those are the lines that reveal whether the unit economics actually work.

This piece lays out how to read a tech S-1 like an operator using real, recent filings: Reddit, ServiceTitan, Rubrik, CoreWeave, Klaviyo, Instacart, and Mobileye. Every figure ties to a specific Form S-1. The framing is analytical: how to think about a filing, not what to do about any stock.

Key takeaways

• Order is the skill. Read revenue mix, then net revenue retention, then SBC, then concentration, then the specific risk factors. Management spins prose; it cannot spin these disclosures without consequence.
• NRR is the unit-economics tell. ServiceTitan disclosed 110%+ net dollar retention across ten quarters, Rubrik 133%, Klaviyo 119% (their Forms S-1). Growth without retention is bought, not earned.
• SBC is the true cost of labor. Rubrik reported $913.9M of stock-based compensation in FY2025 versus $5.7M the prior period (Rubrik Form S-1, FY2024). The footnote, not the headline, tells you what employees extracted before public holders.
• Concentration breaks the narrative. CoreWeave drew 62% of 2024 revenue from Microsoft; Klaviyo drew 77.5% of ARR from Shopify-platform customers (their Forms S-1). Platform dependence is the risk the TAM slide hides.
• Specific beats boilerplate. A risk factor that could apply to any company is noise. One that could only be written by this company is signal.

The S-1 Operator’s Read: the read-order scorecard

The clearest way to read a filing is to fix the order before you open it. Call this the S-1 Operator’s Read: a numbered read-order scorecard that tells you what to open first, what that section reveals, and the red flag to watch in it. The point of naming it is reuse. You can run any tech S-1 through the same seven rows and get a consistent operator-grade triage in under an hour.

This scorecard is an original analytical asset. The figures used to illustrate each row are sourced to the Forms S-1 cited throughout this piece.

#	Open this first	What it reveals	The red flag
1	Revenue mix and growth rate (MD&A)	Where the money actually comes from, and which stream is growing fastest	Low-margin services or usage revenue outgrowing high-margin subscription
2	Net revenue / dollar-based retention	Whether existing customers expand or contract	NRR omitted, buried in supplemental metrics, or trending down
3	Stock-based compensation (income statement + footnotes)	The true, cash-adjusted cost of labor and dilution	SBC rising as a share of revenue, or a large unrecognized balance
4	Customer concentration (risk factors + MD&A)	How diversified the revenue base really is	A single customer or platform above ~20% of revenue
5	Platform / ecosystem dependence	Whether a partner controls the funnel	One ecosystem driving the majority of new customers
6	Risk factors (specific vs. boilerplate)	The risks management is actually worried about	Generic language; the specific risk is the one that matters
7	Use of proceeds and selling stockholders	Whether the raise funds growth or cashes out insiders	Proceeds heavily weighted to secondary sales

Read in this order, the filing triages itself. Rows 1 through 3 tell you whether the business model works. Rows 4 and 5 tell you what could break it that management does not control. Rows 6 and 7 tell you what management is worried about and who is selling.

The banker narrative inverts this. It opens with vision and addressable market, then routes the hard numbers into a “Key Metrics” box you are meant to skim. The operator-read reverses the priority on purpose.

What is the operator’s read order, and why does order matter?

The operator’s read order goes straight to the disclosures management is required to make and cannot dress up: revenue mix and growth, retention, SBC, and concentration, in that sequence. Order matters because it filters signal from narrative before the narrative can frame your judgment.

A banker leads with the story so that every number you see afterward is interpreted through it. The operator does the opposite. You anchor on the unit economics first, then read the prose to see whether the story is consistent with the numbers or papering over them. The same surface-over-story discipline runs through why gross margin is destiny in SaaS: the structural line decides the outcome long before the narrative does.

Revenue mix: the first line item that matters

Open the MD&A revenue table before anything else. Revenue mix tells you which engine drives the business and which one drives the margin, and the two are often not the same.

ServiceTitan’s Form S-1 (twelve months ended July 31, 2024) disclosed roughly 70% subscription revenue and about 25% usage revenue, with the remainder in professional services, on $685M of ARR growing 25% year over year. That mix is the tell. Subscription revenue carries the highest contribution margin; if the usage or services lines grew faster than subscription, blended gross margin would compress even as the top line accelerated.

Contrast that with CoreWeave’s Form S-1 (FY2024), which reported $1.9 billion of revenue, up from $229M in 2023, a 737% increase. The growth is spectacular, but the mix is entirely GPU-infrastructure revenue, the highest-COGS line in the comparison set. A 737% growth rate and a thin, capital-intensive margin profile are both true at once. The growth rate alone tells you nothing about whether the model converts to profit. The cloud-economics version of that tension is mapped in AWS margin pressure and the cloud reset.

The operator question is never “how fast did revenue grow.” It is “which revenue grew, and what does that do to the margin trend.” Read the historical gross-margin line against the mix shift, not the headline CAGR.

Net revenue retention: the unit-economics tell

Net revenue retention is the single most important number in a SaaS S-1, and it deserves the second read because it reveals whether existing customers expand or contract. Growth can be rented from new logos at negative-payback CAC. Retention cannot be faked, because it measures the same cohort over time.

The recent filings give a clean benchmark range:

Company (Form S-1)	Retention disclosed	Period
Rubrik	133% subscription dollar-based NRR	as of Jan 31, 2024
Klaviyo	119% NRR (114% in Q1 2024)	June 30, 2023
ServiceTitan	110%+ net dollar retention	across ten quarters

Sources: Rubrik, Inc. Form S-1 (FY ended Jan 31, 2024); Klaviyo, Inc. Form S-1 (period ended June 30, 2023); ServiceTitan, Inc. Form S-1 (twelve months ended July 31, 2024).

All three are above 100%, which means each cohort of customers pays more over time even after accounting for churn. That is the proof that the land-and-expand motion works without buying growth. ServiceTitan’s revenue per active customer rising from roughly $72K to $78K (2023 to 2024, per its Form S-1) is the same expansion visible at the account level. The distinction between the retention number that includes expansion and the one that does not is the whole subject of gross retention vs net retention in SaaS IPOs, and reading the wrong one is a common error.

The red flag is absence. If an IPO candidate does not disclose NRR, or routes it into a supplemental deck instead of the prospectus, treat that as a signal that the cohort behavior is not flattering. Reddit’s Form S-1 (FY2023), for instance, leans on revenue growth ($804M in 2023, with Q1 2024 revenue of $243M up 48% year over year) but does not present a SaaS-style NRR, because its consumer-advertising model is measured on different cohort mechanics. The lesson is not that Reddit is weaker; it is that you read the retention metric the business model actually generates, and you notice when the metric that should be there is missing.

How do you spot a management red flag in SBC disclosure?

You watch three things: stock-based compensation rising as a share of revenue, a large unrecognized SBC balance in the footnotes, and an SBC spike tied to the IPO event rather than operating growth. Stock-based compensation is where the income statement and the cash reality diverge most, so it earns the third read. SBC is a real cost of running the business that does not show up as cash, which means a company can post a thin GAAP loss while transferring enormous value to employees and founders ahead of public holders.

Rubrik is the cleanest case. Its Form S-1 (FY ended Jan 31, 2024) shows stock-based compensation of $913.9M in FY2025 against just $5.7M in the prior comparable period, with $125.7M of unrecognized SBC as of July 31, 2024. That jump is not the business scaling. It is restricted awards vesting on the IPO event, an enormous one-time recognition that floods the post-offering income statement.

Instacart’s Form S-1 shows the same pattern at consumer scale: roughly $2.6 billion of stock-based compensation in its first quarter as a public company (Q3 2023), driven by RSU vesting tied to the offering, against $764M of quarterly revenue. The SBC charge dwarfed the revenue line for that quarter.

The three indicators to watch:

1. SBC as a share of revenue, trending up. Klaviyo ran SBC at about 13% of revenue (Form S-1), a controlled level. A figure climbing year over year signals dilution outrunning growth.
2. Unrecognized SBC in the footnotes. A large balance is future dilution already promised. It tells you what the next several years of the income statement will absorb.
3. An IPO-event spike. A one-time jump tied to vesting on the offering, like Rubrik’s and Instacart’s, is the moment founders and employees realize value before public shareholders see it flow through earnings.

SBC is the line bankers most want you to “adjust out” with a non-GAAP metric. The operator reads it in, because the dilution is real even when the cash is not.

What is the difference between customer concentration and platform concentration?

Customer concentration is dependence on a few accounts; platform concentration is dependence on one ecosystem you do not control. ServiceTitan’s top 10 customers were about 10% of revenue and Klaviyo’s were roughly 1.4% of ARR (their Forms S-1), but Klaviyo still drew 77.5% of ARR through Shopify. The second flavor is the dangerous one, and conflating them is a common mistake.

The account-level view comes first. ServiceTitan’s top 10 customers were about 10% of revenue (Form S-1), and Klaviyo’s top 10 were roughly 1.4% of ARR (Form S-1). Both are healthy: no single account can sink the business.

Platform concentration is dependence on one ecosystem you do not control, and it is the more dangerous flavor:

Company (Form S-1)	Concentration	Type
CoreWeave	Microsoft = 62% of 2024 revenue; top 2 = 77%	Customer + platform
Klaviyo	77.5% of ARR from Shopify-platform customers	Platform
Mobileye	Top 8 OEMs = 76% of revenue; top 3 Tier 1 suppliers = 73%	Channel + customer

Sources: CoreWeave Form S-1 (FY2024); Klaviyo Form S-1 (period ended June 30, 2023); Mobileye Global Inc. Form S-1 (FY2022 and FY2023).

CoreWeave’s 62% Microsoft dependence means a single counterparty’s GPU-supply decisions can reprice the business. Klaviyo’s 77.5% Shopify-derived ARR means Shopify’s continued recommendation, and its choice not to build a competing CRM, underwrites most of the growth. Mobileye’s reliance on a handful of automakers and Tier 1 suppliers (ZF, Valeo, Aptiv) means design-win cycles at a few OEMs drive the entire revenue line.

The pattern of monetizing through an owned ecosystem is the inverse of these dependencies; it is what platform owners build deliberately, as dissected in Apple Services as the margin engine inside iPhone. The S-1 reader’s job is to spot which side of that relationship the issuer sits on. If the company depends on someone else’s flywheel, the concentration line is the truth and the TAM slide is the fiction.

Risk factors: which ones are boilerplate, which are specific

Risk-factor sections run 30 to 60 pages and most of it is legal boilerplate written to be unreadable. The operator filters for one property: specificity. A boilerplate risk could apply to a hundred companies. A structural risk could only be written by this company about this business.

Boilerplate, skip it: “we face intense competition,” “we may fail to retain key personnel,” “we rely on third-party platforms,” “our results may fluctuate.” These are liability shields, not information.

Structural, read it twice:

• Klaviyo: the risk that Shopify changes terms or builds a competing email and CRM product. That is the 77.5%-of-ARR dependency restated as a contingency.
• CoreWeave: the risk that Microsoft shifts GPU capacity to first-party infrastructure. That is the 62% concentration as a forward-looking threat.
• Mobileye: the risk of losing OEM design wins as automakers consolidate chip suppliers. That is the 76% top-8 dependency expressed as a scenario.
• Reddit: the risk that API pricing or policy changes drive a moderator and creator exodus, which speaks directly to the durability of its content supply.

The test is mechanical. If you can imagine the exact sentence appearing verbatim in a dozen unrelated S-1s, ignore it. If the sentence names a specific counterparty, dependency, or mechanism unique to this issuer, it is the company telling you where the model is fragile.

The bear case: what the operator’s read misses

The strongest argument against this read order is that it over-trusts disclosed metrics and under-weights the qualitative story, and the skeptics get something real here.

Disclosed metrics are management-chosen definitions. “Net revenue retention” is not a GAAP term, so companies define the cohort, the window, and the inclusions themselves. A 119% NRR computed on a favorably gerrymandered cohort is not comparable to a 119% computed conservatively, and the S-1 does not always give you enough to normalize them. The number can be technically true and still flatter the business. The same caution applies to ARR, “active customers,” and adjusted gross margin, all of which are issuer-defined.

The bear case also notes that the operator-read can talk a reader out of a genuinely great business. A high-concentration filing looks fragile on the scorecard, but concentration is also how category-defining companies start. CoreWeave’s Microsoft dependence is a risk and the reason the revenue exists at all. Reading concentration purely as a red flag would have flagged many companies that went on to diversify successfully from exactly that position. Concentration is a risk and a growth vector at the same time, and the scorecard does not weigh the second.

Here is the honest weighing. The operator’s read is the right triage because it anchors on what management is legally constrained to disclose, but it is a filter, not a verdict. It tells you where to look harder. The qualitative work, reading the definitions in the footnotes, judging whether a concentration is a launchpad or a leash, sizing whether a structural risk is near-term or distant, is the part no scorecard automates. Use the read order to find the three things that matter in a 300-page document, then do the slow work on those three.

What operators should take from this

If you read filings as a founder, operator, or analyst, the transferable discipline is not “memorize the benchmarks.” It is the order of operations and the questions each step forces. Here is the playbook, six concrete moves you can run on the next S-1 that drops.

1. Run the S-1 Operator’s Read top to bottom before reading the narrative. Open revenue mix, retention, SBC, and concentration first. Form your thesis on the numbers, then read the prose to see whether it confirms or contradicts what you already found.
2. Normalize the issuer-defined metrics. Find the definition of NRR, ARR, and adjusted gross margin in the footnotes before you compare them across companies. A metric is only as good as its definition, and the definition is never in the headline.
3. Read SBC into the model, never out of it. Treat stock-based compensation as a real cost and track the unrecognized balance as future dilution. The non-GAAP “adjusted” figure is the one management wants you to use; that is the reason to distrust it.
4. Separate customer concentration from platform concentration. A diversified account base with a single-ecosystem dependency is still fragile. Ask which partner could reprice or build the feature themselves, and how fast.
5. Filter risk factors for specificity. Highlight only the sentences that name a counterparty, a dependency, or a mechanism unique to this issuer. Those three or four sentences are worth more than the other fifty pages.
6. Cross-check against the next filing. An S-1 is a snapshot. When the company files its first 10-Q and 10-K, re-run the same rows and watch the deltas: did NRR hold, did SBC normalize after the IPO spike, did concentration improve. The trend across filings is where the truth lives.

That sixth move is the one most readers skip. The S-1 is the pitch; the subsequent filings are the proof. The mechanics of how usage versus seat pricing changes the retention math you are checking in step 1 are worked through in usage-based pricing vs seat-based pricing, and the same infrastructure cost dynamics that move CoreWeave-style margins are mapped in the AI infrastructure market map.

Methodology: how to triage an S-1 in under an hour

• Inputs: the MD&A revenue table, the key-metrics section, the income statement and SBC footnote, the customer-concentration disclosure, and the risk-factor headings. Five sections, in the scorecard order.
• Assumptions: that disclosed retention and ARR figures use a stable, issuer-defined methodology across periods, and that the most recent period is representative. Both are checked in the footnotes, not assumed.
• Sensitivity: a single point of NRR, a few points of mix shift toward low-margin revenue, or a 10-point move in top-customer concentration each change the read materially. Treat each metric as a range and a trend, not a fixed point.
• What this misses: issuer-defined metrics can be gamed within the rules, segment-level margin is rarely broken out, and a snapshot S-1 cannot show whether a concentration is diversifying. The cross-check against later filings is the only fix.

Where this read order is vulnerable to gaming

A method this clean deserves its own counterexamples, because management knows the operator-read too and can stage the filing around it.

Metrics can be defined favorably. As covered in the bear case, NRR, ARR, and “adjusted” gross margin are issuer-defined. A company can choose the cohort window, the customer definition, and the COGS classification that flatter the number. The read order tells you which metric to find; it does not guarantee the metric is computed the way you would compute it.

Timing can be engineered. A company can pull forward a strong quarter into the S-1 window or push a soft one out, so the trailing figures look better than the run-rate. The single-snapshot nature of an S-1 is exactly why the cross-check against the first 10-Q matters: the post-IPO quarters are harder to stage.

Concentration can be temporarily masked. An issuer can sign a late, large customer to dilute a concentration percentage just before filing, or structure a reseller relationship to spread revenue across nominal accounts that share an ultimate parent. The disclosed top-10 figure can understate the true dependency.

Risk factors can hide the real risk in plain sight. The most material risk is sometimes written in the same flat tone as the boilerplate, precisely so it does not stand out. Specificity is the filter, but a careful drafter can make a structural risk read like a generic one.

None of this breaks the method. It bounds it. The S-1 Operator’s Read is the fastest way to find the three things in a filing that decide the business. What it cannot do is the slow, skeptical, footnote-level work on those three. That part is still the operator’s, and it is the part that does not scale.

---

Analysis, not investment advice. Figures are drawn from the public SEC filings cited inline by company and form type (Forms S-1 for Reddit, ServiceTitan, Rubrik, CoreWeave, Klaviyo, Instacart, and Mobileye). Frameworks here are for understanding how to read business filings and tradeoffs, not for making buy or sell decisions.

Want the full toolkit for reading filings like this, the read-order scorecard, the retention and SBC worksheets, and the concentration checklist used above? It’s in the Tech Business Analysis Playbook.