Product-Led vs Sales-Led Growth: The Economics

Product-led vs sales-led growth is not a philosophical split. It is a cost structure matched to deal size, and the filings settle the argument faster than any conference talk does.

Salesforce spent 34.5% of revenue on marketing and sales in fiscal 2026 (Salesforce Form 10-K, FY ended January 31, 2026). Atlassian, the canonical product-led company, spent 21.8% in fiscal 2025 (Atlassian Form 10-K, FY ended June 30, 2025). That looks like a clean verdict until HubSpot enters the table. HubSpot discloses average subscription revenue per customer of $11,414 and still spent 44.1% of revenue on sales and marketing (HubSpot Form 10-K, FY2025). More than either.

The motion does not follow from ideology. It follows from price, gross margin, and how many deals one human being can physically close in a year.

This piece reads that question through 16 public filings and two pricing pages retrieved July 28, 2026. Every company figure ties to a specific 10-K and fiscal period. The framing is analytical: how the arithmetic picks the motion, not what to do about any stock.

Key takeaways

• The archetypes are 12.7 points apart, and the archetype breaks immediately. Salesforce 34.5% of revenue on sales and marketing (FY2026 10-K), Atlassian 21.8% (FY2025 10-K), HubSpot 44.1% on an $11,414 average contract (FY2025 10-K). “Product-led” appears at both ends of the range.
• Product-led companies do not spend less. They spend it on engineers. Atlassian ran research and development at 51.2% of revenue in FY2025, more than double its sales and marketing line (Atlassian Form 10-K, FY2025). Datadog employs approximately 3,900 in R&D against approximately 3,600 in sales and marketing (Datadog Form 10-K, FY2025).
• Percent of revenue is the wrong denominator. Twilio looks leanest of the large names at 17.2% of revenue, but on a 48.9% gross margin that is 35.2% of gross profit, worse than Datadog (Twilio and Datadog Forms 10-K, FY2025).
• The same 250 seats is a $23,730 contract or a $525,000 contract. Jira Standard lists at $7.91 per user per month; Salesforce Sales Cloud Enterprise lists at $175 (both retrieved July 28, 2026). That 22.1x gap, not conviction, decides who gets a salesperson.
• There is a measurable dead zone. Benchmarkit’s 2025 benchmarks found $10K to $50K ACV solutions are often more expensive to acquire than $50K to $100K solutions, and that this is not a one-year exception.
• At scale, everyone runs both. Zoom disclosed Enterprise at 60.3% of FY2026 revenue while total sales and marketing fell to 28.5% (Zoom Form 10-K, FY2026). The endgame is a barbell, not a doctrine.

---

Product-led vs sales-led growth is a cost structure, not a philosophy

Strip the vocabulary away and the choice reduces to one question: can a single contract generate enough gross profit to pay for a fraction of a human being, fast enough to matter?

That framing kills most of the debate. A salesperson is a fixed annual cost with a hard ceiling on throughput. No amount of enablement makes one person close 400 enterprise deals a year. So the motion is decided by two numbers the company mostly controls, price and gross margin, against one number it barely controls, the number of deals a human can physically work.

Product-led wins when the product can be understood, adopted, and expanded without a human in the loop, and that is usually only true when the contract is too small to pay for the human anyway. Sales-led wins when the contract is large enough that a person is cheap relative to what the contract yields.

The trap is stating that test in revenue dollars. Revenue is not what pays a salesperson. Gross margin sets how much of each revenue dollar is available to pay for acquisition at all, and margins across public SaaS run from under 50% to nearly 90%. Two companies with identical contract values and identical sales costs can sit on opposite sides of the line purely because one pays a cloud bill and the other does not.

Everything below is that one test, run against filings.

---

What the filings say: sales and marketing across 16 public SaaS companies

The most-cited claim in go-to-market strategy is that product-led companies spend less on acquisition. Here is that claim measured against the most recent Form 10-K available for 16 public SaaS businesses as of July 2026.

Company	Fiscal period	Revenue	Sales and marketing	S&M % of revenue	Gross margin
DigitalOcean	FY2025 (Dec 31, 2025)	$901.4M	$82.4M	9.1%	59.9%
Twilio	FY2025 (Dec 31, 2025)	$5,067.2M	$873.2M	17.2%	48.9%
Atlassian	FY2025 (Jun 30, 2025)	$5,215.3M	$1,134.5M	21.8%	82.8%
Workday	FY2026 (Jan 31, 2026)	$9,552M	$2,616M	27.4%	not separately tagged
Datadog	FY2025 (Dec 31, 2025)	$3,427.2M	$956.4M	27.9%	80.0%
Zoom	FY2026 (Jan 31, 2026)	$4,868.8M	$1,388.3M	28.5%	77.0%
ServiceNow	FY2025 (Dec 31, 2025)	$13,278M	$4,388M	33.0%	77.5%
Salesforce	FY2026 (Jan 31, 2026)	$41,525M	$14,345M	34.5%	77.7%
MongoDB	FY2026 (Jan 31, 2026)	$2,463.8M	$944.4M	38.3%	71.7%
Klaviyo	FY2025 (Dec 31, 2025)	$1,234.0M	$506.2M	41.0%	74.7%
Snowflake	FY2026 (Jan 31, 2026)	$4,683.9M	$2,062.1M	44.0%	67.2%
HubSpot	FY2025 (Dec 31, 2025)	$3,131.3M	$1,379.4M	44.1%	83.8%
Braze	FY2026 (Jan 31, 2026)	$738.2M	$327.0M	44.3%	67.1%
GitLab	FY2026 (Jan 31, 2026)	$955.2M	$434.7M	45.5%	87.4%
Asana	FY2026 (Jan 31, 2026)	$790.8M	$407.0M	51.5%	89.0%
Figma	FY2025 (Dec 31, 2025)	$1,055.8M	$575.5M	54.5%	82.4%

Sources: each issuer’s most recent Annual Report on Form 10-K as labeled, via the SEC XBRL companyconcept API (us-gaap SellingAndMarketingExpense, RevenueFromContractWithCustomerExcludingAssessedTax, GrossProfit), queried July 28, 2026 and cross-checked against the filings.

The range runs from 9.1% to 54.5%, and the names commonly described as product-led occupy both ends. DigitalOcean, a pure self-serve infrastructure business, sits at the floor. Figma, the most-cited modern product-led company, sits at the ceiling. GitLab and Asana, both open-core or freemium-native, spend more of revenue on acquisition than Salesforce does.

The second column of the story is where the money went instead. Atlassian ran research and development at 51.2% of revenue in FY2025, against 21.8% for sales and marketing (Atlassian Form 10-K, FY2025). Datadog ran R&D at 45.2% against 27.9%, and staffed approximately 3,900 people in R&D versus approximately 3,600 in sales and marketing out of approximately 8,100 total (Datadog Form 10-K, FY2025). Salesforce ran R&D at 14.4% of revenue (Salesforce Form 10-K, FY2026).

Company	Fiscal period	R&D % of revenue	S&M % of revenue	R&D to S&M
Atlassian	FY2025	51.2%	21.8%	2.35x
Datadog	FY2025	45.2%	27.9%	1.62x
Salesforce	FY2026	14.4%	34.5%	0.42x
Figma	FY2025	97.5%	54.5%	1.79x

Sources: Atlassian Form 10-K FY2025; Datadog Form 10-K FY2025; Salesforce Form 10-K FY2026; Figma Form 10-K FY2025. Figma’s 97.5% R&D ratio is distorted by post-IPO stock-based compensation and should not be read as a run-rate.

That is the honest version of the product-led claim. The distribution cost did not vanish. It changed departments. Atlassian buys distribution by building a product that distributes itself, and it pays for that in engineering headcount rather than quota-carrying headcount.

---

Why should sales and marketing be measured against gross profit instead of revenue?

Because gross margin decides how much of each revenue dollar is available to pay for acquisition at all. Twilio spent 17.2% of revenue on sales and marketing in FY2025, the leanest of the large names, but on a 48.9% gross margin that is 35.2% of gross profit, worse than Datadog at 34.9%. The revenue denominator flatters low-margin businesses.

Re-rank the same set on gross profit and the table reorders. This is the version that answers whether a rep can be afforded, because a rep is paid out of gross profit, not out of revenue.

Company	Fiscal period	S&M % of revenue	Gross margin	S&M % of gross profit	Rank change
DigitalOcean	FY2025	9.1%	59.9%	15.3%	unchanged (1st)
Atlassian	FY2025	21.8%	82.8%	26.3%	up 1
Datadog	FY2025	27.9%	80.0%	34.9%	up 1
Twilio	FY2025	17.2%	48.9%	35.2%	down 2
Zoom	FY2026	28.5%	77.0%	37.0%	unchanged
ServiceNow	FY2025	33.0%	77.5%	42.6%	unchanged
Salesforce	FY2026	34.5%	77.7%	44.5%	unchanged
GitLab	FY2026	45.5%	87.4%	52.1%	up 5
HubSpot	FY2025	44.1%	83.8%	52.6%	up 2
MongoDB	FY2026	38.3%	71.7%	53.4%	down 2
Klaviyo	FY2025	41.0%	74.7%	54.9%	down 2
Asana	FY2026	51.5%	89.0%	57.8%	up 2
Snowflake	FY2026	44.0%	67.2%	65.5%	down 3
Braze	FY2026	44.3%	67.1%	66.0%	down 2
Figma	FY2025	54.5%	82.4%	66.1%	unchanged (15th)

Computed from us-gaap SellingAndMarketingExpense and us-gaap GrossProfit in each issuer’s most recent Form 10-K, SEC XBRL companyconcept API, queried July 28, 2026. Rank change is measured across the 15 companies shown; Workday is excluded from both rankings because gross profit is not separately tagged in its FY2026 filing.

Three readings fall out of the re-ranking.

Twilio drops two places, from second-leanest to fourth. A communications platform passes through carrier and messaging cost, so nearly half of every revenue dollar is gone before acquisition spend begins. A 17.2% sales-and-marketing line on a 48.9% margin is a heavier acquisition burden than a 27.9% line on an 80.0% margin.

Snowflake and HubSpot look identical on revenue and are 12.9 points apart on gross profit. Snowflake spent 44.0% of revenue and HubSpot 44.1%. On gross profit, Snowflake is 65.5% and HubSpot 52.6%, because Snowflake’s 67.2% margin carries the cloud bill that a consumption business pays on behalf of its customers.

GitLab moves up five places, the largest swing in the table. At 87.4% gross margin, an expensive-looking 45.5% of revenue converts to 52.1% of gross profit, which puts it ahead of MongoDB, Klaviyo, and Snowflake despite spending more of its top line than any of them. High margin is a subsidy for an expensive motion.

The practical implication is the same one that governs CAC payback measured in months of gross profit: the binding constraint is not what you spend, it is what you have left to spend it from.

---

The GTM Fit Test: the decision matrix for product-led vs sales-led growth

Most go-to-market debates fail because they argue about a single variable. The GTM Fit Test is a five-signal decision matrix plus one arithmetic gate. It is designed to be run in twenty minutes, per product segment, and to name the specific failure mode of choosing wrong.

The five signals are the ones that actually move acquisition cost. Score each one, then apply the gate.

Signal	Points product-led	Points sales-led	Failure mode if you choose wrong
Annual contract value	Under roughly $10,000	Above roughly $50,000	Hire reps under $10K ACV and gross profit per deal never covers the quota carrier. Refuse reps above $50K ACV and competitors take the accounts you never called.
Time to first value	Minutes to hours, entirely self-served	Weeks, requires configuration, migration, or integration	Ship self-serve on a product that needs three weeks of setup and you buy signups that never activate. Add a rep to a product that works in five minutes and you insert friction the buyer did not want.
Who the buyer is	The end user, spending under their own discretionary limit	A committee with procurement, security review, and legal	Sell bottom-up into a category where security review is mandatory and deals stall at 90% forever. Sell top-down where the practitioner already adopted you and you pay a rep to close a deal that closed itself.
Product complexity	One job, visible in the interface, no process mapping	Multi-system workflow that must be mapped onto an existing process	Self-serve on a workflow product produces high trial volume and near-zero conversion. Rep-led on a single-job tool produces a cost structure the price cannot carry.
Expansion path	Seats or usage grow without a conversation	Expansion requires renegotiating a contract	Assume automatic expansion where a contract gates it and your net retention model is fiction. Staff account managers where usage expands on its own and you pay for growth you already had.

Scoring. Four or more signals pointing the same way is a clear motion. A two-three or three-two split is not a hybrid strategy, it is a warning that the segment sits in the dead zone described below and needs to be repackaged rather than staffed.

The gate. Signals are directional; the gate is arithmetic. No score overrides the Rep Coverage Ratio, because a matrix cannot pay a salary.

The Test is deliberately per-segment. Almost every company above would score differently for its self-serve tier than for its enterprise tier, which is exactly why the aggregate sales-and-marketing ratio in the first table is such a blunt instrument. Atlassian’s own 10-K makes this explicit:

To land new customers, we’ve engineered a frictionless flywheel with an emphasis on self-service, making it easy to try and get value first and foremost and then virally expand within an organization. This allows our sales force to focus primarily on expanding and deepening strategic relationships with our existing large enterprise customers.

That is one company running the GTM Fit Test twice and getting two different answers.

---

How do I calculate whether my contract value can pay for a salesperson?

Use the Rep Coverage Ratio, the arithmetic gate inside the GTM Fit Test:

RCR = (ACV x gross margin x deals one rep closes per year) / fully loaded annual cost of that rep

The denominator includes the allocated sales development, marketing, and management cost that a quota carrier consumes, not just base and commission. The interpretation bands:

RCR	What it means	What to do
Below 1.0	The rep destroys gross profit in year one	The motion is structurally impossible. Do not hire an account executive; fix activation or raise price.
1.0 to 2.0	Year one roughly breaks even	You are not running a sales model, you are running a retention bet. Gross retention becomes the load-bearing metric.
2.0 and above	The contract comfortably funds the human	Sales-led is viable, and refusing to hire is leaving revenue on the table.

The important structural point is that the denominator is the fixed constraint, not the numerator. A human’s annual deal throughput has a hard ceiling that no tooling removes. You can raise price, and you can raise gross margin. You cannot make one person close 300 enterprise contracts.

Jason Lemkin arrives at the same denominator from the other direction. His stated requirement is to “clear, say, $300k in revenue from that rep to make the math work,” which produces a roughly $3,000 ACV floor at 100 closed deals a year, and $5,000 once overhead and turnover are absorbed (Jason Lemkin, SaaStr, retrieved July 28, 2026; industry rule of thumb, not a filing).

Here is the RCR surface. Rows are contract value, columns are deals closed per rep per year, gross margin held at 80% and rep cost at $300,000.

ACV	Gross profit per deal	6 deals/yr	12 deals/yr	24 deals/yr	48 deals/yr
$5,000	$4,000	0.08	0.16	0.32	0.64
$10,000	$8,000	0.16	0.32	0.64	1.28
$25,000	$20,000	0.40	0.80	1.60	3.20
$50,000	$40,000	0.80	1.60	3.20	6.40
$100,000	$80,000	1.60	3.20	6.40	12.80
$250,000	$200,000	4.00	8.00	16.00	32.00

Illustrative. Gross margin fixed at 80% and fully loaded rep cost fixed at $300,000 per year, both modeling assumptions rather than disclosed figures. Bolded cells mark the physically plausible pairings: high-ACV enterprise deals do not get closed 48 times a year, and $5,000 contracts do not justify a six-deal rep.

Read the bolded diagonal and the shape of the problem is visible. Throughput and contract value trade against each other, and the product of the two has to clear the rep. Everything above the diagonal is a fantasy quota; everything below it is a company paying a person to do arithmetic that does not work.

Methodology: how to run the Rep Coverage Ratio honestly

• Inputs: your own realized ACV by segment (not list price), your reported gross margin, your observed closed-won count per fully ramped rep, and your fully loaded rep cost including allocated sales development, marketing, and management.
• Assumptions: that a ramped rep’s throughput is stable, that gross margin on new business matches blended margin, and that the fully loaded cost captures the true burden. The $300,000 figure used above is an illustrative modeling assumption, corroborated only directionally by the Lemkin rule of thumb, not a sourced or benchmarked number.
• Sensitivity: the ratio is linear in all four inputs, so a 20% error in throughput moves the answer 20%. Run it at your worst plausible throughput, not your best. Drop gross margin to 60% and every cell in the grid above falls by a quarter.
• What this misses: RCR is a year-one gross-profit test. It ignores expansion revenue, which is precisely why the 1.0 to 2.0 band is described as a retention bet rather than a sales model. It also ignores the ramp period, during which a rep costs full price and closes nothing.

---

Worked example: the same 250 seats, priced two ways

Nothing demonstrates the thesis better than holding seat count constant and letting price do the work. Take a 250-person team and buy it two different products, both at published list price retrieved July 28, 2026.

Deal	List price	Annual price per seat	250-seat ACV	Gross margin	Gross profit per deal
Jira Standard	$7.91 per user per month	$94.92	$23,730	82.8% (Atlassian FY2025)	$19,648
Salesforce Sales Cloud Enterprise	$175 per user per month	$2,100	$525,000	77.7% (Salesforce FY2026)	$407,925

Sources: Atlassian Jira Cloud pricing page and Salesforce Sales Cloud pricing page, both retrieved July 28, 2026; gross margins from Atlassian Form 10-K FY2025 and Salesforce Form 10-K FY2026.

Identical seat count. A 22.1x difference in contract value.

Against an illustrative $300,000 fully loaded rep, the Jira deal needs roughly 15 closes a year just to cover the person, before a dollar of marketing spend. One Salesforce Enterprise deal covers that rep 1.36 times over. This is why the two companies do not run the same motion, and it has nothing to do with either one’s beliefs about buyers.

Sensitivity matters more than the headline. Here is the same comparison at three seat counts, gross margins held at the disclosed figures and the rep held at the illustrative $300,000.

Seats	Jira Standard ACV	Gross profit	Deals to cover one rep	Salesforce Enterprise ACV	Gross profit	Deals to cover one rep
100	$9,492	$7,859	~38	$210,000	$163,170	~2
250	$23,730	$19,648	~15	$525,000	$407,925	0.74
1,000	$94,920	$78,594	~4	$2,100,000	$1,631,700	0.18

Illustrative model. Price and gross margin sourced as above; the $300,000 fully loaded rep cost is a modeling assumption, not a disclosed or benchmarked figure.

Two things become obvious. First, a low-priced product does not stay product-led forever; at 1,000 seats a Jira-priced deal generates $78,594 of gross profit and a rep becomes affordable, which is exactly the enterprise expansion Atlassian’s 10-K describes. Second, at 100 seats the high-priced product still comfortably funds a rep while the low-priced one needs 38 closes, a throughput almost no ramped enterprise rep sustains.

Drop gross margin to 60% and the 250-seat Jira-priced deal falls to $14,238 of gross profit and needs about 21 closes, while the Salesforce-priced deal at 60% margin still covers the rep 1.05 times over. Margin is not a detail in this calculation. It is a multiplier on the whole numerator.

---

At what ACV does product-led growth stop working and sales-led growth start?

There is no single threshold, but the industry frameworks and the disclosed contract values converge on a band. Inside sales becomes plausible somewhere around $10,000 of annual revenue per account, field sales around $100,000, and below roughly $3,000 a quota-carrying rep does not clear the arithmetic at any realistic throughput.

Christoph Janz’s 2014 tiering remains the cleanest map from contract value to motion. Cross-referenced against what the filings actually disclose, it holds up well.

Tier (Janz, 2014)	Annual revenue per account	Motion implied	Disclosed proxy from filings
Elephants	$100,000+	Field sales	Snowflake, roughly $351,400 average revenue per customer (FY2026 10-K, $4,683.9M over 13,328 customers); Datadog, roughly $104,800 (FY2025 10-K, $3,427.2M over ~32,700 customers)
Deer	$10,000+	Inside sales	HubSpot, $11,414 average subscription revenue per customer (FY2025 10-K); Atlassian, under $17,400 (FY2025 10-K, $5,215.3M over more than 300,000 customers)
Rabbits	$1,000+	Inbound and viral	Atlassian’s own disclosed growth metric is customers above $10,000 in Cloud ARR, which sits at the top of this band (FY2025 10-K)
Mice	$100+	Virality required	Jira Standard at $94.92 per user per year is a mouse-priced seat; the account only escapes the band by accumulating seats (list price, retrieved July 28, 2026)
Flies	$10+	Ad-supported	none in this comparison set

Sources: Christoph Janz, Five Ways to Build a $100 Million Business, October 5, 2014, an industry framework rather than a filing; company figures from the Forms 10-K cited. Average revenue per customer is computed from disclosed revenue and customer counts except HubSpot’s, which is a disclosed metric.

Two rules of thumb bound the band from below and above. Lemkin’s floor puts an inside rep at roughly $3,000 ACV minimum and $5,000 for planning purposes. David Skok’s constraints set the other edge: LTV to CAC above 3, CAC recovered in 5 to 7 months at the best companies, and beyond 12 months profitability turns anemic (both retrieved July 28, 2026; industry rules of thumb, not filings).

Note what Atlassian chose to disclose. Its stated growth metric is the count of customers above $10,000 in Cloud ARR, “as it measures our ability to successfully expand within our existing customer base” (Atlassian Form 10-K, FY2025). The company itself picked $10,000 as the threshold where land-and-expand becomes worth measuring, which is the same number Janz picked in 2014 for where inside sales becomes possible. That agreement between a venture framework and a filing disclosure is the strongest evidence the band is real.

---

Is there really a dead zone between $10,000 and $50,000 ACV?

The data says yes, and it is the least intuitive finding in the research. Benchmarkit’s 2025 B2B SaaS Performance Metrics Benchmarks, reflecting 2024 data, found that $10,000 to $50,000 ACV solutions are often more expensive to acquire than $50,000 to $100,000 solutions, and explicitly noted this is not a one-year exception.

Deals above $100,000 showed the lowest new-customer CAC ratio of any band in that dataset. That inverts the intuition that bigger deals are harder to win.

The mechanism is that deal complexity rises faster than price in the middle band. A $30,000 contract in most B2B categories still triggers a security questionnaire, a procurement process, a legal review, and a multi-stakeholder evaluation. You inherit the enterprise sales cycle. You do not inherit the enterprise contract value that pays for it.

For context on the surrounding numbers: the median company in that dataset spent $2.00 of sales and marketing to acquire $1.00 of new customer ARR, up 14% year over year, with the fourth quartile at $2.82 (Benchmarkit, 2025 benchmarks, reflecting 2024 data). Median CAC payback period has risen 12.5% since 2022. Acquisition is getting more expensive everywhere, which makes the band you occupy matter more, not less.

On the GTM Fit Test, the dead zone is the two-three split: too complex to sell itself, too small to pay for the human who could sell it. The response is not to staff harder. It is to move: simplify the product until it genuinely self-serves, or repackage, bundle, and re-tier until the contract clears $50,000.

---

What does Zoom’s 10-K show about running both motions at once?

Zoom is the cleanest natural experiment in public SaaS because it discloses both motions separately, under one brand, with one product. Its FY2026 10-K reports Enterprise revenue at 60.3% of total, up from 59.0% in FY2025 and 57.9% in FY2024, alongside a separately measured Online business with its own churn methodology.

Over those same three years, total sales and marketing fell steadily as a share of revenue.

Zoom fiscal year	Revenue	Sales and marketing	S&M % of revenue	Enterprise share of revenue	Customers above $100K TTM revenue
FY2024 (Jan 31, 2024)	$4,527.2M	$1,541.3M	34.0%	57.9%	3,810 (29.2% of revenue)
FY2025 (Jan 31, 2025)	$4,665.4M	$1,427.4M	30.6%	59.0%	4,088 (31.0% of revenue)
FY2026 (Jan 31, 2026)	$4,868.8M	$1,388.3M	28.5%	60.3%	4,468 (32.8% of revenue)

Source: Zoom Communications, Inc. Form 10-K, FY ended January 31, 2026, MD&A Key Business Metrics; sales-and-marketing and revenue figures via SEC XBRL, cross-checked to the filings.

Read the columns together. The enterprise mix rose 2.4 points, the count of six-figure customers rose 17% from FY2024, and total acquisition spend fell 5.5 points of revenue. That is the barbell working: the self-serve Online base supplies volume and logo acquisition, and the sales organization is aimed only at the accounts whose gross profit can pay for it.

The honest complication is in the same filing. Zoom’s Enterprise net dollar expansion rate has fallen below 100% (Zoom Form 10-K, FY2026). Moving upmarket bought larger contracts. It did not automatically buy better ones. That is the distinction between gross retention and net retention, and it is the exact risk the RCR’s 1.0 to 2.0 band is designed to flag.

The other lesson is about where the self-serve motion actually breaks: not at signup, but at activation, which is what makes a self-serve funnel convert at all. A free tier that acquires accounts nobody activates is a marketing cost wearing a product costume.

---

The hybrid reality: why almost every company at scale runs both

The binary is a useful teaching device and a bad operating model. At scale, essentially every public SaaS company in the tables above runs both motions, because a company with 300,000 customers and 4,000 enterprise accounts is two businesses with two cost structures under one income statement.

The evidence is in the disclosures, not the positioning.

Company	The self-serve end	The rep-covered end	Blended S&M % of revenue
Atlassian	Free tier to 10 users, Standard at $7.91 per user per month	Sales force pointed at large enterprise expansion, per Item 1 of the FY2025 10-K	21.8% (FY2025)
Datadog	Free tier to 5 hosts, Infrastructure Pro at $15 per host per month	4,310 customers above $100K ARR producing 90% of ARR; 603 above $1.0M	27.9% (FY2025)
Zoom	Online segment with separate churn methodology	Enterprise at 60.3% of revenue; 4,468 customers above $100K TTM	28.5% (FY2026)
Salesforce	Free Suite at $0, Starter Suite at $25, Pro Suite at $100 per user per month	Enterprise at $175 and Unlimited at $350 per user per month	34.5% (FY2026)
HubSpot	Free Sales Hub to 2 users, Starter at $7 per seat per month	Professional at $90 plus $1,500 onboarding; Enterprise at $150 plus $3,500 onboarding; partner channel at approximately 49% of revenue	44.1% (FY2025)

Sources: pricing pages for Atlassian, Salesforce, HubSpot and Datadog retrieved July 28, 2026; customer and segment metrics from each issuer’s Form 10-K as labeled.

Datadog is the purest expression. Its consumption pricing starts at a free tier and $15 per host per month, which is why its land motion needs no human at all, and yet 4,310 customers above $100,000 in ARR generate 90% of ARR (Datadog Form 10-K, FY2025). The self-serve tier is not the business. It is the qualification mechanism that tells the sales organization which accounts are worth a person.

That is the barbell in one sentence: acquire the logo at near-zero marginal cost, then deploy humans only against the accounts whose gross profit can pay for them. The blended sales-and-marketing ratio in the first table is an average of those two structures, weighted by mix, which is why comparing two companies on that single line so often produces nonsense.

---

Does product-led growth actually mean lower sales and marketing spend?

Not reliably, and HubSpot is the case that proves it. HubSpot discloses average subscription revenue per customer of $11,414 across 288,706 customers, squarely in the inside-sales band, and still spent 44.1% of revenue and 52.6% of gross profit on sales and marketing in FY2025. That is more than Salesforce on both measures.

There are three readings, and they are not mutually exclusive.

The freemium funnel is a lead-generation cost center, not a sales substitute. A free tier does not remove acquisition cost; it relocates it from the “sales” half of the line to the “marketing” half. Content, brand, paid acquisition, and events all still sit inside the same expense line. On an 83.8% gross margin, HubSpot has more room than almost anyone to spend it, and it does.

Channel is a third motion the binary ignores. HubSpot’s Solutions Partners and partner-referred customers represented approximately 25% of customers and approximately 49% of revenue in FY2025 (HubSpot Form 10-K, FY2025). Roughly half the revenue arrives through neither self-serve nor a direct rep, and partner economics sit inside the same sales-and-marketing line.

Net retention decides whether you have to buy growth. HubSpot’s net revenue retention was 103.5% in FY2025. Snowflake’s was 125% (Snowflake Form 10-K, FY2026). A company at 103.5% has to acquire almost all of its growth; a company at 125% expands into most of it. That difference shows up as acquisition spend regardless of which motion is on the org chart.

The bear-case version in one line: product-led describes how a customer arrives, not what it cost to make them arrive. Any framework that treats “product-led” as a synonym for “cheap acquisition” is reading a label instead of an income statement.

---

The bear case: what the skeptics get right

The strongest objection is that the whole framing is an accounting artifact. A skeptic can read the same tables three different ways, and each reading has real weight.

Sales and marketing is a line item, not a motion. Commissions, brand advertising, events, partner rebates, developer relations, and in some companies customer success all live in that line, in different mixes with different capitalization policies. Two companies with identical go-to-market realities can post ratios several points apart purely from classification. Cross-company comparison of this line has genuine noise, and the noise is not randomly distributed; it correlates with how a company chooses to describe itself.

The private-company data cuts against reading public ratios as a norm. Benchmarkit found VC-backed private SaaS companies spending 47% of revenue on sales and marketing against 33% for PE-backed companies, and R&D at 34% of revenue for private versus 23% for public (2025 benchmarks, 2024 data). The ownership structure moved the ratio by 14 points. That is larger than the entire Salesforce-to-Atlassian gap this piece opened with. If capital structure explains as much as motion does, motion explains less than the framing implies.

Revenue per employee does not follow the ratio the way people assume. Salesforce generated roughly $498,000 of revenue per employee in FY2026 against Atlassian’s roughly $378,000 in FY2025. The company spending 12.7 more points of revenue on sales produced more revenue per head, because Atlassian’s cost simply moved to R&D at 51.2% of revenue. Zoom is higher than both at roughly $655,000, Snowflake roughly $517,000, Datadog roughly $423,000, HubSpot roughly $352,500 (computed from revenue and headcount disclosed in each issuer’s most recent 10-K). There is no clean relationship between acquisition ratio and headcount efficiency.

Here is the honest weighing. The bear case does not break the arithmetic; it bounds the interpretation. RCR is still the correct test for whether a specific segment can fund a specific rep, because that calculation uses your own realized numbers rather than someone else’s expense classification. What the bear case correctly destroys is the comparative use of these ratios as a scoreboard. Use the filings to see the range and the structure. Do not use them to grade a company you do not have segment data for.

---

Where this model is vulnerable

Five limits, stated plainly.

Consumption pricing detaches ACV from the contract at signature. Snowflake and Datadog land small and grow into seven figures. The contract value that eventually pays for the rep is a forecast at the moment the rep is assigned, not a fact. Snowflake’s 125% net revenue retention makes that forecast reasonable; a company at 100% has no such cover. The choice between usage-based and seat-based pricing changes what the RCR numerator even means.

The channel is missing from the binary entirely. HubSpot’s roughly 49% partner-sourced revenue is neither product-led nor sales-led. Channel economics have their own coverage math, their own margin drag, and their own retention profile, and none of it fits a two-column matrix.

Gross margin itself is an accounting construct. What sits in cost of revenue versus operating expense varies by company judgment. Support, hosting, professional services, and amortization of capitalized software all get classified differently. Since gross margin is a multiplier on the entire RCR numerator, that classification noise propagates directly into the answer.

AI agents are repricing the denominator right now. If the fully loaded cost of covering an account falls, the ACV floor for a “human” motion falls with it, and the dead zone narrows or moves. Salesforce now lists Agentforce 1 Sales at $550 per user per month (retrieved July 28, 2026), which is itself an argument that the seat is being redefined rather than merely repriced. That repricing is the subject of how AI is breaking per-seat SaaS pricing, and it is the single largest source of model risk in this piece.

Salesforce does not disclose a customer count or an ACV. Every Salesforce contract-value figure in this article is modeled from list price and an assumed seat count, and labeled as such. Real enterprise deals carry discounts, multi-cloud bundles, and multi-year terms that list pricing does not show. The 22.1x price gap between Jira Standard and Sales Cloud Enterprise is a list-price gap, and realized gaps are smaller.

---

What an operator with real cost data sees

One thing the filings cannot show you is the input side of the gross margin that drives the whole calculation. I can, from a consumer SaaS platform I operate, with the specifics anonymized and the numbers limited to what is actually documented.

The platform runs on ARM64 compute with an edge CDN in front of it rather than on a hyperscaler. The measured result is roughly 80% to 90% lower infrastructure cost for equivalent performance, and $0 egress against roughly $0.09 per GB on the large clouds. A single production cell is held under a hard cost gate of about EUR 25 per month, enforced with an explicit ban list of expensive managed services.

That is a go-to-market decision disguised as an infrastructure decision. Every point of gross margin bought at the cost floor is a point added to the RCR numerator. A product that would sit at 60% margin on a hyperscaler and 85% on owned substrate has a 42% larger acquisition envelope at identical price and identical churn.

The pricing side is the same lever from the other end. The ladder on that platform is four tiers with roughly 1.7x steps between rungs, a deliberate high anchor tier, and 44% to 50% effective annual discounts to push annual mix. The step multiple is what determines whether a customer segment lands in the mouse band or clears into deer territory, which the GTM Fit Test then reads as a signal.

The honest limitation: that platform is pre-launch, so there is no real CAC, LTV, or net retention number to report, and I will not invent one. The transferable claim is structural rather than measured. Cost floor and tier design are the two inputs an operator fully controls, and both of them move the RCR before a single hire is made.

---

What operators should take from this

The instinct is to pick a motion and then build the company to match it. That ordering is backwards. Compute the arithmetic first, then let it name the motion, per segment.

1. Run the Rep Coverage Ratio before you write a comp plan, in gross profit dollars. Use realized ACV by segment, not list price, and your true fully loaded rep cost. If you have never computed the denominator honestly, you do not know whether your sales team is a growth engine or a subsidy.
2. If RCR is under 1.0, do not hire an account executive. Hire a growth engineer and fix activation instead. Below the line, an additional rep makes the unit economics worse in direct proportion to how well they perform, because every deal they close loses gross profit.
3. If RCR sits between 1.0 and 2.0, instrument gross retention monthly and treat it as the gating metric. In that band, the model only works if year two is nearly free. Watch gross retention specifically, not net, because expansion revenue can mask churn that will surface exactly when you scale the team.
4. Price the free tier as a qualification filter with a defined graduation trigger. Datadog’s free tier to 5 hosts and Atlassian’s free tier to 10 users are not marketing giveaways; they are the mechanism that identifies which accounts deserve a human. Define the usage threshold that routes an account to a rep, and instrument it, before you staff the team.
5. Refuse to sit in the $10K to $50K dead zone. Either simplify until the product genuinely self-serves, or repackage and re-tier until the contract clears $50,000. Tier-step design is the lever here: a deliberate step multiple between rungs is what moves a segment out of the band rather than staffing your way through it.
6. Report CAC payback separately for self-serve and rep-assisted cohorts, and segment comp plans by ACV band rather than geography. A blended payback number hides which motion is funding the company, and a single comp plan across ACV bands ignores that throughput per rep, not territory, is the constraint being managed.

---

Are product-led and sales-led growth converging?

The filings say yes, and they are converging from both directions.

Salesforce’s sales-and-marketing ratio has fallen three consecutive fiscal years: 36.9% of revenue in FY2024, 35.0% in FY2025, 34.5% in FY2026 (Salesforce Forms 10-K). Over the same period it built out a self-serve ladder that now starts at a Free Suite and runs through Starter at $25 and Pro Suite at $100 per user per month. The sales-led archetype is buying itself a product-led bottom.

Atlassian moved the other way: 20.1% of revenue in FY2024 to 21.8% in FY2025 (Atlassian Forms 10-K), while its 10-K describes a sales force whose purpose is “expanding and deepening strategic relationships with our existing large enterprise customers.” The product-led archetype is buying itself a sales-led top. Its own risk factors concede the dependency in the other direction:

Our business model for low-touch customers is based in part on attracting a high volume of customers through free trials, limited free versions, and affordable starter licenses.

High volume is a requirement, not a virtue. It is what a low-touch model must produce to survive its own price point.

Datadog already sits at the destination, running a free tier and a $100K-plus enterprise book simultaneously, with 90% of ARR concentrated in 4,310 accounts out of roughly 32,700 (Datadog Form 10-K, FY2025).

So the endgame is neither motion. It is a barbell in which self-serve acquires the logo at near-zero marginal cost and humans are deployed only against accounts whose gross profit can pay for them. The question this piece opened with does not disappear in that world. It gets asked per segment instead of per company, which is harder, more often, and with better data.

Product-led vs sales-led growth was never a values question. It is a division problem, and the denominator is a person.

---

Analysis, not investment advice. Company figures are drawn from the public SEC filings cited inline by issuer and fiscal period, and from published list pricing pages retrieved July 28, 2026. Third-party benchmark research and venture-industry rules of thumb are labeled as such and are not filings. Frameworks here, including the GTM Fit Test and the Rep Coverage Ratio, are for understanding go-to-market business models and tradeoffs, not for making buy or sell decisions.

Want the full toolkit for reading filings like this, the Rep Coverage Ratio model, the GTM Fit Test scorecard, and the segment CAC-payback worksheet used above? It’s in the Tech Business Analysis Playbook.