A corpus-only study of 201 companies, each scored on ten go-to-market motions across six funding stages and twelve categories.
Eight chapters on the corpus-wide record — the findings that survive every caveat, before we go sector by sector. How motion moves by round, how companies actually reach buyers, how they grow and charge, what works and what doesn't, and what changed in the last year.
The tidy story goes like this: a young software company starts product-led, lets users self-serve, bolts on sales-assist around Series B or C, and becomes a full hybrid once enterprise procurement shows up. A clean sequence, one stage at a time.
The record does not support it as a general rule. Across 201 companies scored on ten motions, the centre of gravity is already sales-led in five of six stage bands — including Seed. By the time a company leaves enough public evidence to study, it is already carrying a sales motion. The pure product-led phase, if it existed, happened off-camera.
Product-led-sales is the second-strongest stage-weighted motion at every stage from Seed through Growth. Product still matters enormously — as a wedge, a credibility device, an expansion surface. It is simply not, in most companies, the centre of gravity. Pure product-led leads for only 8% of the corpus (16 of 201). The other 92% put a human in the loop.
Most GTM commentary ranks a leaderboard of the biggest companies and reasons backward. That corpus is ~80% late-stage and public, and it has converged on the same hybrid — so the picture comes out muddy and non-transferable. We did the opposite: we designed a sample to fill a matrix.
The analytical unit is a (stage × category) cell. Companies were selected to populate those cells, weighted toward the ICP — Seed-to-Series-B founder-sellers and their first few reps — not chosen for size. Stage is the primary axis, because motion is genuinely stage-dependent: validation at Seed, building a repeatable channel at Series A, proving it across quarters at Series B, optimizing at Growth. The insight the corpus was built to expose is the transition — at which round, in which category, the centre of gravity shifts, and what triggers it.
Each company is stamped on four axes: stage (seed → public), category (twelve, from devtools to regulated health), ACV band × buyer (self-serve developer through enterprise InfoSec), and an outcome signal — winner, steady, or stalled/down-round. Motion itself is scored as a ten-way vector summing to ~1, over company-neutral archetypes (product-led, product-led-sales, sales-led, marketing-led, community/developer-led, partner-ecosystem-led, procurement/trust-led, founder-network-led, usage-led, embedded-API-led).
The single least-fakeable decision in the study: 15–20% of every stage band is a stalled or down-round company — 27 of 201 overall (13.4%). Without a failure cohort you can only learn what winners did; you can never say what doesn't work, which is half the point.
A study that only looks at winners can describe success; it cannot diagnose failure. So the failures were sourced on purpose.
Every company was first sourced from public, stage-tagged lists as a hypothesis — a provisional stage and outcome. Then each ran through a fixed pipeline. Research rebuilt the record against an evidence-required schema: every load-bearing claim carries a dated public source or is moved to a named list of gaps — stage is verified from evidence, not assumed, and unknowns are declared rather than guessed. Adversarial verification then took a second pass whose only job was to refute the thesis; where it found the record over-claimed, a repair step corrected it — that pass caught, among others, a company filed as "Series B" that had in fact already been acquired. The output is a structured record per company, aggregated into cells.
No number in this report comes from free web search or an unstated benchmark — every figure traces to the aggregated corpus, which is what makes it a primary source. And efficacy here is an observed-pattern signal, not a controlled experiment: it records which motions concentrate among stronger outcomes, not that a motion caused an outcome.
Corpus. 201 structured research units across 67 (stage × category) cells. Motion is a 10-way vector per company; cell figures are weighted centroids.
Verification is uneven. ~129 units passed adversarial verification; ~72 are research-grade (single-pass, evidence-required but not adversarially refuted). Treat single-cell claims accordingly.
Known biases. Survivorship (public evidence is richer for bigger, louder companies — stalled firms are included but quieter by nature); public-evidence-only selection; and the uneven verification above. All three are named; none are corrected away.
N=201
Sales-led is the weighted top motion in five of six stage bands, and it climbs steadily — from 18.9% of the Seed motion centroid (N=29) to 27.3% at Public (N=17). Product-led-sales holds the second tier throughout, peaking at 16.6% in Series B.
Seed companies are not mostly self-serve experiments. Across 29 Seed units, sales-led is already the top motion (18.9%), ahead of product-led-sales (13.9%) and marketing-led (13.4%). Series A and B don't introduce sales so much as make it legible: the conspicuous change is product-led-sales climbing 14.5% → 16.6%, a product surface increasingly used to qualify, route, and support selling. By Growth the stack is broad but still tilted to sales (22.4%); at Public, partner-ecosystem-led rises to become the second motion (14.7%).
Nobody runs a single motion. The average company's strongest play carries just a quarter of the weight — the other three-quarters is the strategy.
N=201 across 12 categories
The average says sales-led. The category map says where that average hides useful timing. Most lanes present as sales-led early with adjacent support; a few show a genuine round-by-round handoff worth naming.
If a product-to-sales handoff exists anywhere in the corpus, it lives in devtools (N=15): product-led at Seed, moving to product-led-sales by Series A and B. Its category centroid leads with product-led-sales (18.3%), then product-led (15.5%) and community/developer-led (14.9%) — the only row where the developer motions out-weigh sales.
Security and compliance companies carry a sales-plus-trust posture from Series A onward — procurement/trust-led runs well above its corpus baseline. When the buyer is InfoSec, the trust envelope is not a late-stage add-on; it is the entry motion.
Fintech reads product-led at Seed but moves to sales-led by Series A and stays there. The corpus can't name the ACV threshold that triggers it — it doesn't expose per-cell ACV — but the flip is early and durable.
Across every stratum, one variable moves the motion more than stage or category: who signs. When a developer sits in the buying group (78 companies), sales-led falls to 0.17 and product-led rises — against 0.25 sales-led for everyone else. No other buyer type discriminates, because nearly every company lists practitioner, department-lead, exec, and InfoSec together. The developer is the exception that proves the rule: product-led is a buyer property, not a stage.
1,026 channel observations
Motion vectors are an abstraction. Underneath them sit concrete channels — and when you count only the load-bearing ones, rated dominant or strong, the picture sharpens into something a founder can act on.
The corpus does not, on the whole, reach buyers by letting them sign up. Direct enterprise sales is the single most load-bearing channel, and the two that follow — partners and a distinct procurement/trust channel — are the mechanics of getting through a security review, not of demand generation. Self-serve exists, but as a wedge feeding a sales motion far more often than as the motion itself.
The dominant channel in modern B2B isn't a signup page — it's a security questionnaire, answered well.
N=201
Two mechanisms decide whether a landed account becomes a franchise: how it expands, and how it is billed. On both, the corpus has moved off the classic seat-based subscription.
Expansion splits three ways — usage and volume, new products and modules, and additional seats — with usage/volume just ahead. Billing tells the same story: the pure per-seat subscription barely leads volume- and transaction-based pricing, and usage/consumption is a substantial third. For a large share of the corpus, revenue now grows with the customer's usage, not their headcount — which quietly changes what "expansion" even means.
N=201
Outcomes are not evenly distributed across stages. The corpus has a clear proving ground and a clear graveyard.
Here is the corpus's most useful negative result. Compare the average motion of the 88 winners with the 27 stalled companies and they are, to the decimal, the same posture — sales-led around 0.22, product-led-sales around 0.15. You cannot read a company's fate from its motion mix. "Which motion should we run" is the wrong question: the winners and the failures ran the same ones.
The faint signal that does exist points one way, and it is worth naming. Winners carry slightly more procurement/trust weight (0.12 vs 0.10); the stalled cohort carries slightly more marketing-led (0.13 vs 0.12). On the margin, trust and proof beat volume and noise. It is a small effect on thin numbers — but it is the only directional tilt the outcome data offers, and it agrees with everything else in the corpus.
Winners and failures ran the same motions. What separated them was fit and timing — not the play they picked.
The 27 stalled or down-round companies are not concentrated in a single motion — you cannot blame "too much PLG" or "too much sales." What they share more often is a stage-mismatch: a motion running ahead of, or behind, the trust its buyer required — self-serve machinery aimed at a procurement buyer, or an expensive sales motion with no proof to hand the reviewer. Motion without enough trust for the buyer is the pattern that recurs on the failure side.
founded-year on 189 of 201 · 179 documented shifts
A snapshot can't watch a company change over time. But it can sort companies by when they were founded, and read the change events each has put on record in roughly the last year. Both point the same way.
Sort the corpus by founding era and the motion drifts, cleanly. The oldest cohort — founded before 2015 — leads sales-led at 0.25, with partner-ecosystem second. The newest — 2022 and later — is down to 0.20 sales-led, with product-led-sales risen from 0.12 to 0.16 and founder-led motion up with it. Newer companies still put a human in the loop, but the human is increasingly a founder plus a product surface, not a partner channel and a full sales org.
Read the 179 documented go-to-market changes from the last twelve months and one thing dominates: companies expanded their product surface, and did it with AI. 61% shipped a new product or platform; 42% launched or repositioned specifically around AI. Far fewer touched the sales machine — 16% moved upmarket or added sales, 15% added partnerships, and only a handful changed pricing or opened self-serve. The headline change of the year is not a new motion; it is AI-shaped product expansion bolted onto the motion a company already had.
The motion of the last year didn't change. The product did — and it grew an AI surface.
It is tempting to read outcomes by vintage and conclude newer companies win more: 2021-and-later founders show 1% stalled against 22% for pre-2021. That is survivorship, not success — a company founded in 2023 has not had time to fail or take a down-round, while the older cohort has. Read the vintage trend for what companies do, never for how they end. On the change data, the outcome clock simply hasn't run yet.
The corpus-wide average is a useful lie. Underneath it, each sector runs a recognisably different motion — and the two most instructive, security and devtools, run near-opposite ones for the same reason: their buyer. This part maps all twelve at a glance, then goes deep on those two.
12 categories · n = 10–22 each
Every sector's motion mix on one axis, ordered from the most sales-led to the most product-led. The spread is the point: "software company" hides a factory floor of different machines, and two of them barely resemble each other.
Two sectors are worth stopping on — not because they are the biggest, but because they are the clearest natural experiment the corpus contains. Security and devtools sell to opposite buyers, and run opposite motions as a direct result.
n=22 · InfoSec buyer on 100% of companies
Security is the cleanest proof in the corpus that motion follows the buyer. Every one of the 22 security companies sells, among others, to InfoSec and procurement — the only sector where that buyer is universal. When the person who signs the contract is the same person paid to run the security review, the trust envelope stops being a late checkbox and becomes the entry motion.
Sales-led still leads the sector's centroid (0.23), but procurement/trust rides second (0.18) — and, unlike every other sector, it is there from the start: 0.22 at Seed, 0.19 at Series A, where elsewhere trust barely registers before Growth. The procurement channel is the second-most load-bearing in the sector (12 of 22 companies), behind only direct sales. This is a market reached through a security questionnaire, not a signup page — and the questionnaire is answered before the deal, not after.
Across rounds, trust stays high and early while the sales machine assembles around it. By the public stage, sales-led jumps to 0.33 and partners rise to 0.17 as the motion scales — but trust does not grow into the picture the way it does elsewhere, because it was never absent. It is the ante paid from the first deal.
The exemplars decode the motion precisely. ConductorOne does not sell access reviews; it reprices identity as "governed work," and the security review is what qualifies the buyer in the first place. Veza sells an authorization graph as the missing security data layer — trust rendered as infrastructure, which is why ServiceNow wanted its identity context inside AI controls. Sublime Security runs a practitioner PLG wedge — free analyzer, open docs, community — whose real job is to prove transparency against black-box incumbents, then converts into an enterprise replacement sale. In all three, the free or product surface exists to establish trust; the revenue lives in the governed enterprise motion.
In security, "prove it" comes before "buy it." The motion is trust-first and sales-closed — and it is structural, dictated by a buyer who is paid to be skeptical.
n=15 · developer buyer on 100% of companies
If security proves motion follows the buyer, devtools proves it from the opposite pole. It is the single sector in the corpus where sales-led is not the top motion — it sits fourth (0.12), behind product-led-sales (0.18), product-led (0.15), and community/developer-led (0.15). The reason is security's reason, inverted: every devtools company sells to a developer, and the developer is both the user and the first buyer. Adoption is bottom-up, hands-on, and usually free; the org purchase is a consequence, not the entry.
Devtools is where you can watch the textbook transition happen. Product-led leads at Seed (0.21); product-led-sales takes over at Series A and B (0.21, 0.20); sales-led finally leads at the public stage (0.20). Round by round, a product motion hands off to a sales motion — the exact arc the rest of the corpus refuses to show. The developer channel is uniquely central here (8 of 15, second only to direct sales), and ACV is overwhelmingly "mixed" — the signature of a free self-serve base feeding an enterprise ceiling.
Even at its most bottom-up, devtools converges on a sales motion — because the enterprise upgrade forces it. GitLab is "easy to misread as open-source PLG, but the economic center is enterprise standardization: the free surface creates gravity, the money concentrates in governance-heavy Ultimate." CodeRabbit installs free at the pull request, but sells neutrality — the independent, agent-agnostic quality gate an enterprise can mandate. Momentic makes engineers own quality inside the repo, then expands on CI volume and organisational control. The free developer surface is the wedge; the compliance-and-governance layer is the business.
Product-led isn't a strategy you choose. It is what a developer buyer allows — and even then, it converges on a sales motion the moment the org, not the individual, has to sign.
A ten-way motion vector is a useful abstraction and a lossy one. A single label — "sales-led" — hides five or six genuinely different machines. This part opens the biggest labels and names what is inside them, from the companies' own channel evidence.
5,646 documented channel mechanisms
Take the 111 companies whose dominant motion is sales-led. Their channels tell you what that sales motion actually is — and it is overwhelmingly relationship- and proof-driven, almost never cold.
The most useful thing a motion label can do is dissolve. "Run a sales motion" means nothing until you know which of six it is.
Averages describe the middle. The edges are where the learning is. These are the companies whose motion vector sits farthest from what their (stage × category) neighbours do — each a documented, source-verifiable case of running against the grain, and each with a reason.
the three largest cell-outliers, read in full · individually verifiable
Averages describe the middle; the edges teach. Rather than catalogue every outlier, here are three companies that each overturn a different assumption — each a single, source-verifiable record, and the highest-confidence material in the report.
Series A · Data / AI infrastructure · expected sales-led, runs founder-networked. Rain builds compute-in-memory AI accelerators, and its cell — Series-A AI infrastructure — runs sales-led. Rain runs on the founder's network instead. Its strongest asset is not a product funnel but access: elite AI investors, advisors, and strategic buyers, cultivated personally. At the hardware frontier, where there is no repeatable motion yet and every design win is bespoke, relationship access substitutes for go-to-market. That is also the fragility the record names outright — the reported failure point is precisely the conversion of that access into a repeatable hardware sale. The lesson: capital and relationship access can be the motion at the frontier — right up until the moment it has to become one.
Growth · Regulated health · expected sales-led, runs marketing-led. Regulated health reads as a slow, institutional, sales-led sale. Solace is the opposite: a Medicare-covered patient-to-advocate conversion funnel that runs, mechanically, like a consumer performance-marketing machine — Webflow/Statsig CRO, Meta and Google, influencer hiring, conditioned SEO. The first conversion is not a demo; it is an insurance-eligibility check. The regulated reimbursement engine is the moat; the motion is high-velocity consumer acquisition. The lesson: a regulated category does not dictate a regulated motion. The buyer and the payment mechanism do — and here a public payer quietly underwrites a consumer funnel.
Series B · Horizontal productivity · expected sales-led, runs product-led. By Series B, horizontal productivity has mostly converged on sales. Gamma stays product-led, and the reason is instructive: its distribution is baked into its output. It is easy to misread as an AI PowerPoint; the sharper truth is that Gamma makes the artifact itself distributable — a web-native page generated from messy input, shared as a link, embedded in workflows, copied by creators. Every deliverable is a piece of marketing. When the product's output is the channel, product-led growth doesn't run out of road at scale — it compounds. The lesson: the most durable product-led motions are the ones where using the product distributes it.
Six patterns recur across every cut of the corpus — sector, stage, channel, outcome, vintage. None is a controlled result; each is the compression of everything above into something a founder can carry into a room.
Stage barely moves the motion; the buyer moves it. When a developer is in the buying group, sales-led falls to 0.17 against 0.25 for everyone else — the single largest swing in the corpus. Design the motion around who signs, not around your round.
Direct sales, partners, and a distinct procurement/trust channel are the three most load-bearing ways companies reach buyers; self-serve and developer channels are a real but distinct minority. The dominant channel in modern B2B is not a signup page — it is a security review, answered well.
Motion does not predict outcome: winners and stalled companies run the same mix. The only tilt that survives is small and directional — winners carry slightly more procurement/trust weight (0.12 vs 0.10), the stalled slightly more marketing. On the margin, proof beats noise; nothing bigger separates success from failure.
Expansion splits across usage/volume, new modules, and additional seats — with usage just ahead — and in billing, volume/transaction and usage pricing together rival the per-seat subscription. For a large share of the corpus, revenue now grows with the customer's usage, not their headcount.
The average company's strongest motion carries just a quarter of its weight (0.26); 152 of 201 are pure hybrids and only 7 are monoline. "Pick your motion" is the wrong instruction. The winning object is a stack, tuned — and it tightens toward sales, round by round.
Of 179 documented shifts in the last year, 61% shipped a new product or platform and 42% launched or repositioned around AI; only 16% changed how they sell. The motion of the year did not change. The product did — and it grew an AI surface. Meanwhile the pure-sales playbook is aging out: newer cohorts lean off sales and partners, toward product and founder.
This report is an agenda, honestly scoped — not a closed argument. It is worth being precise about what it can and cannot yet carry.
The corpus-wide findings are solid: sales-as-default, the hybrid-stack reality, the trust-and-proof vocabulary, the Series-B proving ground. These rest on the full 201 and survive the caveats. The cell-level matrix is thinner: the design target was N≥15 per cell, but the current aggregate has a maximum cell N of 6, a median of 3, and zero cells at threshold. Every per-cell claim here is therefore directional — strong enough to set the agenda, not broad enough to close it.
The arithmetic is simple: 67 analytically interesting cells at N≥15 is roughly a thousand companies. That is the distance between this study — a legible, honest first map — and a conclusive one. The map is drawn correctly; it is not yet drawn densely.