The Series A Proof Engine

1. How to read this document

Three kinds of number appear here and they are not interchangeable. The label travels with the number everywhere.

The companion file Series-A-Proof-Engine-Model.xlsx carries the same model as live formulas. Change a driver on tab 1 and every tab recalculates. It calculates on open, so open it in Excel rather than previewing it.

2. The answer, on one page

$3M of ARR at $30 a seat is 8,333 paid seats live at month twelve. That single number drives everything else in this document.

At ten paid seats per paying workspace that is 833 paying workspaces, which requires 17,544 signups across the year. Two of the nine channels put signups into the product without anyone visiting the website, so the site only has to carry 161,821 visitors rather than the 292,000 a pure web funnel would demand. That is what makes the 6% visitor to signup target reachable, because everything left on the site is intent traffic.

Paid media cannot carry this alone. Per click acquisition clears the payback ceiling only because four of the nine channels cost nothing per click. Blended across every channel the cost is $49 a signup against a $108 ceiling, giving a CAC of $976 against an allowable $2,160, and payback at 5.4 months.

Total spend is $856,368, of which $518,968 is working media and $337,400 is contracted people and tools. No salaries sit in it. That is $0.28 of spend per $1 of new ARR, and it stays under the $1M paid threshold that would otherwise force a brand allocation.

The finding that matters most. Every comparable product in this category monetises power users through a tier ladder with a 3x step and a 10x step. Hyrax has one paid step at $30 and overage switched off by default. With one step, blended revenue per seat is exactly $30 and the funnel must be twice as large as it would otherwise need to be. Section 5 has the evidence and section 8 has the arithmetic.

3. The specification. Thirteen metrics, and what each one rules out

The metric set is not neutral. Each target eliminates options before any choice is made.

Two of the thirteen are product commitments, not marketing commitments. Activation and daily use are won in onboarding and in the first run. No amount of spend moves them, and no channel in this plan can compensate for missing them.

4. The pricing reality

Read live from hyrax.dev/pricing in a real browser on 2026-09-15, because cached fetches of that domain return stale copy.

The category, read live on the same day

What the overage research actually found

No vendor in that set publishes the percentage of seats that exceed their allowance, average overage dollars per seat, overage as a share of list price, or usage derived revenue per user by plan. No independent benchmark measures it either. Benchmarkit surveyed 490 SaaS professionals in July 2022 and found 77% of vendors charge overages, but that is a vendor policy figure and not customer incidence. LogiSense and Benchmarkit in 2025 found 26% charge actual usage over commitment, 58% allow excess without charging and 14% impose a hard cap. None of those answer the question.

The closest proxy is net revenue retention, and it is not a substitute. OpenView in 2021 reported 120% for usage based companies against 110% for traditional. High Alpha in 2025 found hybrid at 105%, subscription at 102% and pure consumption at 99%. Bessemer has published 137% average net dollar retention for companies implementing usage based pricing without stating the sample. The direction is sample dependent, and 120% net retention does not mean 20% average overage, because that figure absorbs seat additions, tier upgrades, contraction and churn together.

Conclusion. Do not model overage as a revenue line. In this category overage is a margin guard, not a growth lever. Copilot and v0 hand out credits worth as much as or more than the price. Lovable prices a top up at a 20% premium to the included rate. Every one of them puts the real power user revenue in a higher tier instead. True revenue per seat is therefore a blend across a ladder, and with one paid step that blend is exactly $30.

5. The bottoms up funnel

Read from the bottom. Every rung carries the rate applied and where that rate comes from.

Rung eleven does the most work in this document. Marketplace installs and assistant surface signups never pass through the site, so the site carries roughly 65% of signups. That takes the traffic requirement down to 161,821 and leaves only intent traffic on the site, which is the only way a 6% visitor to signup rate is reachable.

Scale reference. The one complete funnel published anywhere in the research file converts 0.625% of visitors to trials, 2,500 from 400,000 visits (p.21). At that rate the same signups would need 1.8M visitors. Worth knowing before anyone checks: the file states 6.25% trial conversion while its own table shows 25 customers from 2,500 trials, which is 1%.

6. Unit economics and the ceiling the plan has to clear

Payback inside twelve months means acquisition cost cannot exceed one year of gross profit per workspace.

The blend is the mechanism. Four of the nine channels cost nothing per click, so as the fixed cost channels compound the blended figure falls while the variable channels stay flat. That is exactly the behaviour your metric "CAC holds steady as spend rises" describes.

Where the ceiling bites. At five paid seats per workspace the ceiling falls to about $54 a signup and no per click channel clears it at all. At ten it is $108 and the plan works. Seats per workspace is the first number to instrument and the largest single risk in the model.

7. The monthly build

Signups ramp at 14% a month from a base of 643 and are scaled so the year totals the ladder requirement. Revenue is shown as an exit run rate at each month end.

Why the ramp is back weighted, and it is not a choice. The only published quarterly expectation ladder in the research file is explicit: quarter one measures articles and consistency, quarter two impressions and traffic, quarter three ranking, quarter four leads, and "do not expect leads for the first three quarters" (p.23). Search and content are stated as a nine to twelve month build. So marketplace, network and paid carry the first half and the compounding channels carry the second.

One honesty note on the ramp. This model converts signups to paid in the same month. Adding a one month conversion lag pushes exit ARR to roughly $2.7M, which is why the plan is built to land at $3.1M rather than exactly $3.0M. The Excel model lets you test it.

8. Sensitivity. The three numbers that move the answer

The ladder case is not invented. Every comparable runs a 3x step and a 10x step off its base: Copilot at $10, $39 and $100; Cursor at $20, $60 and $200; Replit at $20 and $100; Windsurf at $20 and $200. Applying that shape to a $30 base gives roughly $90 and $300. At a mix of 70% base, 25% mid and 5% top, blended revenue per seat is $58.50, which halves the seat requirement and halves the funnel.

The mix behind $58.50 is an [ASSUMPTION]. No vendor publishes its plan mix. The tier prices are [LIVE] from seven vendor pages. Whether to add steps is a pricing decision, not a marketing one, so this plan is built entirely on the live single step and the ladder is shown only as the largest available upside.

9. The channel portfolio. Nine channels, with the mechanic on each

Four of the nine cost nothing per click. That is what makes the payback metric true.

Organic and owned comes to 63.1% of signups, clear of the above 30% requirement, which the file rates as excellent at that level (Accel p.77).

10. The channels that are cut, and the number that cuts them

11. What the companies that actually did this did

Every figure here is from the research file. The pattern across them is in the last row of each block.

The three patterns that repeat

One. Distribution lives inside the product, not beside it. Marketplace installs, public reviews in other people's repositories, an open sourced predecessor, a shareable demo artifact, documentation other projects embed. In every fast case the product itself is the channel.

Two. One channel proven before any second one. Not one channel forever. One at a time, with a hard box on the test. The file is explicit that until that channel is found and stabilised, nothing else marketing does matters.

Three. The narrow segment came before the spend, every time. Clay churned nearly all its customers to get narrow. One company went from 35,000 targets to a trigger based universe with 60% and above response. Another went from spraying to 2,500 companies of which 300 qualified, and those 300 produced the first $8M in 12 months.

And what the failures did

RethinkDB. From their own post mortem: you are in the market your users think you are in, not the one you say you are in. Their users were "willing to pay less for the lifetime of usage than the price of a single Starbucks coffee." Their conclusion was that developer tools need roughly ten times the leads per sale of other software. [FILE] p.113

Snyk's first two years. Roughly $4M raised against $100K of revenue, with thousands of enthusiastic developers who would not pay. They then reached a $2.6B valuation in 30 months once the buyer was identified. Developer enthusiasm is not revenue until you know who signs. [FILE] p.114

Jasper. First mover advantage, strong brand, $125M raised, and a product that was prompts in a text box supported by strong search marketing. It was undermined the moment the underlying capability commoditised. A content and search motion wrapped around a thin product is the named failure case for a company this size. [FILE] p.70

Webflow, on the mistake itself. "I really did used to think that if you build it they will come... I wish I had spent a lot more time, including in Year One, to really think about distribution and how growth flywheels work." [FILE] p.82

12. The people. Thirteen in house, mapped to the rungs

Every rung of the funnel has a named owner. Four of the nine channels belong to engineering, not marketing.

Founding AE. Owns rung 4.

Not an acquisition role at $30 a seat. Three jobs in order. Expansion inside paying workspaces, which is seats and credit consumption and the whole of the net revenue retention target. Product qualified accounts, worked at low volume and high quality: the comparable capped this at 100 a month rather than 1,000 and those converted at 3x to 5x a marketing qualified lead, [FILE] General Catalyst p.97. And the top tail, the handful of workspaces large enough to justify a call.

Quota design, or this breaks. Set an explicit share of quota against product driven revenue, "say, 20% of overall quota", because otherwise reps chase the few large deals and ignore 100 small ones, [FILE] p.98.

Marketer. Owns rungs 12 and 14, and half of 6 and 8.

Owns strategy, story, messaging, positioning, the site, the channel portfolio and every contractor. Five funds prescribe exactly this shape, one senior generalist with specialists contracted, and the file calls it the strongest convergence in the research, [FILE] p.13 and p.120. Strategy, story and messaging never leave the building: "if that's not driven by the internal team, it will never be successful", [FILE] Greylock p.102, and "never hire an external firm to help you with messaging, positioning, or to help create your category", [FILE] a16z p.30.

Why the quality of this one seat outranks nearly every other. "The difference between a 50th percentile controller and an 80th percentile controller is not going to change the trajectory of your business, but the difference between a 50th percentile someone doing growth, and 90th, the trajectory of your business is changing." [FILE] p.104

Engineering and product, seven seats. Own rungs 10, 11 and half of 8.

This has to be a named allocation rather than goodwill. Four workstreams: the marketplace listing and two click install path; MCP and editor extensions plus documentation quality; the activation path, defined as a workspace reaching a finding it wants fixed; and the shareable artifact plus a captcha on signup. Roughly one engineer equivalent of the seven, spread across four streams, is an [ASSUMPTION]. The only engineering allocation figure in the file is 60% to 70% of capacity reserved for growth, and it comes from a handbook built on companies between $250M and $30B, so it fails the stage filter, p.93. Do not quote it.

Four part time leaders at 10%. Own rung zero.

About 16 hours a week combined. Spend it only on what nobody else can do. The 50 accounts leading the category. Portfolio introductions asked for by specific portfolio. Design partners, three to six months each at a discount, [FILE] Greylock p.109, screened for tolerance of an unfinished product, with the validation test being "when design partners try to explain the product and almost sell it back to you, that's when you know you've hit something", p.105. On camera and podcast guesting, which the file names directly as the founder influencer pattern: "A US customer needs to know who you are. They go and search on LinkedIn. They go and search on GPT", p.92. And reading the forums directly, weekly, not through a report.

Contractors, attached to rungs. This is the non media budget.

Never contracted. A brand agency, where "$10k/month is on the lowest side" and you will still be deprioritised, [FILE] Index p.55. And the strategy, story and messaging themselves.

Never hired. Specialist marketers in house, prescribed against by five funds. A communications person, which the file puts at post Series B at the earliest and only if the founder will give the work real time, p.102. Anyone above the marketer. And your best community member, who "is more valuable to us as a community member that we keep a looser relationship", p.79.

When to add, with the trigger rather than a date

The capacity gap, named rather than hidden. The one complete funnel in the file ran 400,000 monthly visits with an in house marketing team of three people, none of whom wrote, p.22. This plan needs 161,821 visitors across twelve months with one marketer. That gap is exactly why the mix is weighted the way it is. The organic share cannot be a writing programme, because one person cannot run one. It is a page system built once, a marketplace listing maintained by engineering, an assistant surface that wins on documentation quality, and a community that answers its own questions.

13. The tools, by job

Named where the file names them. The allowance is on the budget line, not per vendor, because vendor quotes are not in the research file.

14. Timeline and phasing

Five phases. Two of them run at the same time, and the file is explicit that they should.

Month by month, what starts and what gets measured

15. The budget, line by line, month by month

No salaries. The thirteen people are already committed at over $3M a year. This budget buys contracted specialists, channels and tools only.

The three checks on the total

One. It stays under the brand threshold. Paid media is $518,968, well under the $1M at which the published allocation rule changes from 100% performance to a brand split. [FILE] First Round p.11

Two. It clears the payback ceiling. $976 of CAC against an allowable $2,160, payback at 5.4 months against a target of under twelve.

Three. The efficiency, stated with its tension. $0.28 of spend per $1 of new ARR. The file's only efficiency anchor is $1.50 per incremental $1 of ARR being achievable and under $1 being hard, from a study of 122 unicorns that fails the stage filter, p.84. By that anchor $3M of new ARR would imply over $4M of spend. Either the revenue target is aggressive for the money, or the product led motion delivers an efficiency no fund publishes. Say it rather than let it be found.

The honest gap. No fund in the research file publishes a marketing budget benchmark for a ten person company that just raised $10M, and the file says so directly at p.125. This total is built up from the funnel and from the published allocation rules, not quoted from a benchmark.

16. How each of the thirteen metrics gets hit

The reporting order, which is not the order above

The file gives an explicit order for reporting to a board or an investor. Sales accepted leads delivered by marketing first. Then pipeline from those leads, as a weighted dollar value and as a share of all pipeline. Then daily over monthly users. Then organic inbound as a share of pipeline. Then cost to get a customer and its payback. Then revenue kept and grown from existing customers, with win rate, as the lagging proof. Clicks, downloads, subscriber counts and repository stars are directional only and never the goal. [FILE] p.14 to 15

Two substitutions worth arguing for: "NPS scores are a better indicator of AI success than domain authority", [FILE] p.101, and one winner stopped tracking repository stars deliberately, [FILE] p.79.

17. Optional items. Reference only, and not in the budget

Listed so that each exclusion is a decision with a number attached rather than an omission. None of this is in the total in section 15.

18. What not to do. The exclusion list, grouped

On money

• No brand agency.
• No visual rebrand as a growth lever.
• Do not trade organic against paid as though they compete for the same money.
• Do not defer marketing analytics past $1M of spend.

On channels

• Do not add a second channel before the first plateaus.
• Do not run outbound before the target is known.
• Do not build a three to six month growth roadmap, because "you're going to miss the window".
• Do not write community or forum posts with a model.
• Do not cold pitch reporters.
• Do not announce the funding round by default.
• Do not copy another company's tactics as a prescription.

On product and community

• No free trials in B2B. Use a money back guarantee. Freemium, which is what the live pricing is, is a different thing and is correct here for a product whose time to value varies.
• Do not let sales touch self hosted or free users in a way that reads as harassment.
• Do not claim community built if you wrote it yourself.
• Do not let large customers pull attention off the community, because "at that point it's too late. You lost it".
• Do not build another generic chat interface. "The prompt is a bug, not a feature".

On reading your own numbers

• Do not chase a growth rate, which "can be juiced through paid marketing, or one-off deals that are hard to replicate".
• Do not read a viral revenue spike as durable demand. Those buyers "will likely churn".
• Do not treat trial revenue churning at 30% a month as real revenue.
• Do not judge content on immediate conversion.
• Do not chase contract value from accounts outside the target.
• Do not optimise for repository stars.
• Do not let split tests substitute for product vision.

On positioning

• "AI-powered" is not a position.
• Do not ship the pitch deck to the website.
• Do not base positioning on a competitive analysis.
• Do not fill the hero copy with jargon.
• Do not lead a first meeting with features before the narrative.
• Remember you are in the market your users think you are in, not the one you say you are in.

19. Risk register. What breaks this plan, in order of impact

20. Sourcing. Every claim to its page

Page numbers refer to the investor marketing research dossier. Live reads are dated. Assumptions are in section 21.

Channels

Cuts and exclusions

People

Funnel, metrics and money

Live reads, dated

21. What is not in any source. The assumption list

This is the shortest and most important section in the document. Everything here is mine and any of it can be challenged.

Three things the research file gets wrong or cannot support, stated so nobody is caught out

One. The file states 6.25% of trials convert while its own table shows 25 customers from 2,500 trials, which is 1%. Use 1%. p.21

Two. The $8 to $10 of pipeline per $1 of marketing spend sits inside a piece subtitled "from $10 million to $25 million in ARR". It passes the date filter and fails the stage filter. If it appears anywhere it has to be labelled as the standard being built toward. p.61 to 62

Three. Two timing figures in the file do not apply to this plan and will be raised if they are not addressed first. The eight year figure is about reaching scale, not about clearing a Series A bar. The four year figure is specifically the product led to enterprise sales transition, which this plan never makes. p.91