ElevenLabs is adding ARR nine times faster than it did in its first 20 months

Divide each leg by the months it took and the curve gets concrete: about $5M of new ARR a month across the first twenty ^[23], $10M a month across the next ten ^[24], $26M a month across the five after that ^[25], and roughly $45M a month across the most recent six ^[26]. One month of the fourth leg does the work of nine months of the first ^[27]. The last leg on its own, 330 to 600-plus in half a year, compounds to something near 230 per cent year on year if it holds ^[28].

The two commission rules pull against each other, and that looks deliberate. Quota at twenty times base salary puts a rep's base at five per cent of the number they carry ^[14][29], the plan is uncapped, and Reina says the company committed publicly to lowering quota if the market showed the number was set wrong ^[14]. Commission is then paid only on recurring contracts, and a proof of concept pays zero no matter its size ^[15]. So the plan withholds money for work that ended in a pilot while paying money for work an agent finished, because when an agent unlocks revenue on an account the human who owns that account still collects ^[16]. Reina's framing is that you are paying twice, once for the agent and once for the person, and that the alternative is a rep quietly working against the system meant to help them ^[16]; he now recommends the rule to his portfolio companies ^[17].

Which leaves the saving unlocated. Double-pay holds commission expense per dollar of new ARR flat, so the money has to come out of the SDR and customer success headcount never hired, and the published case rests on two demonstrations rather than a cost line: an AI SDR that calls an inbound back immediately converted better than a human replying inside thirty minutes, and an AI customer success manager working the SMB and mid-market longtail found upsells nobody had time to chase ^[20]. No figure is given for headcount avoided, nor for net revenue retention, gross margin, churn or customer acquisition cost ^[22].

The geography mechanic is the one I would steal, or rather the tax half of it. Every market launch got a written thesis, a channel mix and an expected result inside three to six months before it opened ^[18], and outside the core markets the channel was chosen by the tax code, since withholding taxes and local invoicing made direct selling worse economics than handing volume to a reseller ^[19].

On the grants program the return is stated and the cost is not. Over ten per cent of enterprise revenue came out of the cohort for a long stretch ^[11], against a cost given only as three months of usage per account ^[12], with no dollar figure, no enterprise revenue line and no share of total ^[31]. That is a ratio missing its denominator, which is a different object from a payback.

Two other readings fit the same numbers. Voice model demand may have pulled the curve and the five decisions are simply the ones remembered afterwards, and Reina himself calls the grants play a gamble they had no idea would work ^[32]; or the playbook is partly product, its author having been employee number four and the company's first investor before moving to a $15M solo GP fund whose pitch to founders includes this material ^[6][8]. My read is that the comp rules are the transferable part, because you can test them in your own plan next quarter, while 41 months to $600M-plus is a benchmark to admire rather than underwrite ^[5]. What would break that read: a grants cohort whose ten per cent traces to a handful of accounts that would have signed anyway ^[11], or a competitor matching the curve on a capped plan that pays commission on pilots.