Growth systems

Distribution Is a System, Not a Campaign

Matt Swulinski’s case for AI-era SaaS growth is deliberately unfashionable: product-led growth still matters, but product love without a measured, creative-intensive distribution machine eventually runs into a ceiling. The machine is not “performance marketing” bolted on late. It is the operating system that turns a good product into repeatable demand—and tells you when not to spend.

Source: 20VC with Harry Stebbings — “How to Build a $100M Growth Engine: Lessons from Wispr Flow & Superhuman”, with Matt Swulinski · 80:47 · Video ID: bm8rMM4Bxz8

Swulinski comes to the argument as an operator rather than a theorist: he was growth hire number one at Wispr Flow, previously worked on growth at Superhuman, and now leads growth at Viktor, an AI coworker product. He also describes having run an agency across e-commerce, SaaS, and enterprise B2B. That mixed background is the point. His proposed operating model imports e-commerce’s ruthless loop—creative volume, direct response, and conversion instrumentation—into SaaS, while preserving the product experience that makes a self-serve motion worth amplifying.

PLG does not disappear when agents arrive; the buyer path changes

The first correction is to the idea that agents somehow invalidate product-led growth. Swulinski argues that the old PLG disciplines—self-service discovery, low-friction evaluation, a product that sells itself through use—remain intact. What changes is that a meaningful portion of evaluation will increasingly be performed by agents acting for users or teams. Companies therefore need to understand how their tools and APIs are found, compared, and selected without a human clicking through every step.

That is a subtle claim. It is not “replace the funnel with AI.” It is a warning that the funnel now has another participant. A company that only optimizes the human-facing landing page may miss the structured signals, documentation, external reputation, and product clarity that help an automated researcher recommend it.

Superhuman provides the counterweight to a product-only story. Swulinski recalls an exceptional product, a strong founder, and word of mouth that worked—until the company reached the edge of the audience reachable through that motion. Craft created the right to grow; it did not eliminate the need to acquire beyond the founder’s natural graph.

Borrow e-commerce’s mechanics, not its indiscriminate spending

Swulinski’s sharpest prescription is that SaaS should learn from e-commerce. E-commerce teams are accustomed to treating paid acquisition as an accountable production system: creator networks supply user-generated creative, teams make many variants, and performance is observed all the way through add-to-cart and purchase. In his telling, SaaS companies often delay paid acquisition because it feels less pure than organic or product-led growth—and then learn about positioning, funnel friction, and audience far more slowly than they need to.

Paid, used well, is not simply an amplifier. It is a rapid research instrument. A team can test a message, a creative angle, and an onboarding path in a week; a podcast appearance, content program, or long editorial cycle may take much longer to yield comparable signal. His practical default is a narrow core: Meta, Google, and lifecycle—email, SMS, push, and the in-product paths that turn an acquired user into an activated one. In his view, that combination can carry an early company to its first $1M–$10M of ARR before it scatters attention across every possible platform.

The necessary qualification: “start paid right away” does not mean buy traffic before the product has demonstrated any pull. Later, Swulinski names roughly 50 conversion events as the minimum useful signal for an algorithm to learn from. The coherent version of the advice is: begin testing early, but scale only once tracking and conversion signal make the tests interpretable.

The first growth hire may be measurement infrastructure

Before a dollar reaches Meta or Google, Swulinski wants a company to build the plumbing. His deliberately provocative estimate is that 90% of SaaS companies get their analytics wrong before spending: attribution is assembled from scratch, server-side and client-side events are incomplete, and subscription events do not reliably make it back to the ad platform. E-commerce enjoys more turnkey tooling; SaaS too often treats that operational gap as a detail.

The consequence is more than a messy dashboard. If Meta matches only about half of the people who subscribe, its optimization model cannot learn who is actually valuable. CAC rises, the company concludes that paid does not work, and the real problem—an incomplete feedback loop—is hidden beneath a channel diagnosis.

The conversion event should initially be the most consequential event the business can measure reliably: a subscription, a trial, or a download. As product behavior and tracking mature, the optimization target can move further down funnel. Viktor, for example, optimizes around a user finishing onboarding and adding the product to Slack or Teams, rather than treating a shallow click as success.

Economics also need a vocabulary suited to AI products. Swulinski allows that a well-funded company may accept a 1:1 LTV:CAC ratio while it is building distribution, with something closer to 3:1 as a later aspiration. But when inference or token usage is material, raw LTV is flattering. The more honest comparison is LTV to gross profit after those variable costs. Free trial credits belong in CAC, too: they are marketing spend even if they are delivered through the product.

Creative has become the targeting layer

The e-commerce import becomes most concrete in creative production. Swulinski says that a $100,000 Meta budget may require 400–500 new creatives per month to avoid plateauing. At Viktor, he describes creators making three or four videos per week, with agencies and an internal team extending the output. The important unit is not a single “winning ad.” It is the rate at which a team can manufacture and learn from distinct creative packages.

That emphasis reflects his reading of Meta’s current direction: platform automation is taking over more conventional media-buying decisions, so the creative itself increasingly does the audience-shaping work. A camera shake, an awkward opening, or a visibly unpolished first second can interrupt a feed pattern; but the durable lesson is not to imitate one accidental-looking winner. The portfolio needs variation across people, ages, settings, hooks, and narratives. Copy one successful package too literally and, he says, performance eventually “craters.”

Creator economics are part of the capacity equation. He recommends spending roughly 30% of budget on partnership ads and observes that some teenage UGC creators make $20,000–$30,000 per month. The platforms themselves concentrate spend behind the few ads that work—an 80/20 pattern—so a growth team needs a broad enough creative surface to discover those few without betting the entire account on them.

Automation belongs in the factory, not necessarily in the face of the ad.

Swulinski is skeptical of fully AI-generated video, which he thinks audiences can still spot as “slop”; he would limit it to perhaps 5% of an account. Yet he expects agents to do 80% of growth execution within three years. Those statements are compatible when the division of labor is clear: automate research, production coordination, variants, routing, reporting, and feedback; preserve real human voice and taste where authenticity is the product.

Scaling means finding the ceiling—and measuring the parts paid did not cause

His growth playbook is aggressive but not naive. A team should be able to see a possible path to viable economics in two or three weeks, then spend a full three months improving creative, copy, landing experience, page speed, and app-store assets. This is not a one-week verdict on a channel. It is a bounded learning period with enough iteration to distinguish an unworkable market from a broken implementation.

Once a program works, Swulinski favors deliberately pushing into the ceiling: chart spend against acquisition or ARR with the relevant conversion lag, multiply budget dramatically to expose saturation, then pull back and repeat. He describes doing exactly this at Wispr Flow, increasing spend fivefold to learn how elastic the system really was.

But he gives the counter-rule in the same conversation: without an incrementality model, a company can simply be donating money to Meta or Google. Holdout tests and marketing-mix modeling matter because observed conversions are not identical to caused conversions. A spike in dashboard attribution can coexist with demand that would have arrived anyway. The operating standard is therefore not “scale whatever the platform reports,” but “stress-test demand, then measure the lift honestly.”

Channels compound when they reinforce a narrative, not when they are added as a checklist

Wispr Flow’s channel mix is a useful example of this portfolio logic. Swulinski says Google Ads—not conventional SEO—was the principal engine, spanning non-branded search, Performance Max, and YouTube. Short educational video built demand; later search and display activity harvested some of that demand. In other words, the channels were not independent silos with separate credit claims.

The same logic governs audience expansion. When a channel fatigues, he does not recommend immediately abandoning it. Keep evergreen activity running, then unlock a new audience with its own creative, funnel, and economics. At Viktor, the sequence includes agencies, e-commerce brands, and SMBs. The positioning began horizontally—an “AI employee for everyone”—partly to discover who showed up, then moved toward verticalized messages and flows. Harry Stebbings presses him on whether the broad phrase is unclear; Swulinski accepts the ambiguity as the cost of trying to establish a category. That is a real trade-off, not a slogan: category breadth can create optionality, while vertical clarity is what turns a discovered ICP into conversion.

Make the product’s constraint do referral work

Referral programs fail when they feel like a generic growth widget. Swulinski’s examples are effective because the reward is immediate, understandable, and offered at the moment of need. Superhuman’s “give one month, get one month” created users with hundreds of credits. Wispr Flow surfaced sharing near a user’s word cap, when additional capacity had direct value.

Viktor follows the same design principle near token limits: users can earn credits by posting on LinkedIn, or receive a revenue share for companies they refer. He reports that some eight-person teams spend $15,000–$20,000 per month on the product. Whether or not that figure generalizes, the design principle does: the best referral prompt is not a detached invitation to advocate. It converts a present constraint into a beneficial, low-friction exchange.

Paywalls should follow the same logic. Put them after the magic moment, not in front of it. A user needs to have experienced the reason to pay before a restriction can clarify value rather than simply stop evaluation.

AEO needs outside evidence, not a second pile of keyword pages

Swulinski’s view of answer-engine optimization is similarly broader than a content-production hack. Generating pages in the style of old SEO will not be enough. The system needs useful site content, but it also needs external narrative that can be cited and trusted: YouTube reviews, Reddit discussions, and credible PR coverage. In this model, PR is not merely a temporary traffic spike; it is a durable third-party reference point.

His organic diagnostic is a useful guardrail against paid-channel triumphalism. At maturity, he wants roughly 35–45% of acquisition to come from word of mouth and organic discovery. If paid is paused and the business collapses, that may reveal neglected SEO, AEO, reviews, or brand demand—not a reason to declare paid intrinsically bad. Paid should create learning and reach; the rest of the portfolio should make that reach increasingly less dependent on buying every marginal visitor.

The durable advantage is a growth team that can build feedback loops

Swulinski’s organizational argument is harsher than his channel advice. He wants “specialized generalists”: people who can understand the moving parts of a system, identify routine work, automate or reroute it, and use outcomes to improve the next cycle. His interview test is not whether someone can chat with ChatGPT; it is whether they can map inputs and outputs and construct an AI feedback loop. He says fewer than 1% of candidates show the depth he wants.

His newsletter-sponsorship system at Wispr is the tangible case. It identified inbound opportunities, sought rate cards, researched audiences, made a first-pass negotiation, handled contracts, copied links and tracking, and used historical results to recommend whether a sponsorship should continue. The value is not that one workflow became magical. It is that a formerly serial, coordination-heavy job became a measurable system with human judgment concentrated at the decisions that mattered.

His provocative claim that companies should “probably fire most” marketers who are not systems thinkers is better read with his own qualification: let people build systems first, then observe who becomes 10×. The useful question is not whether a role sounds traditional. It is whether the role is connected to a feedback loop that can improve, compound, or be partly delegated.

The operating checklist

Product craft makes the system worth compounding

The memorable line is that distribution is the only moat. Taken literally, it overstates the case: distribution without a product people love buys short-lived attention, and Swulinski’s own examples depend on a magic moment, usable conversion signal, and a reward users actually want. His stronger contribution is to reject the false choice between product and growth.

In markets where products can be copied quickly and AI lowers the cost of execution, distribution becomes the system that discovers an audience, translates value into many credible creative forms, converts attention through a working funnel, and feeds learning back into the next iteration. Product craft is what lets the system retain the users it reaches. Neither half compounds properly without the other.

Editorial note: Claims, figures, and operating recommendations above are attributed to Swulinski’s reported experience and views in the source interview; they are not presented as independently audited results.