Swulinski: the e-com paid-ads playbook is SaaS's $100M growth engine

Summary

• Matt Swulinski’s core thesis is that “the e-com playbook is the right playbook for SaaS”: hundreds of UGC creators, massive creative volume, and paid spend where “every single cent needs to equal a purchase or an add to cart.” He applied it to scale paid at Superhuman before the Grammarly acquisition and then at Wispr Flow — “that’s really what put it on the map” — because “distribution to me is the only moat” when you open X and see “100 new products, five in your category, two that have absolutely just cloned your website” daily.
• Against the consensus that paid is “a dangerous drug” to defer, Swulinski says start it “right away” — “paid is the easiest way to validate that you have PLG.” On a $3–5M seed, ~$100K focused on Meta and Google, with lifecycle also spun up, can validate messaging, funnels, and positioning “all within a week”; but 90% of companies fail to set up the martech first, leaving Meta with “ghost people” and misdiagnosing bad tracking as “paid doesn’t work for me.”
• Post-Andromeda, “the creative is the targeting” — manual audience-setting and media-buying tinkering were substantially displaced, making creative strategy the central job. A $100K/month Meta budget needs “at least 400 to 500 new creatives a month” or it will plateau and get outcompeted; Victor runs a paid-percentage-of-spend creator program where “kids that are, like, 17, 18, 19 are making 20, 30K a month just making a couple ads for us.” Full AI-generated video is “slop” — maybe 5% of the account at most.
• On scaling, deliberately blow the engine up: Wispr 5X’d its budget from one month to the next to find the ceilings, then pulled back with a map of what was incremental. Google Ads was Wispr’s best channel — “everything performs there” — with YouTube education videos feeding non-branded search and PMax; X ads are the anti-channel: “I’ve yet to meet a SaaS head of growth or performance marketer that says that X ads print.”
• Referral and affiliate programs should be tangible, usage-aligned, and placed at the “magical a-ha moment” — then paywall, “because you wanna open up your pocketbook.” Victor’s referral program can give referred-company revenue share in credits, with a 20% example; its separate affiliate program pays 10–15% revenue share, and some affiliates earn $20–30K/month. Affiliate drives 10–15% of monthly acquisition; free trial credits must be counted in “fully loaded CAC” or “you’re not really calculating your acquisition cost.”
• His hot take on teams: “probably fire most of your marketing team that is not a systems thinker” — the JD changed and companies are “brute forcing people into these new JDs.” Fewer than 1% of candidates pass his bar (“chatting with the thing is not a workflow”); he ran Whisper Flow’s entire $3–5M budget execution solo via a Claude Code “marketing OS,” and predicts companies become “board of directors” structures within three years — 20% human strategy, 80% agent execution — with marketing unicorns potentially commanding ML-researcher-style pay.
• The investor-relevant gap he flags: SaaS has no out-of-the-box equivalent of e-com’s Triple Whale/Elevar attribution and conversion-tracking stack — “I’m waiting for startups to be made that fill this gap” — and Stebbings offers to fund an AI-systems bootcamp “today with millions of dollars.” Meanwhile healthy scaled mix is 35–45% organic; if turning paid off craters growth, “that means you have other problems.”

Deep dive

1. The e-com playbook is the right playbook for SaaS

• Swulinski’s through-line from Superhuman: the product was hyper-refined, but growth was founder-led referral while “e-com’s been doing the UGC programs, hundreds of thousands of ads” — and internal resistance to paid was real. His conversion: “my philosophy is that the e-com playbook is the right playbook for SaaS,” where “every single cent needs to equal a purchase or an add to cart,” applied late at Superhuman pre-Grammarly acquisition and then wholesale at Wispr Flow — “that’s really what put it on the map.”
• The justification is competitive: “distribution to me is the only moat” — open X any day and there are “100 new products, five that are in your category, two that have absolutely just cloned your website.”
• On the agentic shift, he argues PLG discipline transfers: the same rigor applied to a perfect self-serve human funnel now applies to “how is an agent doing research, how is it picking the tools and the APIs” — so “the best companies focused on PLG right now will also be best positioned for the agentic layer.”

2. Start paid immediately — it’s the fastest validation engine you have

• Stebbings’ challenge — everyone says wait, paid is “a dangerous drug that you can get hooked on.” Swulinski: “Right away… paid is the easiest way to validate that you have PLG.” Brand and organic “just takes too much time”; on paid “you can refine messaging, do creative testing, test your funnels, test your positioning all within a week.”
• Channel discipline: “the core three of any acquisition engine — Meta, Google, and lifecycle,” covering video intent, search intent, and the lifecycle “net” (email/SMS/push). “You can scale to your first million, 10 million ARR just off of those three things.” Everything else — TikTok, Reddit — is the agent trap: “if you do a million channels and you do them all poorly, it’s not gonna help you out.”
• Sizing: on a $3–5M seed, ~$100K is “a good start budget to essentially validate that if we put money, people are gonna actually want my product” — modulated by ARPU and funnel length.

3. Before the first cent: 90% of companies are “spending into the air”

• His flagged market gap: e-com has Triple Whale (out-of-the-box mutual-exclusivity attribution deduping Meta/Google/lifecycle double-counting) and Elevar-style conversion tracking in 15 minutes; in SaaS “all of this is homegrown at every single startup” — ClickHouse, Hex, first/last-click, built by engineers. “I’m waiting for startups to be made that fill this gap.”
• The mechanism of failure: platforms do exactly what your signals tell them. With a 50/50 match rate on Meta or an unmaxed enrichment score on Google, converters are “ghost people” — the algorithm targets randomly, CAC spikes, and “you’ll say, ‘paid doesn’t work for me.’ Most of the time people haven’t done the actual setup correctly before they can say that.”

4. Optimize for one event, and get honest about fully loaded CAC

• Pick the single most important event and buy it: at Flow, the download (desktop client plus iOS made tracking hard); at Vanta, the far-down-funnel act of adding Vanta into Slack or Microsoft Teams — “we can have a couple thousand dollars in CAC and the unit economics still work” at high ARPU. Early on, focus purely on acquisition cost; product owns retention.
• On ratios: 1:1 LTV:CAC can be acceptable depending on how much has been raised — “you want to run one to one to get as many users using your product and believing in you as soon as possible” — with ~3:1 the eventual target. With token costs, LTV:CAC “is usually not enough”: use LTV gross profit, and count Modal and inference, not just “my Anthropic bill” — “have a good finance leader that’s gonna gut check” the cost items.
• The line worth underwriting to: “we call this fully loaded CAC. Our free trial credits get summed with marketing spend. If you’re not doing that, then you’re not really calculating your acquisition cost.”

5. After Andromeda, the creative is the targeting — and you need 400–500 a month

• Meta’s Andromeda update “changed the targeting algorithm where the creative is the targeting” — manual audience-setting and media-buyer tinkering “went out the door,” with the job shifting toward creative strategy. His volume rule: “let’s say you have 100K Meta budget, you probably need at least 400 to 500 new creatives a month, otherwise you’re gonna plateau and get outcompeted.”
• How Victor gets there: a creator program paying a percentage of ad spend (creators do 3–4 videos a week, a couple hundred of them), five agencies, plus in-house — and “we have kids that are, like, 17, 18, 19 that are making 20, 30K a month just making a couple ads for us.” UGC never appears on the creator’s page, so anyone can make it; ~30% of spend should go to partnership ads. Yes, 80/20 holds — “that’s also just how the algorithm works: it finds the best ad and pumps all the spend at it.”
• What wins: pattern disruption — “a rough shake of the camera that seems like it’s a mistake, but it got your attention” — but the real unit is the portfolio: “different ages, different genders, different settings, different hooks… If you only focus on what works and you pump just that, performance will crater.”
• On AI creative (Stebbings cites Cliff Weitzman, whom he identifies as the founder of Speechly, turning one ad into thousands of variants): variations are “the superpower of AI in creative,” but full AI videos “are slop” — maybe 5% of the account. Both agree the algorithms may deprioritize AI content: “you’re 100% right.”

6. The landing experience, page speed, and the two-to-three-week verdict

• Prerequisites before launch: analytics, conversion tracking, and at least 50 conversions already flowing so the algorithm knows the best customer. Then “within two to three weeks… you’ll be able to say ‘yes, I can start printing money today’ or ‘I have to go to the drawing board’” — but give the full system three months, because ad, copy, landing page, page speed, and App Store screenshots are “micro levers that compound.”
• Stebbings’ shade at PostHog’s site (“Beijing’s version of UI”) gets a defense: “that’s a pattern disrupt thing for them… the target audience appreciates the weirdness.” The general test is brutal though: “If I read nothing else but your headline, do I know what you do?” — and no main CTA “a scroll and a half away.”
• Page speed matters for AEO and SEO: crawlers make “an easy call to see how long your page takes to load,” it’s core to top-10 ranking, and “any second of improvement is a drastic jump in conversion.”

7. Go as hard as you can until it blows up — then scale back with the map

• The Wispr method: “go as hard as you absolutely can and then see it blow up, and then understand where and how it blew up, pull back, and use that as information.” They 5X’d budget from one month to the next rather than tuning slowly — you don’t have six months when you need 30–40% month-over-month growth — and learned “Meta needs to be second to Google… these newsletters were shit, this podcast doesn’t work.”
• Chart total spend against acquisition and ARR, accounting for the lag between spending and conversion — Matt gives a 14-day example — then ask how elastic the spend-to-revenue relationship is.
• Google Ads was Wispr’s engine: “non-branded search, PMax, YouTube ads, everything performs there,” with far more fine-grained control (India vs. US CAC) than post-Andromeda Meta. For a hard-to-explain product (“I talk, it turns to text — but where?”), 30-second-to-one-minute YouTube education videos ran to hundreds of millions of impressions, then “fed into non-branded search, PMax… all those pieces worked together.”
• YouTube deserves its own program: best ads are 16:9, so Victor had Victor build an app wrapping any Story-format UGC video into a static template with customer logos, G2 rating, and CTA — “we turned all of our Story ads into YouTube ads… Welcome to YouTube.” Scripts and filming are a separate team from UGC.
• The discipline question: “how incremental is your spend?” — once it works “it’s really easy to dial it up to 13 on a scale of 10” and hand free money to Meta and Google. North of $1M/month, run an MMM with holdouts; at scale, organic mix should be 35–45%, and if turning paid off kills growth, “that means you have other problems — you have forgotten about the other half of the job.”

8. Why Superhuman asymptoted, and the tagline fight over “AI employee for everyone”

• Superhuman’s ceiling: premium product, immediate paywall, founder ICP — “the early adopter tech founder is not an infinite audience.” The unlock is proactively opening ICPs (e.g., turning Superhuman into a sales engine, “recent opens” being his favorite feature). When Harry compares the email competitor referred to as “Fixed Scale” by him and “Fixer” by Matt with Superhuman and asks whether Superhuman “fucked up paid,” Matt initially says “yes,” then qualifies the comparison by pointing to the different market, Meta/e-commerce playbook, enterprise deals, and sales-led growth. Notably, Flow had “a very large component of our ARR was enterprise before we had a single AE” via self-serve team licenses.
• Scaling is audience-by-audience: Victor launched broad on purpose to see who showed up (agencies, e-commerce brands, SMBs), then worked the list in order, building a repeatable funnel per ICP. Modern ICP research is a prompt: “simulate this person — what do they read, how do they make decisions?” then build creative, landing, and onboarding for that person.
• Stebbings’ pushback as an investor in the company — “the AI employee for everyone, I don’t know what the fuck that means… are you doing my tax?” Swulinski’s defense: “there’s no other positioning that really makes sense” against Claude and ChatGPT, and now with use cases banked they can invert to “I hired Victor and here’s what he did.”

9. Referrals and paywalls: make it tangible, and strike at the magic moment

• Superhuman’s give-a-month/get-a-month worked because it was tactile and discoverable — some users banked “hundreds of months they were never gonna pay.” Align the ask to usage limits: at Flow, surface the referral right as the 2,000-word limit approaches — “that k-factor is what makes or breaks that.” Victor pays creators via CPM in credits for LinkedIn posts and gives referred companies a percentage of recurring revenue in credits, with a 20% example. This is sensible when “eight-person teams are spending $15,000, $20,000 a month” and Victor is willing to spend $1–2K CAC.
• The failure modes bracket a sweet spot: either intangible (“refer your friends… you get swag”) or “hyper-complicated 20 tiers” of gamification no one uses.
• Paywall placement follows the “magical a-ha moment”: get users there fast, keep them in the honeymoon, “and the moment you have one workflow that goes ’this is game-changing’ — paywall. Because then you wanna open up your pocketbook.” That’s what drove Flow’s organic LinkedIn rocket-ship moment.

10. AEO runs on YouTube reviews; X is the polluted channel; affiliate is the sleeper

• AEO’s most important inputs aren’t a page factory — companies are producing “100 to 200 pages a week” of “generated AI slop” — but YouTube, Reddit, and the social narrative. Long-form YouTube reviews “rank for the long tail” and are “a really high citation on ChatGPT,” making them a top early lever. TechCrunch and Product Hunt still matter, but as “founder initiation” and citation fuel, not primarily as traffic.
• TikTok: “in any of the three core companies I’ve been at so far, it hasn’t worked yet.” X is his most polluted channel — identical shock-and-awe launch videos (“we spent way more time on our launch video than our product”) — and “I’ve yet to meet a SaaS head of growth or performance marketer that says that X ads print. If someone has, please tell me.”
• Most underappreciated: affiliate — external non-customers with an audience, paid 10–15% revenue share. At Victor it’s “the highest ROI channel,” driving 10–15% of monthly acquisition, with some affiliates making $20–30K a month; “if they feel like they can earn money even before they’re your customers, they will become your customers.”

11. Fire the non-systems-thinkers: one person plus agents beats the five-person team

• His hiring 180: a year ago he’d want the world’s best Meta media buyer with 10 years of scale; now an AI-native systems thinker “plus experience will outcompete someone that just has experience.” Hot take: “probably fire most of your marketing team that is not a systems thinker… stop brute forcing people into these new JDs.” The tell inside teams: “the A players are becoming S-tier players and the B players are becoming D players.”
• The bar is a quasi-engineering interview — how do you use AI workflows, including in your personal life? Bad answer: a ChatGPT project you chat with — “chatting with the thing is not a workflow.” Good answer: self-improving loops like his Whisper newsletter system, where a Claude Code agent triages sponsor emails, negotiates rates, ingests contracts, writes copy, builds tracking links, and decides renewal from historic performance across 70–120 newsletter providers. Fewer than 1% of candidates clear it — but that’s fine, because “a three, four, five person influencer team today is one really good person with an amazing suite of agents.”
• Proof of concept: until December 2025 he was “the only person doing execution on a three, four, five million dollar budget” at Whisper — a million a month on Google, hundreds of thousands of keywords — running “90 to 95% of my work through the AI” on a Claude Code marketing OS he built without knowing how to code, adopting Claude Code in September “before it was good.” His compounding hack: a “session end” skill distilling every session into an Obsidian vault so “every session compounds” — “what do you have on your computer saved in your AI output… is the treasure trove for the next couple years.”
• The macro bet behind Victor: today work is 80% manual/20% agents; “in three years companies will essentially be like board of directors — 20% strategy and thinking, agents do 80% of the execution.” Incumbent B2B CMOs who don’t get this will see growth “stand still” while adopters take share; marketing unicorns like Tobin, who built Whisper’s original UGC viral program before Polymarket, may get their ML-researcher pay moment “over the next year.” Stebbings’ coda — he’d fund a six-week, $10K AI-systems bootcamp “today with millions of dollars”; Swulinski: “you’re 100% right… you can do and then understand, versus sit in a room and learn.”

Verification Notes

• The transcript alternates between “Whisper Flow” and “Wispr Flow”; the email competitor is transcribed as “Fixed Scale”/“Fixer,” so those names are not further normalized here.