Teamday
← Back

Matt Swulinski: The E-Commerce Growth Playbook for AI SaaS

20VC with Harry Stebbings
businessagentsfuture-of-workknowledge-workstartupinterview

Why Matt Swulinski's Paid-First Growth Playbook Matters for AI Companies

Matt Swulinski ran growth at Superhuman and Wispr Flow, two of the fastest-growing product-led growth (PLG) products of the last few years, and now leads growth at Viktor, an "AI employee" startup. In this 20VC conversation with Harry Stebbings he lays out a growth model borrowed from e-commerce rather than classic SaaS, and describes how he runs it with agents instead of a large team.

The e-commerce handbook for SaaS: Swulinski's core claim is that SaaS should copy direct-to-consumer marketing. "My philosophy is that the e-commerce playbook is the right one for SaaS, because if you look at e-commerce, every penny should translate into a purchase or an add-to-cart." That means hundreds of creators, a wide variety of creative, and a balanced set of channels. He also rejects the common advice to wait on paid ads: paid is "the easiest way to validate that you have a PLG" model because messaging, funnels and positioning can be tested in a week.

Measurement before spend: Before the first dollar goes out, the tracking has to work. He points to Meta's match rate and Google's enrichment score: poor conversion signals mean the platforms optimize blindly, CAC balloons, and founders wrongly conclude ads don't work. Unlike e-commerce, SaaS has no plug-in equivalent to Triple Whale, so each startup builds attribution in-house. He estimates 90% of companies skip this step.

Creative is the targeting: After Meta's Andromeda update, creative itself decides who sees an ad, which eliminated the traditional media buyer. "If you have a budget of 100,000 in Meta, you will probably need at least 400 to 500 new creatives per month. Otherwise, you'll stagnate." Those come from a creator program paying a share of ad spend, several agencies, and an in-house team. He says teenage creators earn $20-30K a month, that 80% of results come from 20% of creators, and that AI is best used for variations: fully AI-generated video is "garbage" and should be roughly 5% of an account.

Google as the main engine: At Wispr Flow, Google Ads (unbranded search, PMax, YouTube) was the top channel because one platform covers the whole funnel and allows CAC control by country. Short YouTube videos educated buyers on a hard-to-explain product, then fed search and display conversions. He reuses vertical UGC on YouTube by dropping it into a static Figma template.

Systems thinking is the new hiring bar: "Many teams are trying to make their teams AI native and are failing. Why do they fail? The people in those positions don't think systemically." His interview test is whether a candidate has built a self-improving workflow, not whether they chat with an assistant. His own example: an agent that handles newsletter sponsor requests end to end, negotiating, generating tracking links, reading performance, and deciding whether to renew. He ran this from a folder of skill files and markdown in Claude Code, on cron checks at 9:00, noon and 6:00.

5 Takeaways from Swulinski on Growth in the Agent Era

  • Validate with paid on day one - A $100K budget on Meta and Google, with lifecycle messaging, is enough to test whether people want the product; organic content runs in parallel, not first.
  • Fix tracking before scaling - Server-side conversion signals and a real attribution stack decide whether paid works at all.
  • Optimize the deepest cheap event - Use downloads early; move to the bottom-of-funnel event (for Viktor, adding the app to Slack or Teams) as ARPU justifies it. Count free trial credits as fully loaded CAC.
  • Diversify the acquisition mix - Referrals tied to usage limits, affiliate programs paying 10-15% revenue share (10-15% of Viktor's acquisition), and YouTube reviews that feed answer-engine citations.
  • Hire for systems thinking - A 3-5 person influencer team becomes one person running agents; A-players become S-players and weaker ones fall behind.

What This Means for AI-Powered Marketing Teams

Swulinski's bet is that within three years companies look like boards where 20% of the work is strategy and 80% of execution is done by agents, with roughly the reverse today. The concrete lesson for marketing organizations is that the constraint is no longer channel expertise but whether someone can map a role as a system, hand the repeatable parts to agents, and keep a feedback loop running. Teams that do this will outrun larger incumbents that only bolt a chatbot onto existing jobs.