Sandy Diao on Why Data-Driven Growth Killed Descript's Web Launch

The Web Launch That Tanked Everything

Sandy Diao spent 2 years convinced that Descript's desktop-only constraint was killing growth. The funnel data screamed it: massive drop-off between signup and app download. When the team finally shipped a frictionless web version, she thought she could take a vacation.

Conversion rates dropped across the board. Activation down. Monetization down. Revenue down.

We launch web, we drive traffic to it, conversion rates drop across the board, activation goes down, monetization goes down, revenue goes down. And I'm just scratching my head wondering, like, how could this possibly happen? Web is frictionless.

— Sandy Diao

The lesson: someone who downloads a desktop app has already invested time and intent. They're committed. Web visitors click around and bounce. The data showed a problem, but the customer context told a different story. This is the difference between being data-driven and data-inspired.

How to Build a Growth Hypothesis That Doesn't Suck

Most growth teams treat hypotheses like shopping lists. They see a competitor doing something, add it to the backlog, and call it strategy. Sandy saw this firsthand at Pinterest when the team copied Facebook's 7-part onboarding email series. Weeks of work. Zero lift.

A good hypothesis has two components: a data point and a customer story. The data gives you the "what," but the story gives you the "why." Without both, you're just cargo-culting tactics from companies serving totally different markets.

A good growth hypothesis always has those two components to it. The first component is an insight, and the insight has to be data-inspired or backed by some kind of context. And then the second part of a good growth hypothesis is they have some kind of story around the customer and your understanding of what that data point means.

— Sandy Diao

At Descript, this meant understanding that desktop app users weren't dropping off because of friction—they were self-selecting for commitment. The company wasn't data-driven; it was data-inspired. And that distinction saved them from chasing ghosts in the funnel.

Channel Fit Comes Before Channel Scale

Everyone wants to know their ROAS. Sandy says you're asking the wrong question. Before you optimize for 3x returns, you need to know if a channel can work at all. And "work" means two things: conversion rates that match your intended customer, and incremental spend that drives proportional returns.

PR and influencer hits feel like wins, but they're one-off blasts. You can't build a system around them. Real channel fit means you can turn the dial and predict what happens next.

Channels don't die because the platform stops working. They saturate because you've exhausted an audience. At Descript, early paid campaigns targeted podcasters. When that audience capped out, the team expanded to short-form and long-form video creators. Same channel, new audience, step-function growth.

If you're starting out and you hit breakeven or better, you're in good shape. Then you optimize. Most teams kill channels too early because they expect 3x ROAS out of the gate. It never happens. You're juggling audience targeting, creative formats, and budget allocation. Give it time.

Static Ads Still Beat Video on Meta

Video content dominates every growth playbook in 2024. But on Meta, static ads still outperform video for driving app downloads and deep product engagement. Why? Because video interrupts the entertainment flow.

Think about your own behavior. You're scrolling, you hit a video ad, you watch it—are you really going to pause your feed, download an app, and complete an onboarding flow? Unlikely. Static ads let you open a new tab, browse, come back later. Less interruptive, better for conversion-optimized systems.

If you actually think about it, a conversion optimization system is basically saying, I'm trying to find a user who's likely to take a certain action. When you're watching videos as a consumer yourself, how likely is it that you're going to basically click a button, pause all the entertainment that's going on in your social app, and then go far into a product software experience?

— Sandy Diao

Creative makes the biggest difference in paid performance. More than audience targeting, more than budget. Test format before you scale spend.

Affiliates Drove 25% of New Users at Descript

Descript didn't seem like a natural fit for in-product growth loops. It's often a solo experience—one creator, one project. But Sandy noticed something: customers were publishing full-length tutorial videos on YouTube without being asked. Some linked back to Descript, some didn't.

The team reached out and asked creators to add links. Traffic spiked. So they launched an affiliate program with a 15% recurring commission. The ROAS was incredible, and the program scaled to self-service. Affiliates drove over 25% of new users that year.

The insight: content creators have built-in audiences. Every video they publish is a distribution opportunity. Descript didn't need traditional referral loops—they needed to align incentives with creators who were already talking about the product. That's power law distribution. Focus on your unfair advantages, not every channel under the sun.