Product Updates
Custom Verdicts For Every Cohort: Hero Lab's Next Evolution
Move beyond a simple "winner." Hero Lab has the guts to tell you which hero wins and the smarts to tell you exactly what to send, if something else is better.
Hero Lab introduced cohort-level creative scoring earlier this summer. Until now, it generated and tested email hero images. And it delivered dozens of on-brand variants and predicted which ones would actually perform before you ever hit send.
That’s a pretty big feat, yes. But now there's more!
Instead of averaging one winner across your whole list, Hero Lab runs each image through a per-cohort competition. A cohort is a group of subscribers who behave alike, say, recent buyers of a certain category, or people who only ever click on sale messaging. Now in Hero Lab, every cohort receives its own favorite creative, instead of settling for a blended average. This is powered by the same predictive scoring model behind Surge that runs on data from how 10,000+ ecommerce brands perform in the wild.
See the Winners, See Who Sees What
The Backstroke interface now offers a few new views:
Winners shows the top-performing creative for each cohort at a glance.
Who Sees What lays out exactly which image goes to which audience.
You can also switch over to All Variants to see everything Hero Lab generated, download images or save any of them straight to your library.
Today, Hero Lab starts to answer a harder question: not just which new creative wins, but whether you should send anything new at all.
We're rolling out Control Benchmark + Verdict, the first version of Hero Lab that can tell you to keep what you're already sending. Here's how it works.
You designate a control. This is the hero you'd send if you weren’t doing any AI testing.
That gets scored against every generated variant on the same model and the same cohorts.
Hero Lab then recommends which hero image to send, whether that is your control image or another it generated.
Think of the Control Benchmark image as the bar, instead of a contestant. It skips Hero Lab's text/crop treatment entirely, so a finished design gets scored exactly as your team made it.
From there, the Winners view leads with a single, plain-language verdict, instead of a ranked list of scores. In this view, Hero Lab will tell you one of four things:
Send the full set of new variants,
One new image beat your control,
Your control is predicted to win,
Or there's no clear winner yet.
For the two outcomes where a new image wins, Surge automatically builds a ready-to-use dynamic image block for Klaviyo. From there, cohorts whose winning variant beat the control use the new creative, and every other cohort keeps seeing your original control image.
This is the clearest look yet at where Hero Lab is headed: from only generating options to doing that and recommending the one worth sending.
Built for Multi-Brand Teams
If you're running email across several brands or client accounts, you already know the burden of constantly re-confirming which account you're in. Backstroke's navigation just got rebuilt to lighten that load.
Navigation now collapses to icons and gives your workspace more room. When you expand it, it floats over your content instead of shoving the page sideways. Every account carries a persistent color identity across the top of the app and account switching now lives in one clear control, instead of separate menus.
Built to Be Trusted
Here are a few less flashy, but still important updates:
Cohort imports are more dependable. Cohort membership now keys directly off a person’s identity in Klaviyo, so a subscriber in the right segment can't miss being part of a cohort. Imports are atomic, so a failed import can't wipe out existing cohort data and every import now reports matched, unmatched and duplicate records.
Your email editor protects in-progress work. If the editor loses connection, Backstroke now restores from the latest version you were actually working on, rather than risking a stale reload.
A security upgrade, backed by a bigger commitment. We hardened image processing and upgraded core infrastructure, with no impact on customer accounts. It's part of a larger security effort, alongside Backstroke's SOC 2 Type II certification.
Backstroke Works Where You Already Work
MCP (Model Context Protocol) is the emerging standard that lets AI assistants like Claude connect securely to outside tools and pull in real data. Backstroke's MCP server can now be added directly to Claude, thanks to new Dynamic Client Recognition support. That means you can bring Backstroke's predictions—trained across 10k+ of ecommerce brands—into the AI assistant you're already using.
We're also testing something we're calling Smart Shortlist, which allows Hero Lab to automatically screen generated variants before scoring them. Smart Shortlist catches obvious misses like text over faces or unreadable copy before they ever reach you. So instead of reviewing a large list of options, you’ll see images that are actually worth your time. We’re sitll building, so more on that soon.
All in all, Backstroke is getting smarter across every instance where AI should carry the weight. And it’s getting simpler and more supportive everywhere email marketers still need to make the call.
Want to know what to send before you have to actually send it?
