Stensul makes email learn from its own blocks. Subject-line tests are too small

You know the drill. The team has tested two subject lines against each other for years. Winner: variant B, 0.8 points more opens. Lesson for the next campaign: none. Nobody knows whether it was the word, the length or the day of the week.

That is the gap Stensul is going after.

What Stensul announced

On November 6, Stensul, an email creation platform for enterprise marketing teams, unveiled its November release. The key points, per the press release:

• Self-Improving Email (early access): each email's performance is tied to the specific content blocks inside it. Marketers and AI agents are meant to spot which patterns work better across personalized and localized variants and carry them into future sends. Review and approval stay with people.
• Subject lines and preheaders per segment inside a single campaign, written by hand, suggested by AI, or both.
• Pre-send compliance checks that flag missing required content while the email is being built.
• Salesforce Marketing Cloud Next: the integration is generally available.

What is missing: performance numbers, pricing and a general-availability date. These are announcements, not results.

The unit of learning just got smaller. Everything behind it changes

Here is what I find interesting. The thing a team learns from shrinks from the campaign to the block. Hero image, offer box, testimonial, button copy. Each block gets its own performance history.

Which means: if you want to work this way, you first need a clean block library. Same blocks, same names, reused instead of rebuilt every time. I build my own emails for i4u.business exactly like that, from blocks, by script. The reason is boring: you can only compare what repeats.

For a lot of teams, that is the real wake-up call. AI is not the bottleneck. The bottleneck is the email that gets hand-crafted as a one-off in an agency template every single time.

The catch: the optimizer only knows clicks

A system that keeps reusing whatever got clicked yesterday eventually lands on the email everyone sends. Discount up top, countdown, red button. Good for click rate this quarter. Bad for the brand over three years.

Ehrenberg-Bass research on distinctive brand assets has said it for years: recognition comes from consistency, not from optimizing every single touchpoint. If an agent picks the blocks, someone has to decide which elements are never up for grabs.

Then there is measurement. Since Apple's Mail Privacy Protection, opens are a shaky signal. Score blocks on opens and you are optimizing on noise. In my view, only clicks and conversions are a usable basis, and only once your list is big enough that differences are not random.

What I would do now

• Run a block inventory. Count the blocks that genuinely recur across your last ten emails. Fewer than five? There is nothing for an AI to learn from yet.
• Lock your brand elements. Logo placement, tone, one recognizable block. Fix those before any tool is allowed to optimize.
• Measure on clicks, not opens. And write down what each block is supposed to do.

Bottom line: Self-improving email rewards teams that work modularly and punishes everyone else with an email that looks like everyone else's.