Caught in my Nette

I’ve now seen the same stat get quoted by three different marketing blogs this month: AI-driven personalization drives “82% higher conversion rates.”

Here’s the thing… The number isn’t really about AI writing better emails. It’s about AI catching a behavioral signal sooner than a human would. Those are two completely different jobs, and the growth teams I talk to are pointing their AI effort at the wrong one.

Retention as part of Growth

Growth marketing has an acquisition-cost problem right now. Customer acquisition costs are up roughly 222% over the past eight years (Antavo), which is exactly why “just get more people to sign up” stopped being a sustainable growth strategy. This has led to a pivot to lifecycle and retention, which is a right growth instinct.

But look at where the actual lift is coming from. Propel’s research ties a 20-30% churn reduction to lifecycle segmentation aka sorting customers by behavior, not writing them nicer copy. Bain’s older but still-solid number still holds: a 5% bump in retention can drive 25-95% more profit. Scaling upwards of 50% for SaSS businesses. None of that is about email quality. It’s about timing and targeting.

Meanwhile, the eMarketer data on AI in email marketing this year shows AI is now used in 51% of email content creation. Used mostly for copy and image generation (25% of use), personalization (18%), and analysis (16%). This has collapsed production time where only 6% of marketers needed two weeks or more to build a campaign in 2025, down from 62% the year before.

So, we basically got a lot faster at writing emails. What we didn’t get is meaningfully better results from writing them. CTR benchmarks haven’t moved much, and the report’s own conclusion is that “oversight remains essential” because AI copy still needs a human check before it goes out. Faster isn’t the same as better.

71% of consumers say they want personalization, and 76% say they get frustrated when it’s done badly. People aren’t asking for more emails to be written, they’re asking for the right one at the right moment. AI that writes faster gives you more shots at getting it wrong just as easily as more shots at getting it right.

And this isn’t only an email problem. Nearly 43% of Gen Z now say they trust AI agents for shopping decisions, and shoppers referred by ChatGPT convert at roughly 12%, versus 7% from a plain Google search. This signal comes with higher intent but noticeably lower patience for friction. If that’s where B2C discovery is heading, the brands winning retention won’t be the ones with the cleverest subject lines. They’ll be the ones whose systems notice a customer’s intent shift in real time and get out of the way fast.

Getting Noticed

The “4 Forces” framework I ran into recently (CMSWire, using AMRA & Elma’s attention research): customers judge relevance in about 1.7 seconds on mobile now. That’s not enough time for a beautifully-written email to save a badly-timed send.

Which means the most impactful job for AI in your retention stack isn’t “help me write email #11 in this drip sequence.” It’s “tell me the moment this specific customer’s behavior changed, so I can act on it before they’ve already checked out mentally.”

That’s a completely different build. And it’s the one most teams skip, because writing copy feels like the productive, visible work, while signal detection feels like plumbing.

Next Step: Behavioral Trigger Audit

1. Pull your current lifecycle sends and sort them into two piles. Calendar-triggered (day 3, day 7, day 30 after signup) versus behavior-triggered (visited plan change page twice, app untouched for 2 weeks, featurre usage dropped week-over-week). Most lists I’ve seen are still 70%+ calendar-triggered. That’s the tell.
2. For every calendar-triggered send, ask what behavior it’s actually a proxy for. Your “day 7 check-in” email is really trying to catch “active downgrader”, so why wait for day 7 if the behavioral signal (checked “change plan page 3 times) already told you?
3. Use AI for the part it’s actually good at: scoring and pattern-matching across your usage data, not drafting the twelfth version of your re-engagement email. Feed it the behavioral data, have it flag propensity-to-churn signals in near real time, and only then decide what message fires.
4. Keep the actual copy human-written, or at least reviewed. This isn’t a purity thing. When content reads as obviously AI-made, people rate it as “less natural and less useful” and engage with it less, even when the underlying offer is identical. Source: NIM research. You can use AI everywhere in the pipeline except the one place where “sounding like a person” is the whole point.
5. Measure engagement quality, not send volume. The attention research above found a 10% increase in attention correlates with 17% spending growth.

Here’s what that might look like. Say you run a subscription meal-kit app (classic B2C, classic churn problem). Your current flow probably has a “day 14 check-in” email because someone decided two weeks ago that felt like a reasonable cadence. A behavior-triggered version will skip that guesswork completely. Based on the above audit, it might fire the moment someone is inactive on the app beyond a week, because that’s the actual churn signal, whether it happens on day 9 or day 30.

The AI’s job is finding out that skip pattern in real time. The email that goes out once it fires is still written (or at least reviewed) by a person who knows what tone actually gets a skipped order back.

At the very least, go count how many of your lifecycle emails are triggered by a calendar versus a behavior. If you run this audit and the results surprise you either way, I’d genuinely like to know.

Note to self: Underpinning this audit is data availability and accuracy. Guess, I shall write about that soon and how your martech stack impacts the ability for your organisation to be AI-ready.

---

Want to be alerted once a new post is out? Stay “In the Nette”.