Caught in my Nette

AI-driven personalization drives “82% higher conversion rates.”

Well, the number isn’t really about AI writing better or more emails. It’s personalisation. i.e. more about AI catching a behavioral signal sooner than a human would. Those are two completely different jobs.

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.

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 SaaS businesses. That lift traces back to timing and targeting with relevant content.

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 got a lot faster at writing emails, while results stayed roughly flat. CTR benchmarks have held steady, and the report’s own conclusion is that “oversight remains essential” because AI copy still needs a human check before it goes out.

Back to personalisation – 71% of consumers say they want personalization, and 76% say they get frustrated when it’s done badly. What people actually want is the right email at the right moment,not a higher volume of them. AI that writes faster gives you more shots at getting it wrong… just cos the recipient gets fustrated at the overcommunication (cue: nagging parents)

This pattern shows up beyond email too. 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 will be the ones whose systems notice a customer’s intent shift in real time and get out of the way fast.

Getting Noticed

Based on the “4 Forces” framework (CMSWire, using AMRA & Elma’s attention research), customers judge relevance in about 1.7 seconds on mobile now. That leaves almost no room for a beautifully-written email to rescue a badly-timed send.

Which means the most impactful job for AI in your retention stack is telling you the moment a specific customer’s behavior changes, so you can act on it before they’ve already checked out mentally.

That’s a different build than most retention stacks are set up for today. After all, writing copy feels like the visible, productive work, while signal detection feels more 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 sign up) versus behavior-triggered (visited plan change page twice, app untouched for 2 weeks, feature usage dropped week-over-week). Most lists I’ve seen are still 70%+ calendar-triggered.
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 scoring and pattern-matching across your usage data, more than for 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. There’s a concrete reason for this, not just a stylistic preference: 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). AI earns its keep everywhere else in the pipeline, right up to the point 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 replaces that guesswork with a real signal. Based on the above audit, it might fire the moment someone delay the next meal delivery, because that’s the actual churn signal, whether it happens on day 9 or day 30.

The AI’s job is spotting 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.

If you run this audit, I’d genuinely like to know what you find.

---

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