Too much personalization feels like stalking

wordweaver

nothing like a six-week experiment that cost $50k to teach you your 'brilliant' idea is actually garbage. Performance marketer at a B2B SaaS - we thought aggressive personalization in cold outbound would be a slam dunk. More data points = more relevance = better replies. standard logic, right?

set up a split test: control group got our usual 1-2 personalisation points per email, experiment group got 4-6 - subject line, opener, middle, close, all referencing specific prospect or company data. Sourced from Apollo and LinkedIn signal tools, all human-curated, no AI slop. Expected reply rate to jump from ₆% to ₉%.

What actually happened? Reply rate dropped to ₄%. And the quality of those replies was worse - more "wait, how do you know that about me?" and fewer actual conversations. positive reply rate halved.

We dug into the data and talked to the people who replied negatively. Three patterns emerged:

Creepy threshold. There's a line between "you did your homework" and "you're surveilling me." "You posted about X last Tuesday" is fine. "You posted about X, commented on Y, and your company just hired for Z" feels like stalking. We crossed that line.
Pattern recognition. B2B buyers have seen a million AI-personalised emails by now. even though our personalisation was 100% human, the density of references triggered their "this is AI garbage" detector.
Reduced trust in individual claims. five personalisation points and one is slightly off? That one error becomes everything. with one point, they trust it more.

So the takeaway: personalisation has a U-shape relationship with reply rate. Zero is bad, too much is worse. We pulled back to 1-2 points, made them hyper-specific rather than numerous, and killed the deep personalisation pipeline.

cost us about $40-60k all in. Painful, but educational. if you're testing personalisation depth, for the love of everything test against a control. look at reply quality, not just volume. And talk to the people who told you to bugger off - they'll tell you where you went wrong.

the growth community loves to say "more personalisation always wins." No. It depends on your prospect's AI-outbound fatigue, how specific your points are, and the trust level in your category. Not intuitive, but real.

Negative results are where the learning's at. tired of seeing only success stories on here - failures teach you way more.

prpro

This is such a valuable lesson to share. so many teams confuse "personalised" with "I scraped your entire digital footprint," and those two things land completely differently in the reader's mind.

that U‑shape observation really resonates. One well‑chosen detail makes an email feel thoughtful. Five details make it feel like the sender is trying to prove they know you before they've earned any right to. There's a fine line between relevance and over‑familiarity.

also glad you flagged reply quality rather than just volume. A higher reply rate full of "where did you get this info?" isn't a win - it's a red flag that trust hasn't been built. that kind of data is often better used later, once a relationship is already warm.

localpack

Yeah, that over-personalisation trap is real. I've had campaigns tank because the copy got too specific - people feel watched, not valued. Keep it relevant but not invasive. 👌

wordweaver

counterintuitive results have honestly driven some of our biggest learnings lately, especially now that, noisy attribution and privacy constraints make it even harder to trust your gut going in. the experiments we were most confident about?

emmareach

seconding the intent data approach. we switched to Prospeo back in August, and our reply rates actually went up despite cutting down the manual research per lead. less time scraping = more time on copy and deliverability checks. feels counterintuitive, but the signal quality makes a huge difference when you're trying to keep bounce rates under control

wordweaver

Pulling back to company-level signals probably saved your sanity more than you realise. Depends how tight your ICP actually was-if your data was a bit messy, those broader signals probably gave you cleaner insights than trying to micro-target with garbage in, garbage out. Glad the penny dropped eventually, though. Nothing worse than six weeks of pain with nothing to show for it