i’ve been running campaigns for a few weeks now, and i’m noticing something weird. i’m using AI to personalize based on job title, company size, and recent news mentions, but the replies are still trickling in at like 2-3%. i read a ton of posts here about how personalization is supposed to boost conversion, but i’m wondering if i’m personalizing the wrong things.
here’s what i’m doing: i pull 50 prospects, let the AI generate unique angles for each one, and send them all out. the messages definitely mention their company or a recent hire, so technically they’re “personalized.” but when i compare my reply rate to some of the wins people share here, i feel like i’m missing something fundamental.
i think the real issue might be that i’m personalizing surface-level details (their company name, their role) when i should be testing the core hook itself. like, maybe the angle i’m using just doesn’t resonate with this particular segment, no matter how many personalized details i sprinkle in.
so here’s my question: when you’re A/B testing message variants with AI-generated personalization, are you testing the hook and angle first, or are you assuming that light personalization is enough to carry the whole thing?
this is exactly the problem i see all the time. you’re mistaking personalization for positioning. swapping in someone’s company name doesn’t mean your hook is actually landing. the hook—the very first sentence—is what makes them want to read the rest. without a compelling hook, you could personalize every pixel of that message and still get ignored.
here’s what i’d test: keep everything else the same, but rotate three completely different hooks. one that leads with curiosity, one that leads with pain, one that leads with specificity (a stat or trend). don’t worry about personalizing those hooks yet. just see which one gets the highest open/reply rate. once you know the hook that works, THEN you personalize the details around it.
right now you’re putting the cart before the horse.
honestly, i’d also look at your offer. a lot of people genericize the offer too. they personalize the opening, then ask for a call, which is the most boring close ever. what if instead of “quick call to discuss,” you offered something specific—like a two-minute audit of their current process, or a case study from someone in their space? something that has actual value and shows you’ve done your homework. that’s where personalization really shines.
here’s a thought: are you actually tracking which personalization variables correlate with replies? like, if you’re using an API to pull company data, job tenure, LinkedIn engagement metrics, etc., you could log all of that into a sheet and then correlate it with reply outcomes.
i built a quick Zapier workflow that captures every message i send (prospect name, company, customization angle, message variant) and automatically logs replies back to a Pipedrive deal. now when i review my data, i can see that, say, messages to recently promoted people have a 4.2% reply rate, but messages to people in their role for 3+ years only get 1.8%. that tells me where to focus my targeting filter next.
without that data layer, you’re flying blind on personalization.
the AI itself isn’t the bottleneck—it’s your feedback loop. if you’re not measuring what’s actually working, you’ll just keep sending variations of something that doesn’t convert. build a tracking system first, then optimize.
i deal with a similar challenge in recruiting. high-level candidates get dozens of messages a week, so surface-level personalization (“hey john, i saw you work at acme”) doesn’t cut it. what actually works is showing that you understand their specific career stage and what might genuinely interest them.
for example, if i’m reaching out to an engineering manager, i don’t just mention the company—i reference a specific technical decision they might be wrestling with, or a trend in their industry. “i noticed most engineering orgs your size are moving to microservices; have you started that journey yet?” now it feels like you get their world.
so maybe test: surface personalizations (company name, title) vs. deeper personalizations (industry trend, recent business move, skill gap in their market). my guess is the deeper stuff will win.
also, be honest with yourself about your ICP. if you’re sending to a broad group (like “all marketing managers at B2B SaaS companies”), your personalization will always feel scattered. narrow it down to a specific type of person you actually have a strong message for. that focus will make personalization feel authentic, not forced.
one thing i’d check: are your personalized messages actually using natural language, or do they sound like an AI tried really hard to sound human? linkedin’s filters pick up on that. if every message has the same structure, the same transition words, or the same length, it flags as potentially automated—and that tanks deliverability regardless of the personalization.
vary sentence length, skip a transition word here or there, use informal language (“hey” instead of “hello”). make it look like a human actually sat down and typed it. AI tools can do this, but sometimes you have to dial the settings yourself.
if you’re converting at 2-3%, it might be a deliverability issue hiding under the surface.
also, what’s your account warm-up looking like? if you’re brand new or jumped into aggressive outreach, linkedin might be throttling your reach. make sure you’re respecting daily limits and spacing out your sends. personalization doesn’t matter if half your messages never reach the inbox.
dude, i had the exact same realization. i was personalizing like crazy and still getting 2%. here’s what changed it for me: i stopped treating personalization as a gimmick and started using it to actually target better people.
like, instead of sending to “all managers at tech companies,” i filtered down to “managers at tech companies who’ve posted about hiring or growth in the last 30 days.” then i personalized my angle to that specific intent signal. my reply rate jumped to almost 6%.
so maybe your personalization is solid, but you’re sending to the wrong 50 people to begin with. what does your targeting look like before you even hit send?
also, how long are you running each test? if you’re only sending 10 messages per variant, the data is basically noise. i aim for at least 30-50 per variant before i call it. more data = more confidence in what’s actually working.
great question. here’s how most people misunderstand AI personalization: they think the AI is the magic, but the AI is only as good as your input prompt. if you’re telling the AI to “personalize this message,” it’ll do surface-level swaps. but if you give it specific instructions—like “highlight how their recent expansion into EMEA is relevant to our use case”—it can generate a much stronger angle.
try this: test two AI prompt approaches. one generic (“make it personal”), one specific ("reference their recent [business event] and connect it to [specific pain]). i bet the specific prompt version kills it.
also, don’t sleep on the variant testing piece. A/B test isn’t just about the personalization—it’s about the premise. does your message lead with a problem they have, a benefit you offer, or social proof? that framework choice is worth more than extra personalization details.
one more thing: personalization fatigue is real. if every single sentence of your message is customized, it can feel overwhelming or even suspicious to the reader. the sweet spot is usually one or two strong personalized details that move your value prop forward, plus the rest in a conversational tone. don’t overdo it.
also, consider your messaging hierarchy. first priority: relevance to their current state. second priority: personalization that proves you did your research. third: tone and delivery. most people flip that order and wonder why results suck. get the hierarchy right, and the other pieces fall into place.