Hey 👋
I was reviewing a batch of "personalized" outbound drafts this week.
Ten minutes of research per row. Company news. A hiring post. A quote from their last earnings call. The CRM notes looked thorough.
What struck me wasn't the effort. The effort is fine.
It was that almost none of that research made the actual note.
What shipped was five polite paragraphs that could have been sent to anyone in the same industry, plus a first name. Or worse: an AI biography of their career that never named the reason to talk this week.
Last week was signal half-life. This is the next leak: you had a reason, then you buried it under theater.
Here's what that actually looks like.
TL;DR
Company-specific personalization can roughly triple replies from directors+; activity context can ~3× replies and meetings (Gong Labs, 30k+ emails)
87% of buyers say sales emails still miss a relevant org challenge (Gong)
LinkedIn notes: slightly fewer accepts (25.3% vs 27.6% blank) but 8.2% vs 5.3% replies after (~55% lift). AI prompts only moved replies 7.3% vs 7.0% (Belkins, 2026)
Ops A-grade cold email: 5.4% reply vs 3.4% baseline (~58% lift). Name a process, not a "workflow challenge" (Lavender, 231k+ emails)
Signal-triggered outbound still ~2–3× reply vs non-triggered (ORRJO, 2026)
This week's move: one real detail in the opener. Cap the approve queue. Kill theater drafts instead of rewriting them colder.
Weekly finds
Your exec prospect doesn't care that you both like golf. Gong read 30k+ emails. Talk about their company and exec replies roughly triple. Flattery? Not so much.
"Doing great things in [Industry]" fools nobody. ORRJO calls it a mail merge with extra steps. Read this before your next "I noticed..." opener.
Should you add a note to your LinkedIn invite? Belkins tested 11.5M requests. Slightly fewer accepts, way more replies. The AI-written ones barely moved.
Stop telling ops people about their "workflow challenges." Lavender's 231k emails say it plainly: name the actual process and replies jump ~58%.
Your CRM just moved into Claude. Pipedrive's connector is live, and most people use it to read, not write. It still won't find your reason for you.
THE ISSUE
Personalization Theater
You'll like this one if your drafts look "researched" and still get ignored.
Signal-led outbound fixed half the problem: you finally have a reason to write.
Then the note arrives as a book report. Or it arrives looking personalized and saying nothing a stranger couldn't guess from the About page.
That's personalization theater. It burns research time, burns the signal, and teaches the buyer that "personalized" means "longer."
Theater has a costume. Relevance has a reason.
Gong's personalization taxonomy is useful because it separates costumes from reasons.
Individual flattery (hobby, alma mater, "congrats on the promo") can lift IC and manager replies. Executives respond harder to company-level stakes. Activity context can roughly triple replies and meetings because you're referencing something they already did.
Same CRM field labeled "personalized." Completely different goods.
ORRJO's blunt line: "doing great things in [Industry]" is a mail merge with better lighting. Buyers recognise it in one glance.
Practical: Before you approve, ask: would this first sentence still make sense if I swapped in a peer company? If yes, it isn't personalization. It's costume.
The research dump is not the note
Founder-led teams are especially good at this failure mode.
You care. You open six tabs. You paste three facts. Then you polish until it sounds like a LinkedIn About written by a polite robot.
The buyer didn't ask for a biography. They want one clear reason this is worth two minutes now.
Lavender's ops slice: baseline ~3.4%, A-grade that names a concrete process ~5.4%. "Optimize your operational efficiency" is theater. "Onboarding docs for the new self-serve tier" is a reason.
Practical: Cap the research to one usable detail. Write the opener around that detail only. Everything else is notes for you, not copy for them.
AI made the costume cheaper. It did not make the reason better.
Belkins' LinkedIn study is awkward for the "AI will fix messaging" story.
Across millions of touchpoints, AI-assisted prompts moved reply rate from 7.0% to 7.3%. Real at scale. Not transformative.
Gong's buyer line is harsher: 87% still say the email never addressed a relevant challenge. Five paragraphs of "personalization" can still miss the only thing that mattered.
Pipedrive's MCP in Claude is the same plot sideways: CRM gets closer to the assistant (~65% of early use is reads). Faster context. Still a human call whether the note has a reason or a costume.
Practical: Use AI for the brief (what fired, when, one stake). Write or heavily edit the three sentences yourself. If the draft still sounds smooth and generic, it's theater. Kill it.
The approve queue is where theater survives
Approval is healthy. An unbounded queue is where stale research goes to get a second coat of polish.
You leave the draft overnight. The hiring post ages. Your "personalized" opener is now a history paper.
Same rule as last week, different object: the detail has a half-life too. Same-day hot queue. Kill unsent theater instead of rewriting it colder on Friday.
Belkins' connection notes are the soft filter: blank accepts a bit more; notes that name a reason reply a lot more. Volume of "personalized" activity is not the KPI.
Practical: Before approve, glance at when the signal fired and whether the one detail is still true. If the reason is gone, don't rescue the draft with better adjectives. Archive it until a new reason appears.
What we change
No new stack required.
One detail. One. From the signal, not from their bio.
Write short. Three to five sentences. Costume dies in length.
Approve hot. Same-day for hot reasons. Theater drafts don't get a rewrite next week.
Score the opener, not the research notes. If the first line survives a company swap, it isn't ready.
Separate costume metrics from reason metrics. "Personalized sends" without reply quality is a mood.
This week's experiment
Pick twenty drafts in the approve queue.
Tag each opener: reason or theater.
Send only the reason pile (still short, still same-day where the signal is hot). Kill or rewrite the theater pile with one detail and nothing else.
Compare reply rate on reason vs theater for one week. The spreadsheet usually shuts up.
