用 Jev 做 LinkedIn 外联的 GTM Bot
把外联拆成一系列决策:人是否对、信号是否新鲜、消息讲谁,Jev 逐条打分并锁定最佳话术再建活动。
Run LinkedIn outreach with Jev: Jev is TypeSafe's new model, and it doesn't write. You give it a lead or a message plus a narrow question with the answers you'd accept, and it picks one and tells you how sure it is. That's exactly what outreach needs, because outreach is a stack of decisions. Right person? Signal still fresh? Message about them or about us? Which of my three messages will this one person actually answer? You can't make those calls on every lead by hand, so most campaigns skip them and find out from the reply rate. Jev makes every one of them, on every lead and every message, before anything goes live, for cents. So we built GTM Bot on it. It tiers the leads in your @heyreach_io lists, scores every message variant, locks the best one to each person, builds the campaigns in HeyReach, and self-improves based on your campaign performance. Here's the full setup: 1. Two keys and one small file. A TypeSafe key for Jev, a HeyReach key from Settings, Integrations, API. Then a config that says what you sell: one sentence on the product, one on the problem, your target roles, the company type, who to skip. That's all it knows about you, so make it sharp. 2. Run the demo first. Three seconds, under a cent, and you're reading a report of what it does before you touch your own list. 3. Point it at a list. Any HeyReach list (Sales Nav, Clay, CSV) or your 1st-degree network. Give every lead a signal: one line on what they did, and when. A post about your problem beats a like on a random post, and a signal loses half its weight every 21 days, so fresh wins. 4. Write 2 or 3 first messages, not one. It needs options to pick from. 5. Score the leads. Every lead gets asked: right role, senior enough, right company, how close is the signal to the problem you solve, and the disqualifiers (sells what you sell, open to work, on your exclude list). Tier A, B, C or Skip. 6. Score the messages. Every message gets asked: about them or about you, does it use the signal, does it pitch, does it ask for a meeting on first touch, stock phrasing, a merge tag still unfilled. Score and reply chance per variant. 7. Lock the best message to each lead. No more random split putting your best lead on your worst message. Every one of those is a Jev call, and that's where the 2.59 seconds go. 8. Push it into HeyReach. One list and one campaign per tier-and-variant pair. A and B get campaigns, C and Skip don't. Every lead lands with its verdict on it as custom fields, so you can see why it's there. Preview it first, then let it build. 9. The sequence is set for you. Blank connection request, then the message 3 hours after they accept. Blank requests get accepted 27% of the time, ones with a note 22%, and 3 hours is the rule from our 96,051-campaign report. People you're already connected to skip the request and get the message. 10. Launch. In HeyReach, or do what I do: open Claude with the HeyReach MCP and say "start the [GTM] campaigns". 11. Let the inbox teach it. At 30+ conversations it reads the ones you started, scores the lead and the first message behind each one, and learns which ones got a reply on YOUR account. It also prints what moves your replies, habit by habit: using the signal, asking for a meeting too early, going over 60 words. 12. Every Monday: learn from last week, score the new list, build, launch. Each week it knows a bit more about who answers you.
这条是引用着另一条写的:x.com/1223783704578985991/status/210245714749814…