# How We Got Supermemory a 25% Outbound Reply Rate | PumpGTM.com

Canonical URL: https://pumpgtm.com/blog/supermemory-25-percent-outbound-reply-rate

[← All guides](/blog) Outbound Cold messaging

# How we got Supermemory a 25% outbound reply rate

Supermemory's LinkedIn outbound gets a 25% reply rate. The whole method is two decisions: targeting (thousands of Y Combinator companies narrowed to 523 startups building AI agents) and messaging (a two-message sequence that leads with research instead of asking for a meeting). Full walkthrough with the video and transcript.

Namanyay Goel · November 5, 2026 · 13 min read On this page
- [Step 1: targeting, or who is actually worth a message](#step-1-targeting-or-who-is-actually-worth-a-message)
- [Step 2: messaging, or why the ChatGPT version gets you blocked](#step-2-messaging-or-why-the-chatgpt-version-gets-you-blocked)
- [The Supermemory sequence, message by message](#the-supermemory-sequence-message-by-message)
- [What to copy for your own outbound](#what-to-copy-for-your-own-outbound)
- [Questions founders ask](#questions-founders-ask)

Supermemory gets a 25% reply rate on cold LinkedIn outbound because of two decisions: who to message and what the first message says. We narrowed thousands of Y Combinator companies down to 523 startups building AI agents, then sent each founder a two-message sequence that leads with Supermemory's own research instead of asking for a meeting. The video above walks through both in six minutes, using the real map and the real sequence inside [PumpGTM](/blog/what-is-pumpgtm-how-it-works). This is the written version.

[Supermemory](/customers/supermemory) is the memory layer for AI agents, founded by Dhravya Shah. We run their GTM outbound, and this is the same system we run for 50 plus startups in San Francisco. If you are building in developer tools and have your first five or ten customers, this is the part where you stop finding the next ones by hand.

## Step 1: targeting, or who is actually worth a message

Targeting means finding the correct people to talk to and not wasting time on people who do not need your product. If you are a car mechanic, you want people who own a car, not people who commute by bike. Obvious when you say it, and still the thing most outbound gets wrong, because a bigger list feels like more pipeline.

For Supermemory the right people are fast-growing startups building with AI. The starting point was every Y Combinator company on one map, the same data we publish for free at [pumpgtm.com/yc/founders](/yc/founders): industry, launch year, team size, founder names, LinkedIn profiles and X handles for every company.

Then two cuts:

- Remove the very large companies. Supermemory wants earlier stage startups that are still building new products. That took the list roughly in half, to about 3,000 companies.

- Keep only the companies actually building AI agents. That removed more than four fifths of what was left. 523 companies remained.

523 is a list you can reason about. PumpGTM reads each company and explains why it looks like a fit. One of them, Humwork, offers human experts as an API for AI agents. Another, OpenTag, is an early stage team building an AI coworker inside Slack, which means they are managing their customers' memory in one place and will hit exactly the problem Supermemory solves. That reason, written beside the name, is what a good target list looks like. The same cut works on any audience you already own: [the people engaging with your LinkedIn posts](/blog/turn-social-engagement-into-qualified-prospects), or [your competitors' followers](/blog/how-to-find-customers-from-your-competitors).

## Step 2: messaging, or why the ChatGPT version gets you blocked

Before the message that works, the one that does not. If you ask ChatGPT for LinkedIn cold messages for a memory infrastructure startup, you get this:

Hi [first name], came across your work at [company]. I'm building in the memory infrastructure space and connecting with people working on AI systems. Would be great to connect.

Nobody talks like this. "Came across your work at company" wastes the first line. "Would be great to connect" asks for nothing and offers nothing. The follow-up, "I am exploring how teams are approaching long-term memory for AI systems", is a pile of keywords about the sender, not about the person reading it or the problem they have. It makes nobody curious and it reads like a bot, so it gets ignored or blocked. We wrote more on this in [three rules for a cold message that gets replies](/blog/cold-message-that-gets-replies-three-rules) and [how to review AI-written cold messages](/blog/review-ai-written-cold-messages).

## The Supermemory sequence, message by message

This is the sequence running in PumpGTM for the 523 founders, as a [LinkedIn and email play](/plays/linkedin-email). PumpGTM sends the connection request, waits for the acceptance, sends the messages from Dhravya's own account and stops the moment someone replies.

- Connection request. As soon as a founder joins the sequence, they get a connection request from Dhravya. No pitch in the note.

- Wait one day after acceptance, then message one. The first message does not waste a line. It goes straight to Supermemory's strongest argument, file systems versus vector databases, and links a research article the Supermemory team wrote on it. The ask is not a meeting. It is "would love to hear your note on this". The point is to position Supermemory as the expert in memory, so the message leads with research. This one message alone gets a lot of the replies.

- Wait three days, then message two. For the founders who were busy, the second message says Supermemory is deploying a new model and onboarding some early customers, and explains plainly why it is better: the cheapest, fastest and best way to add continual learning and context to an agent.

That is the whole sequence. Two messages, one real reason to answer each, nothing that sounds generated. It has gotten Supermemory a 25% reply rate, with a lot of the replies turning into signups or booked calls.

## What to copy for your own outbound

Cut the list until every company has a one-line reason beside it. Lead the first message with the thing you know that your buyer wants to know, and ask for their opinion on it rather than for their calendar. Give the busy ones one more reason three days later. Then let the system do the sending and the waiting, from your own account, so the reply lands with you.

If you want this set up for your company, [start the 7 day trial](https://pumpgtm.com/?utm_source=blog&utm_medium=cta&utm_campaign=supermemory-dhravya#install) and point it at your market, or [book a call](https://pumpgtm.com/call-founder?utm_source=blog&utm_medium=cta&utm_campaign=supermemory-dhravya) and I will map your targets and write the first messages with you. The YC founder map in the video is free at [pumpgtm.com/yc/founders](/yc/founders).

## Transcript

Today I'm going to show you how to get your next 100 customers by mapping your target market, starting your outbound campaigns and writing great messaging that actually convert. And I'm going to do this with a walkthrough to one of our best customers Supermemory. We're helping their GTM team multiply their productivity and get more meetings booked on the calendar. This video is going to be especially valuable if you're building in DevTools and if you've got your first five or 10 customers and you want to figure out how to start scaling your GTM and outbound motion. And if you haven't heard of me, I'm Namanyay. I'm building PumpGTM.com. We're a Y Combinator backed startup, and we're helping 50 plus of the top startups in San Francisco scale and run their entire GTM outbound motion. First, targeting, how to find the right buyers for your product. Two, messaging, or how to write great messages that convert and get sales and do not sound like AI slop. And I'm going to start with targeting. What targeting means is to find the correct people to talk to and not waste your time talking to people who don't even need your product. Just as an example, if you're a car mechanic, you're focused on getting people who actually have a car. If you're talking to people who commute or use bikes, you're just wasting your time. And that's why good targeting is important. And for Supermemory, this means that they want to work with fast-growing startups who are building with AI. And to explain how we've done the correct targeting for them, I've got this great visualization loaded up, which is thousands of Y Combinator backed companies in one map. We can scroll in, we can see by industry, I can switch it to look at it by launch here. I'm seeing some of familiar names here, Instacart, PagerDuty, DoorDash, and we can keep going and keep going into this great 3D visualization of thousands of Y Combinator company. But of course, even though Supermemory is a great product, talking to all of these companies is not feasible and they might not have a need. So first, we're going to eliminate some of the very large companies because right now Supermemory wants to focus on earlier stage startups who are building new products. That leaves us with approximately 3000 companies. We've essentially reduced the list by half. The next we can take is to focus on only those who are actually building AI agents. Again this eliminates over four fifths of the companies. We are left with 523. And with PumpGTM, we can figure out what exactly they're building in. So just as an example, we've got Humwork towards the right who are offering human experts as AI agents. And we've got PumpGTM explaining to us why this looks like a good fit. We've got their founder names, their LinkedIn and their X handles right here, easily accessible. And this 3D map is really useful because I can actually look at the entire landscape in one easy go. I can click into different company names, I can learn about their founders, I can get their exact team size. All of it is data that I've made available for free on pumpgtm.com slash yc slash founders. You can check it out and use it for yourself without any conditions. A great company that I've spotted on this list is OpenTag. They're building an AI co-worker inside Slack. They're an early stage company. They're probably managing their entire customer's memory in one place for their co-workers. And Supermemory is likely a product that can help them scale or deal with larger and larger amounts of memories. Now when we selected the right kind of company to talk to, the second question becomes about our messaging. And this essentially means writing the perfect cold message that gets their interest and helps you get a sale. Now before I tell you what to do, I'm going to show you exactly what not to do and explain why it doesn't work. If you're using ChatGPT to generate your cold messages, I'm sorry, but you're not going to get the reply rates you need because all of them sound exactly like the same generic AI slop. I'll show it to you right now. I've opened up ChatGPT and I'm going to ask it. Hey, I'm building a memory infrastructure startup. I found a target list of my ICP. What LinkedIn cold messages should I send them to get some meetings booked on my calendar? Now, if you read this yourself, you're going to see why these messages will not work. Hey, first name came up across your work at company. I'm building in the memory infrastructure space and connecting with people working on AI systems. Would be great to connect. All right, came across your work at company. You're just wasting the first messages and the text that you have. Would be great to connect. No one talks like this in real life. Please do not use this in your messages. You will simply get blocked. Again, I am exploring how teams are approaching long-term memory for AI systems. All of this is this unnecessary amount of keywords. It's unnecessarily about me versus them or the problem that they're trying to solve. and none of it makes anyone curious or interested at all. Instead, it sort of bores them and makes you think you're an AI bot. Instead, I'm now opening up PumpGTM and here's the messages we write for Dhravya and Supermemory. If you don't know about us, PumpGTM basically helps you automate the entire part of connecting and sending them a sequence of messages. In Supermemory's case, as soon as the founder joins this sequence, we send them a connection request. After it's accepted, we wait for one day. In the first message itself, it's important not to waste any time and directly show your best strengths. So in this message, we're straight up talking about file systems versus vector databases and linking a great research article that the Supermemory team came up with. Rather than asking for a sale or 15 minute meeting, we're just asking to share a note. Because we want to position Supermemory as the expert in memory. That's why we're leading with research. Usually even this single message gets a lot of replies. But just in case that the founder is busy, we're sending them another message after waiting for three days. Here, we explain that we're going to deploy a new model at Supermemory and they're onboarding some early customers. And we're clearly explaining why it's better. It's the cheapest, fastest, and the best way to add continual learning and context to the agent. This messaging sequence has been extremely effective. It has gotten us a 25% reply rate with a lot of them signing up for the product or ending up booking calls. Now all of these little pieces, figuring out the correct targeting, writing great messages and setting up your entire GTM outbound engine is what I'm doing for a few more startups. Comment pump below and I'll choose some people to help out and I'll help you map your time, set up some messages and try and get you some results. I'm also going to be talking a lot more about GTM topics. If there's anything you want to hear more about, just drop a comment. We're already trusted by some great teams. You've got some logos up here. If you want your GTM setup as well, go to pumpgtm.com and check us out. There's a seven day free trial. Thank you.

## Questions founders ask

How did you get from thousands of YC companies to 523? Two filters on the YC company map. First, remove the large companies, because Supermemory wants earlier stage startups still building new products; that halved the list to about 3,000. Second, keep only the companies building AI agents, which removed more than four fifths of the rest and left 523. PumpGTM then writes a one-line reason beside each company explaining the fit.

What is a good reply rate for cold LinkedIn outbound? Most founder-sent cold outbound we see lands well under 10% when the list is broad and the message is generic. Supermemory's sequence gets 25%, and the difference is the list (523 companies with a stated reason each) and a first message that leads with research the founder wants to read.

Why does the first message ask for a note instead of a meeting? A founder who has never heard of you will not give you fifteen minutes, but will tell you what they think of a research article on file systems versus vector databases. Asking for their opinion positions Supermemory as the expert in memory and starts a real conversation. The meeting comes out of the reply.

Why don't ChatGPT cold messages work? They all sound the same: "came across your work at company", "would be great to connect", a sentence of keywords about the sender. Nothing in them is about the reader or a problem the reader has, so they read as a bot and get ignored or blocked.

How long is the sequence? A connection request, one message a day after it is accepted, and a second message three days later for the founders who were busy. Two messages, sent from the founder's own LinkedIn account, and the sequence stops for anyone who replies.

Can I use the YC founder map myself? Yes. It is free at pumpgtm.com/yc/founders with no conditions: every Y Combinator company with industry, launch year, team size, founder names, LinkedIn profiles and X handles.

[Original source](https://x.com/NamanyayG/status/2118407497983643803) · 2026-11-05 · Adapted and expanded for PumpGTM

[Namanyay Goel](https://nmn.gl/) · Founder, PumpGTM

Founder of PumpGTM (YC). Writes about founder-led GTM, LinkedIn outbound, and the systems behind it.

[X](https://x.com/NamanyayG)[LinkedIn](https://www.linkedin.com/in/namanyayg/)

On this page

- [Step 1: targeting, or who is actually worth a message](#step-1-targeting-or-who-is-actually-worth-a-message)
- [Step 2: messaging, or why the ChatGPT version gets you blocked](#step-2-messaging-or-why-the-chatgpt-version-gets-you-blocked)
- [The Supermemory sequence, message by message](#the-supermemory-sequence-message-by-message)
- [What to copy for your own outbound](#what-to-copy-for-your-own-outbound)
- [Questions founders ask](#questions-founders-ask)

Share

Founders doing this with PumpGTM

[Supermemory Supermemory put founder-led LinkedIn outreach on autopilot](/customers/supermemory)[Docmost Docmost follows up with trial signups and abandoned checkouts on LinkedIn](/customers/docmost)[Browzer Browzer reaches DevRel leaders with outreach that reads like Rahul wrote it](/customers/browzer)

## Find and engage with buyers across channels

Give your AI the super strength to find you leads and message them till you get the sale across LinkedIn, email, X, and more.

## More guides

[### How we'll add $250K ARR to a $100M company in 30 days A $100M company has 20,000 X and LinkedIn followers and 300 of them are high-value signups who never became customers. Here is the 30 day founder-led outbound plan we set up with PumpGTM to turn that into $250K ARR, plus the webhook that messages every new signup the moment they sign up. 6 min read](/blog/add-250k-arr-to-a-100m-company-in-30-days)