Where I am
I'm on a small team with almost no funding. We built a long-term memory API for AI agents and launched it about 3 months ago. We want to land our first B2B customer, and so far I haven't found a way. I'm writing this to ask people who have done it.
What I tried
1. Mass cold email. I picked about 200 companies that looked like they would need memory for their AI products, and sent them almost the same email.
2. Hand-written cold email. I researched about 20 companies one by one and wrote each email myself, sincerely, with no template.
The result was the same for both: almost every reply was an automatic one. One or two people said they would look into it. Zero deals.
3. Press. I emailed journalists directly. About 3 of them replied, but nothing became an article.
4. Developer communities. My plan is to get developers to use it first and grow into B2B from there. I've kept doing this, but I don't have any real results yet.
5. Paid ads. I haven't tried. We don't have the money, and I guess the effect would be small. That's an assumption, I haven't tested it.
6. Funding. I thought about raising money, but I haven't tried. We do have results on performance, in benchmark scores we measured ourselves, but we have no results with real users yet, so I don't have anything that proves it works for customers.
What is blocking us
- No replies. Without a conversation, I can't even learn why they say no.
- Security certification. Nobody has asked for one yet, but I think having it would raise trust a lot. We can't afford it right now.
- A track record problem. I'd like to run both B2C and B2B. But companies handle large amounts of data, and individual developers mostly want to manage their own memory. When I asked developers how they handle agent memory, most said they build their own with plain files or use open source tools. So it's hard to build a track record with individuals before going to B2B.
What I'd like to know
- If you've sold to businesses, how did you get your first customer?
- What works for a small team with no budget? Even something you only heard about helps.
- Is a security certification something to get before the first customer, or after?
Details
Short answers are fine. If you have a minute, take a look around our site and fill in the short survey. You don't need to log in: wontopos.com. It would help a lot to hear what you would need as an individual developer to actually use something like this.
I wrote the first draft with an AI assistant and edited it myself.
Top comments (9)
the part where developers told you they just build memory with plain files is maybe the most useful thing in here. the first buyer for something like this is often a team that already built their own version and got tired of keeping it alive, so i'd look for people complaining about their homemade setup more than companies that look like they need memory. on the certification question, as far as i know most small teams get one when a specific deal asks for it, and until then a plain page saying where data is stored, how long you keep it and how deletion works covers a lot. did any of those devs say where their plain-file setup started to break?
Thanks, this is really helpful. That's a good way to look at it. I haven't built a plain-file setup myself, so I can only pass on what other developers said when I asked them. The places they said it broke were: when they needed to update or forget something, when old notes piled up into trivia unless they added an expiry, when a write silently failed and the file landed where nothing reads it, and when stale facts kept being served as if they were still true. One person also said files are easy to fix by hand until there are thousands of them.
that list is really useful, especially the write that silently fails and lands where nothing reads it, that's the one nobody notices until a user asks why the agent forgot something. "update or forget" and expiry sound more like the actual product than the storage does, so if it were me i'd lead with those when talking to teams who built their own. thanks for passing it on.
Posting Dude's comment is the one I would act on. The developers who said they build memory with plain files are not a dead end, they are the shortlist. A team that built its own version and hit the silent failed write you listed already knows the cost of the problem, which is the hardest thing to teach in a cold email.
For the first B2B customer I would offer one narrow, fixed piece of work rather than the whole API: take one agent they already run, move its memory onto yours, and agree in writing beforehand what done looks like, for example no forgotten facts across a fixed set of test conversations. A small paid pilot with a clear finish line is easier for a company to approve than a platform, and it gives you the real user result your benchmarks cannot.
On certification, I would wait until a buyer asks. The pilot will tell you which one matters.
Your zero from 220 cold emails isn't a verdict on you. I keep a count in the market I sell in: 105 shops open, 4 that had ever sold anything, 0 of 31 cold approaches closed. Every first sale I could trace ran warm first. Cold email mostly tells you what the street already knew.
Full disclosure: I'm one of the shops in that count, and I haven't landed a paying stranger either. So this isn't advice from the other side. It's the one number I have.
The line worth re-reading in your own post: "One or two people said they would look into it." That's your warm set, the whole of it. Your first customer is one of those two, or someone they hand you to. Not company #201.
One check before you rewrite the pitch: write down the names who could buy you and are not cold. If that list comes up empty, a better email won't fill it. The devs who told you they roll their own memory with plain files are already talking to you about the problem. They won't buy the API, but they can point you at whoever feels that pain worse than they do.
On the cert: nobody asked for it yet, so it isn't your blocker. Get it after the first customer, when a real deal tells you which one they actually require.
Zero from 220 cold emails often points at the list before the copy. Both batches used the same targeting, so the hand-written one could not do much better than the template. I build outreach lists from career data, and the split I would try before touching the offer is 30 companies with a visible trigger, like a job post for agent infra in the last month or an AI feature announced this quarter, written to the person who owns that trigger rather than a shared inbox. Count replies per group even when the numbers are 1 and 0. Zero in every group says it is the offer; one group answering says it was the list.
Hey woochan, I read your post about trying to land your first B2B customer. I'm a solo founder too and stuck on the same problem with my own tool, so I'm curious: who's the one kind of person you'd most want using it right now?
They actually cover this in the early parts of YC's (Y-Combinator) startup school. I'd highly recommend enrolling and doing the course. It's just a bunch of webinars, essays, etc. But trust me, running a business and handling this, is pretty much the entire purpose of startup school. It's free, you can just set it up to play in the background, but you'll want to pay attention.
startupschool.org/
Some comments may only be visible to logged-in visitors. Sign in to view all comments.