Same agent, different list: 238 cold emails to Japanese schools, 13 replies, 2 meetings

Last week I wrote about 1,027 cold emails that got 7 replies. In that post I mentioned a second product that we sell with the same agent. I used it as a comparison. This post is about that second run. The product and the list The second product is Speech Monster. It is a speaking practice app for students. A student talks to an AI for three minutes a day, and the AI trains the student to answer in a fixed order: conclusion, reason, example, conclusion. Schools use it for interview practice and for presentation practice. The customers are schools, prep schools and boards of education in Japan. The emails are in Japanese. This is one big difference from the LeadAce run, and I will come back to it at the end. The agent is the same. LeadAce builds the prospect list, reads each organization's website, writes one email per organization, sends it from our own mailbox, and records every reply. The numbers The numbers below are from August 28. The agent is still running, and I will write up the newer sends separately. All time, July 12 to August 28: Emails and contact forms sent: 238. One per organization Bounced: 1 Replies from a human: 13, which is 5.5% Of those 13: 6 were interested, 4 were a no, 3 were unsubscribes Meetings: 2. One turned into several calls. Neither turned into a contract Last 30 days, July 30 to August 28: Sent: 167, about 5 a day Replies from a human: 11, which is 6.6% Interested: 4 In the same period, the LeadAce run sent more emails and got only a few replies, and all of them were a no. The code was the same. The research step was the same. I was the person pressing the button in both runs. How the list was built The agent had 6 ways to find organizations. This is what each one returned. Prep schools that advertise interview coaching, found by web search. This had the best reply rate. Small schools that do one-on-one interview practice replied most. Career support offices at colleges and technical colleges, found from the public school directories. This was the only source the agent's own scoring rated above the baseline. Schools named in education press releases and case studies. Medium. Exhibitors at an education technology expo. Too few sends to judge. Boards of education, from their public sites. 20 sent, 0 replies. Member lists of private school associations. 29 sent, 0 replies. The agent archived this one on August 20. I see one pattern here. The sources that worked were the ones where the email could go to a department that has this problem, like a career center. The sources that failed were lists of institutions with only a general address. What the emails asked for None of these emails asked for a meeting. I was surprised when I read them again. The closing line asks for one sentence back. It also tells the reader what to write if the answer is no. Something like: if this is not your department, just reply "wrong department". If this is not the time, just reply "maybe later". So replying takes one line, and saying no takes two words. This is why the no replies are useful. "Wrong department" tells the agent to look for another contact at the same school. "Maybe later" puts the school back in the queue for a later cycle. If the reader says nothing, the agent learns nothing. Each email also starts with one specific thing from that organization's website. For example, a course the school runs, a purchase notice a prefecture published in May, or a seminar the principal gave at an expo. The agent finds these in the research step. The LeadAce run used the same step. What the replies said I am not naming the schools. Here is what the replies looked like. Two meetings. One was with a private school that does one-on-one interview coaching. We met several times, and they told us in detail how they run interview practice. It did not become a contract. The other was one introduction call with the career center of a junior college. It stopped there. One technical college career office printed our materials and put them where students can see them, and shared the PDF with the homeroom teachers of two grades. Two asked for materials before deciding. One of them said the fit with their support program looked small, but asked for a test account anyway. One said they use a different app already. One said that paid apps are hard to recommend to students. This is a signal about price, and the agent recorded it as a rejection reason. One board of education replied with exactly the two words we suggested: wrong department. Three unsubscribes. Two of them came from shared mailboxes, an admissions PR address and a career office mailing list. When you send to a shared address, the reply comes from whoever reads it that day. What was different from the LeadAce run The reader could use the product. A school can start using an app. Nothing in their setup blocks it. In the LeadAce run, the reader needed a Mac and a Claude Code plan, and my list never checked for that. The ask was one line, and the email said what to write if the answer was no. The LeadAce emails asked people to try a product and gave them no link. That is a much bigger ask, and there was no easy way to do it. The sender was our company, from our company domain, with a website at the other end. The LeadAce run sent from a separate domain that had no website. The market. Japanese institutions answer email, even when the answer is no. I cannot separate this from the other three with this small amount of data. I want to be clear that this difference is real. Points 1 to 3 are things I control. Point 4 is not. What I learned is that 1 to 3 were all wrong in the LeadAce run at the same time, and the agent could not fix any of them by writing better. What did not work Boards of education: 0 for 20. Now the agent only targets the officer who runs a specific program, and skips the general inbox. Association member lists: 0 for 29. Archived. Shared mailboxes: two of the three unsubscribes. The list needs a department or a person, not an info@ address. Price: one reply said a paid app is hard to put in front of students. One reply is not a trend. But it is the kind of no that a sales tool should record, and ours did. What I take from this This is not a great result. Two meetings and no contract do not make a business. But the two runs used the same agent, and one of them produced conversations while the other produced nothing. The writing was the same in both. What was different was who was on the list, what we asked for, and whether the reader could act on it. If you are testing an outbound tool on your own product and getting nothing, I would check those three things before you change the copy.