Relationship tools remember every email and miss the only conversation that mattered.
Relationship software reads your inbox beautifully. The conversation that started the relationship never reached it.
Ninety seconds at the booth
Tuesday, 3:40 in the afternoon, an exhibit hall that smells like carpet glue and drip coffee. A program manager stops at your booth, reads your badge, and talks fast. Their team is stuck on a vendor contract that renews in March. They hate the current tool's export. They'll be in your city next month and would take a call if you send something short.
Ninety seconds. Then they're gone.
By the following Monday, your relationship software has done its job perfectly. It knows the name, because you connected on LinkedIn. It knows the company. It knows nothing about the March renewal, the export complaint, or the offer of a call, because none of that was ever typed, sent, or scheduled. The most useful ninety seconds of the week left no trace for it to find.
The inbox horizon
Call it the inbox horizon. Automated relationship tools see everything that crosses into a digital channel: the email thread, the calendar invite, the LinkedIn connection, the Slack DM. Past that line, they're blind. And for a new contact, the relationship almost always starts on the far side of it.
These tools are very good at what they do. Point one at five years of Gmail and it'll surface the last thing a client said about pricing, the date of your last call, the promise you made in a reply and forgot. That's real value for people already in your orbit.
But the conversation that decided whether someone enters your orbit at all happened standing up, with no keyboard nearby. Nothing crossed the horizon until you sent the first email. If you sent it.
Why better recall doesn't fix the conference pile
The category pitches itself on recall, and recall assumes the memory already exists somewhere. For an existing client, it does. It's sitting in a thread. The software's job is retrieval, and retrieval is a solved problem.
A contact you met yesterday has no thread. The job there is capture, and capture has a deadline measured in days. The recall framing treats every relationship as a search problem, when the new ones are a writing problem: somebody has to get the context out of your head and into a record before it fades. No amount of indexing does that for you.
So the tools look strongest exactly where you needed them least. With the people you already know well enough to email.
What is relationship memory, and which half gets captured?
The category has started defining its own terms, which helps. Intriq built a concept page around relationship memory as a distinct category, and its definition has four parts: how you met, what they care about, what was last said, and what was promised Intriq defines relationship memory as keeping what you know about a person, what was said, what you promised, what matters next, recalled before you meet ([source](https://www.intriq.app/what-is-relationship-memory/)). Supermemory published a context-memory guide in March 2026 The guide 'Supermemory: Adding Long-Term Memory to AI Apps' is published at https://betterstack.com/community/guides/ai/memory-with-supermemory/ (Nov 3, 2025) ([source](https://betterstack.com/community/guides/ai/memory-with-supermemory/)), part of a wave of writing that treats memory itself as the product.
Run that four-part definition against what passive aggregation can actually see. The last two parts live in email, so a tool that reads your inbox gets them. The first two mostly live in a conversation. How you met is a room, an event, a moment. What they care about is usually something said out loud, often something nobody would put in writing to a vendor they met ten minutes ago.
The decay problem makes the gap worse. Memory research going back to Hermann Ebbinghaus in 1885 describes how quickly unrehearsed detail fades, and a conference week is the worst case: dozens of similar conversations, no rehearsal, a flight home. We covered that curve in [The 72-Hour Memory Cliff](/articles/72-hour-memory-cliff/). The short version is that the half the software can't see is also the half that disappears first.
Part of relationship memory
Where it usually lives
Passive aggregation sees it?
Captured in the room?
How you met
The event, the booth, the introduction
Rarely (a calendar entry at best)
Yes
What they care about
Said out loud, seldom written
No, not until they write it down
Yes
What was last said
Email, messages, call notes
Yes
Yes, for the first conversation
What was promised
Email replies, meeting follow-ups
Yes, once it's typed
Yes, if noted on the spot
Four parts per Intriq's relationship-memory definition; the channel mapping is our analysis.
Forty conversations, one Monday
Take a three-day trade show where you have forty real conversations. By Thursday night, maybe half of those people have scanned your badge or connected on LinkedIn. Your relationship tool dutifully creates records for them: name, title, company, a headshot.
Monday you sit down to follow up. Every record looks the same. The tool can't tell you which of the forty mentioned a budget cycle, which one asked for a case study, which one was polite and was never going to buy. You're triaging by title and gut feel, and that's how the warm leads end up getting the same template as the tire-kickers.
The other half never produced a digital trace at all. No scan, no connection. As far as the software knows, those conversations didn't happen.
What saves the week is whatever you wrote down at the time.
Three lines typed between conversations beat a perfect index of an empty inbox.
How to judge a contact tool if the horizon is real
If the inbox horizon holds, the evaluation question changes. Stop asking how much a tool remembers. Ask how fast it lets you record what just happened, while the other person is still within earshot and your next conversation is already walking up.
That puts the weight on seconds. A capture flow that needs you to open an app, find the right screen, and type a paragraph will lose to a crumpled card in your pocket every time. Whatever you use has to take the name, the where, and one line of context in about the time a handshake takes.
It also changes what good follow-up looks like. A draft built from what they care about reads like you listened. A draft built from a LinkedIn title reads like you scanned a badge.
A two-minute test for any contact tool
Add someone you met today who has never emailed you. Then ask the tool what they care about. If the answer stays blank until you type it yourself, the tool is a retrieval layer, and you still need a capture habit.
Where Met sits
Met was built for the far side of the horizon. Event Mode keeps the room captured while you keep talking, so the context goes in during the event instead of in a hotel room three days later. Scanning a card gets you the name and title. The note you add in the moment is the part that matters, which is why we've argued that [OCR is a feature, not the product](/articles/ocr-is-a-feature-not-the-product/).
Follow-up drafts come from that captured context, so the email mentions the March renewal and the export complaint by name.
Storage is iCloud-only. What people told you at a booth was said in person, and it stays in your own account.
None of this replaces an inbox-reading tool for the relationships you already have. Met covers the ones you're starting.
What we're watching
The obvious next move for aggregation tools is to reach across the horizon: a quick-capture button, a voice note, a badge scan that prompts for one line of context. If that happens, the category gets better and this argument gets narrower.
We'll be watching whether those features ship as a real capture flow or as a text box bolted onto an index. The difference will show up on the first Monday after a conference.
Capture the conversation while you're still in the room. Get Met before your next event.
Relationship software is excellent at the people already in your inbox. Met is built for the ones who aren't there yet: Event Mode captures the name, the where, and the line of context while you're still in the conversation, follow-up drafts are written from what the person actually said, and everything stays in your own iCloud account. No ads, no interruptions, no contacts on someone else's server.
