Your follow-up is a cold email with a name pasted in.
Contact data and relationship memory are two different problems. The category built almost entirely for the first one, which is why fast, correct follow-ups still land cold.
Thursday morning, eighty clean records, nothing to say
Eighty cards came home from the conference in a jacket pocket.
You scanned every one of them Sunday night. Names spelled right, titles correct, companies deduped, all of it tagged with the event. By any measure the capture worked.
Now it's Thursday. Open a compose window and try writing to the fourth one down.
VP of Partnerships. Mid-market logistics platform. And then nothing. You know what was printed on the card. You don't know why the two of you talked for twenty minutes, or which one of you brought up the reorg, or what you said you'd send over.
So you write the safe version. "Great meeting you at the conference, would love to continue the conversation." You send it inside the week, from a real record, to a correctly spelled human being.
And it lands as a cold email. Because it is one.
Contact data and relationship memory are two different disciplines
Contact data is the printed layer: name, title, company, email, phone. It's stable, structured, and machine-readable, which is exactly why the whole category learned to capture it well. A card scanner can do it. A LinkedIn export can do it. Your phone's contacts app has done it for fifteen years.
Relationship memory is the other layer. How you met. Who introduced you. What they said they were worried about. The fact they're moving teams in Q3. That their daughter just started college. Intriq, one of the tools now indexing against this framing, calls it "soft, useful context," which is the sharpest language currently in the market for it.
Call the distance between the two the soft-context gap.
The gap isn't a data-quality problem. Your records are fine. It's that the field a follow-up actually runs on was never captured by anything, and the tools that promised to solve "contact capture" quietly scoped themselves to the half that fits on a card.
"Capture faster" was the wrong thing to optimize
Look at how the category competes and the assumption becomes visible. Scanner apps benchmark against OCR accuracy, multi-card batch speed, seconds from photo to saved record, number of languages recognized. Personal CRMs compete on integrations and reminder cadence. All of it is throughput on the printed layer.
That made sense when the bottleneck was typing. It isn't anymore. Recognition on a standard business card has been effectively solved for years, and shaving another second off it changes nothing about Thursday.
We've argued a version of this before, that [OCR is a feature, not the product](/articles/ocr-is-a-feature-not-the-product/). The soft-context gap is why. When a category optimizes the part that's already working, it produces tools that are excellent at the wrong job and users who can't explain why their pipeline still feels thin. Speed of send was never the variable. Nobody ever ignored a follow-up because it arrived on Thursday instead of Tuesday. They ignored it because it had nothing in it.
What does relationship memory actually mean?
The concrete test is which questions a record can answer when you sit down to write. Contact data answers one of them. Relationship memory answers the rest.
The evidence that this is a real, unoccupied category isn't one vendor's positioning. It's convergence. Intriq is indexing a dedicated Relationship Memory page. PeopleRecall and Whodat? are both live in the App Store building against recall rather than storage. Supermemory published a Context Memory Guide in March 2026 aimed at exactly this layer. And practitioners have been documenting homegrown systems for the same problem, which is the clearest signal of all. People build their own scaffolding when the tooling has skipped a job they still have to do.
Four independent parties circling the same word in the same year is not a trend. It's a gap being discovered from four directions at once.
The question you'll ask on Thursday
Contact data answers it
Relationship memory answers it
Who is this and where do they work?
Yes
Yes
Why did we end up talking at all?
No
Yes
What did they say they were dealing with?
No
Yes
What did I promise to send them?
No
Yes
Is anything about to change for them (role, funding, move)?
No
Yes
Who introduced us, and what do I owe that person?
No
Yes
Field-level comparison of what a scanned card yields versus what writing a follow-up actually requires.
The same contact, written two ways
Take the VP of Partnerships. Both emails go out Thursday at 9:40. Both are grammatically clean, both use her real title, both reference the event.
Version one: "Hi Dana, great connecting at the conference last week. I'd love to find time to continue the conversation and share more about what we're building."
Version two: "Hi Dana, you mentioned your integrations team is getting absorbed into platform in Q3 and that you're trying to lock partner commitments before the reorg lands. You asked me to send the two-page version. It's attached. Happy to be a Q3 signature rather than a Q4 one if that's useful."
The difference between them isn't writing skill and it isn't send speed. Every specific in version two came from a twenty-minute hallway conversation, and none of it was recoverable four days later from a scanned card. Version one is what's left when the soft context evaporates. It's not a worse email. It's the only email that record could produce.
A follow-up without soft context is a cold email with a name pasted in, no matter how fast you sent it.
Relationship memory has to be captured at a different moment
If the thesis holds, the fix isn't a better follow-up workflow. It's a capture habit that fires somewhere the category currently isn't standing.
Contact data survives the trip home. It's printed, so it waits patiently in your pocket until Sunday night. Soft context doesn't wait. It starts degrading the moment you turn to the next person, and by the time you're at the desk with the stack, most of it is gone. That's the mechanic behind [the 72-hour memory cliff](/articles/72-hour-memory-cliff/), and it's why batch-processing cards after an event is structurally incapable of producing warm follow-ups. You can't batch a memory you no longer have.
So the capture moment moves. Not the hotel room at 11pm. The thirty seconds after you step away from the conversation, still standing on the floor, while the reorg detail and the promise you just made are still in working memory. Three lines is enough. Why we talked, what they're dealing with, what I owe them.
That's the whole discipline. It's unglamorous, it's a behavior change, and it's the part no scanner can do for you.
The capture window is the conversation, not the inbox
If your capture step happens after the event, you are only ever capturing the printed layer. Soft context has to be written down while you're still in the room, or it isn't captured at all. Any workflow that defers everything to a post-event batch has already lost the field that makes the follow-up land.
What a tool built for the second problem looks like
A tool designed around relationship memory makes different tradeoffs than one designed around contact throughput. It has to be fast enough to use mid-floor without ending your conversation, which means capture and context go in together in one pass, not in two sessions separated by four days. It has to treat the note as the primary record and the card as the attachment, not the reverse. And it has to keep the whole thing somewhere you're willing to put a comment about someone's job search, which is a storage question, not a feature question.
That's the shape Met is built to. Event Mode exists so the room stays captured while you keep talking, and everything lives in your iCloud rather than on a vendor's server, because the notes that make follow-ups warm are also the notes you'd never want mined.
We come down on the strong version of this claim: the category's next real advance isn't recognition accuracy, it's a capture habit that gets soft context recorded in the thirty seconds that matter. Four independent tools have started naming the gap in the last year. Watch which ones actually move capture into the conversation, and which just add a notes field to a scanner and call it memory.
Capture the context, not just the card. Get Met.
Met was built for the person who leaves a conference with eighty cards and four days of decaying context. Capture the person and the reason you talked in a single pass on the floor, keep it in your own iCloud instead of on a vendor's server, and write Thursday's follow-up from what was actually said rather than what was printed.
