21 August 2026
An action item dies for structural reasons, not because people are lazy. It has no owner's name attached to it, no date, or it lives in a document that closes the moment the call ends and never gets opened again. Fix those three things and most of the follow-through problem disappears on its own.
"We should really get a dashboard for this" and "I'll build the dashboard by Friday" can appear four sentences apart and get written down the same way: a bullet point starting with a verb. One of them is a task. The other is a mood. The person taking notes, whether human or a model listening to the call, has maybe two seconds to decide which is which before the conversation moves on, and most note-taking systems don't ask them to decide at all — they just capture the sentence.
The tell isn't the verb, it's the grammar around it. A real commitment has a subject and a time reference: someone said "I" or their name got assigned to it, and there's a "by" or a "before" attached. A wish has neither. It floats. If you go back through a transcript and pull out every sentence that starts with "we should" or "someone needs to," you'll find that almost none of them survive contact with the following week, and almost none of them were meant to — they were the sound a group makes while thinking out loud.
Human note-takers under-report this distinction for a specific social reason: writing "Dave agreed to do X by Thursday" when Dave actually said "yeah, probably, I'll see" feels like putting words in Dave's mouth in front of the whole team. So the note gets softened to "Dave: X" with no date, which reads as a task but carries none of the information a task needs. The softening is polite in the room and useless the following Monday.
This is also why action items clump at the end of meeting notes under a heading like "Next steps," divorced from the sentence that created them. Once a commitment is separated from its own context — who said it, what tone they said it in, whether anyone else in the room nodded — there's no way to tell, three weeks later, whether it was a promise or a suggestion. The note preserved the words and lost the only part that mattered.
A transcription model doesn't soften language out of politeness — it has no stake in the room. But it has a different failure mode: it's very good at finding verb phrases that look like tasks and mediocre at knowing which speaker actually owns one, especially in a meeting where three people talk over each other about who's doing the follow-up. "I can take that" said by the wrong voice, misattributed by a beat of overlapping audio, produces a wrong owner with total confidence.
This is the part worth checking rather than trusting. AVAY's meeting AI writes down an action item only when it can attach a name and, ideally, a date to the sentence that created it — if neither is said out loud, it flags the line as open rather than inventing an owner to make the note look tidier. That's a deliberately unglamorous choice: a follow-up with a gap in it is honest about where the conversation actually left things, and a follow-up with a guessed name in it is a liability nobody notices until the deadline passes.
An action item that makes it to completion almost always has three things attached to it, and the absence of any one of them is enough to kill it. The name has to be a person, not a team — "marketing will look at this" has no owner, because everyone on marketing can assume it's someone else's. The date has to be a date, not a season — "soon" and "next sprint" both mean whatever the reader wants them to mean when they reread the note. And the item has to be visible somewhere the owner will actually open before the deadline, not just archived in a notes doc that gets written once and read never.
That third condition is the one most tools skip, because it's the least interesting to build. A perfectly worded action item in a document nobody reopens is functionally identical to no action item at all. The fix isn't a better sentence, it's a better location: resurfacing the item at the start of the next meeting where it's relevant, or in a channel the owner already checks, rather than trusting them to remember which doc it's in.
AVAY keeps a running list of decisions and open items as the call happens, and it can pull an item back into the next related meeting if a connector links the two. But it can't invent an owner or a date that nobody said, and it won't chase someone by email — the follow-up still depends on someone in the room actually saying who and by when, out loud, on the call. What changes is that the gap becomes visible immediately, on screen, while the conversation that created it is still happening and someone can just fill it in.
| Sounds like | Treat it as | |
|---|---|---|
| "I'll send that by Thursday" | Owner and date both stated | A real task — log it as-is |
| "We should look into this" | No owner, no date, no time reference | A discussion topic, not a task |
| "Someone should own this" | Passive construction, no named person | Ask who, right then, before moving on |
| "Yeah I can probably take that" | Hedged commitment, tentative tone | Confirm the date before the call ends |
Listen for a name and a time reference attached to the sentence, not the verb. "We should fix the onboarding flow" is a suggestion; "I'll fix the onboarding flow by Wednesday" is a commitment, even though both start the same way. If either piece is missing, ask for it out loud before the topic moves on — it takes five seconds in the meeting and saves a dead task later.
No, and treating every one as a task is part of why follow-up lists get ignored — most of them are the sound of a group thinking, not a decision. Convert a "we should" into a task only when someone accepts it by name and gives a date; otherwise leave it as a discussion note, because a list padded with untracked aspirations trains people to stop reading it.
It can attach an owner when the person says their own name or clearly accepts the task out loud, but it will misattribute or leave a gap when ownership is implied rather than stated, especially with overlapping speech. A tool that flags the gap rather than guessing a name is more useful than one that always fills it in, because a wrong guess looks correct until the deadline arrives.
Because the third requirement — a location the owner actually reopens — gets skipped. A dated, owned item sitting in a notes document that's written once and never revisited behaves exactly like an item with no owner or date at all. It has to reappear somewhere the person already looks, like the agenda of the next relevant meeting, or it competes with everything else in their inbox and loses.
A named person, a specific date, and a place it will be seen again before that date arrives. Drop any one of the three and the item is functionally decorative — it exists in the notes but has no mechanism forcing anyone to act on it.
An action item doesn't die from laziness — it dies from a missing name, a missing date, or a home nobody reopens; fix those three structural gaps and follow-through stops being a discipline problem.
AVAY is a video meeting platform that transcribes the call itself — no bot joins, because there is nothing to join. Start one at avay.ai, read how each part works in the documentation, or see what it costs.