One agency on the call had 60 open tasks and could not say where most of them came from. Another owner opened his list while the rest of us watched. It said 99 plus, and his own guess was that two thirds of those were reminders that had stopped being tasks a long time ago.
Neither of those is a lazy agency. That is what a task list looks like in a small shop after a year or two. It has stopped telling the truth, everybody in the building knows it, and that is why nobody reads it.
Why it rots
Three things, and none of them is a discipline problem.
Several hands touch one request. The CSR opens the task, the producer answers the customer by email, the VA files the carrier document. The work got done and nobody owned the close, because closing it was never anybody’s job.
Tools create tasks of their own. An email bot opens a follow up. A call logging rule opens another. A renewal sequence drops ten reminders on a list that already had eighty. Each one was reasonable the day somebody set it up. Together they bury the list.
And once the list is mostly noise, people stop reading it. That is the real cost. The one task that matters, the cancellation notice or the claim question nobody answered, is in there, under sixty that do not.
The morning brief
The fix one agency landed on is not a cleanup weekend. Cleanup weekends hold for about a month. It is an audit that runs every weekday at 7 AM, before anybody is in.
It reads everything it can see. Open tasks and service tickets in the management system. Email, inbound and outbound, including the VA’s mailbox and the other staff mailboxes it has delegated access to. The team chat channels. Where a call transcript is tied to a task, it reads that too.
The first run is read only. One ops lead put it this way: it doesn’t go in, it doesn’t close tasks, it doesn’t change anything. It reads, it checks, and it reports.
Then it ranks what is left. Two questions: how long has this been open, and could the house be on fire over it. A claim, a cancellation, a non-renewal, or anything with money attached goes to the top no matter how new it is.
The best example from the call was not a task at all. The owner had asked his team a question in the chat channel weeks earlier. It got buried under the day’s traffic and nobody ever answered it. There was no task for it anywhere. The brief found it on the first morning and put it at the top, because an open question from the owner with no reply is exactly the kind of thing a task list cannot see.
| The list today | After a morning audit |
|---|---|
| 60 open, and nobody knows which ones matter. | Ranked by age and risk, with the evidence attached. |
| Closed when somebody remembers. | Closed when the evidence says so and a person says yes. |
| A question in the team chat, buried, never answered. | Surfaced first thing, at the top of the list. |
| Thirty tasks on one person, ten of them call follow ups nobody checked. | The transcript read, nothing needed, ten closed in one pass. |
Close with proof, not with faith
Here is the part that makes it trustworthy.
When the audit thinks something is done, it does not say “this looks done.” It shows the evidence with a timestamp. On this date, the VA sent this document to this person. Here is the carrier’s reply. Then it asks: here is your evidence that it’s completed, do you want me to mark it completed? One yes and it closes.
The person running this was clear that he is not ready to let it close tasks on its own. That is not an apology for the setup. That is the setup.
The agency wide audit
Once the owner’s own list was clean, the same prompt went across everybody, once a day, with one addition: look for patterns.
Patterns are where the bulk cleanup lives. One staffer had 30 open tasks, and 10 of them were call follow ups. The audit read the call transcripts, saw nothing had been asked for on the call, and they closed all ten in one pass. A service ticket had been open for 360 days. The audit found the email that actually fixed the problem, months earlier, and the ticket went away.
What it does not do is mass close the busiest producer’s list. Most of what is on that list is real. So instead of closing, it pulls: the small requests that do not need a producer get flagged and handed to a VA or a CSR, and the producer’s list gets shorter without anything falling through the cracks.
The manager’s job changed shape too. Check-ins used to be “did you get to that?” Now it is “verify this, I think she’s done it,” and Claude goes and checks, because it can see every channel in the building and the manager cannot. One owner said that being able to see their whole 365 suite gave them a level of oversight they had never had. The task list is only where you notice it first.
The setup is not in Claude
The audit is only as good as what it can read. If it can only see your own mailbox, it will tell you your own tasks are done and miss everything the VA finished. The access is granted on the Microsoft 365 admin side, not in Claude, and it takes about five minutes.
- Go to admin.exchange.microsoft.com.
- Open Recipients, then Mailboxes.
- Select the mailbox and open Mailbox delegation.
- Choose Read and manage (Full Access).
- Add the members who should see it. Do the shared boxes too: service@, quotes@, contact@, and any task@ box, not only the people.
- Test it. Ask Claude “can you see [name]@ and service@?” and make sure the answer is yes before you trust the first run.
What breaks
Three things tripped people up, and they are worth knowing before you trust a flag.
- Adding somebody’s mailbox in desktop Outlook with their password (File, then Add account) is not delegation. Claude reads the backend, not your Outlook. If it was set up that way, Claude cannot see it.
- Claude will sometimes say “I can only see your mailbox.” It is wrong. Remind it that it has delegated access and it will go look.
- The AgencyZoom API can cut off the bottom of long HTML emails. Some carrier cancellation and endorsement notices put the customer name at the very bottom, so the audit could not match them to a customer and raised six or eight false escalations. That is noise, not a fire, and once you know the shape of it you can tell the difference in a second.
Where to start
Read only, on your own list, for a week or two. Tweak the prompt when it ranks something wrong or misses a channel. When you trust what it is telling you, have Claude rewrite the prompt for each employee and hand it to them to run on their own machine. That is exactly how it rolled out at the agency on the call. The owner’s own line on the result was that, as far as tasks go, he doesn’t open the list anymore. He reads the brief.
Here is the starter prompt. Keep it read only, and keep the rule that it asks before it closes anything.
Every weekday at 7 AM, run a read-only audit of my open work. Do not close, edit, reassign, or send anything.
Read:
- My open AgencyZoom tasks and service tickets
- My email, inbound and outbound, plus any mailbox I have delegated access to
- The Teams chats and channels I am in
- Call transcripts or notes tied to open tasks, if available
For every open task or ticket:
1. Look for evidence it is already done (a sent email, a reply, a note, a call transcript). Quote the evidence with its date and who it was from.
2. If it is done, list it under "Ready to close" with the evidence. Ask me before closing any of them.
3. If it is not done, rank it by how long it has been open and how bad it is if it slips. Anything involving a claim, a cancellation, a non-renewal, or money goes to the top.
4. Flag anything in email or chat that asked a question and never got an answer, even if there is no task for it.
Give me: Top 5 to handle today, Ready to close (with evidence), Unanswered questions, and Everything else, oldest first.
If you think you can only see my mailbox, check again. I have delegated access to other mailboxes.