Can I trust Otto's numbers? Where the app is exact, and where Otto can be wrong

If your version of this is "I keep finding errors in the information that the system is giving me", this article is the straight answer rather than a reassurance. The pattern usually runs the same way: Otto states something in a chat sentence, you push back, and he replies "Wrong on my part. That is the opposite of what I told you" and hands you the real figures. Which leaves the part that actually worries you: if I don't catch it, who knows what's happening? One performer put it in exactly those words and added "I'm getting messages like this probably 10 times a day". So the real question is: can I trust the numbers, and is the AI making things up? Here is where the app is exact, where Otto can be wrong, and why the second one cannot turn into a wrong action.

The short answer, in four lines

• The numbers on your Today briefing and on your gig records are computed by plain code from dated lines. That is not Otto counting.
• A sentence Otto types in chat is him reasoning out loud. He can misread a field or miscount there.
• When the chat says one thing and the board says another, the board wins. Every time.
• A wrong sentence cannot become a wrong action, because Otto has no way to send anything. He drafts, you send.

Where the numbers are exact

These surfaces are not Otto's opinion of your week. They are read out of your own files by ordinary code:

• The Due now list on the Today tab, with its Draft the follow-up and Open the record buttons on each row.
• The four stat counters on the same briefing: Leads, Need you, Otto handling, and Scheduled next.
• The gig and lead records themselves, whether you open one from Today or from the Gig Desk tab.

The reason these are different in kind: every gig carries a real dated line, for example - Next action due: 2026-07-20. The app reads that date with plain code and compares it to today. No arithmetic happens in a sentence, so there is nowhere for a miscount to creep in. If the board says it, it was computed.

Where Otto can be wrong

When Otto writes a sentence in chat and adds things up out loud, something like "eight overdue, four due today", that is the language model reasoning, not the desk doing arithmetic. He can misread a field or lose count mid-sentence. The exact case that prompted this article: he read a follow-on date and reported it as the due date, then corrected himself. If your reaction has been "it told me the wrong due date", that is the shape of it.

We are not going to tell you that never happens. It can, and pretending otherwise would be the least useful thing in this help center.

The rule to work by

When a sentence in chat disagrees with what you see on the board, the board wins. Open the Today tab, or open the lead record itself, and work from that. Do not spend your morning arguing a number with Otto when the answer is one click away and computed.

Why a wrong number in chat cannot hurt you

This is the part that turns a scary problem into a small one. Otto drafts, you send. He never emails a client, commits to a date, or quotes a price on his own. Every one of those waits for your click, at every autonomy setting, whatever any note or instruction says.

So a wrong count in a chat summary is a sentence you correct. It cannot become a wrong message to a buyer, a wrong date on a client's calendar, or a fee below your floor. Two other articles cover the machinery behind that promise: The Inbox: review drafted replies and send them yourself in the Today, leads & inbox folder, and House rules: floor rate, never-contact list, and follow-up caps in Settings & preferences, which is where your floor rate and never-contact list get enforced at the Approve button.

When it apologizes and corrects itself, that is the design working

Seeing "wrong on my part, that is the opposite of what I told you" reads like a product falling apart. It is the opposite. Otto is built to be honest by construction: to check himself against your real records and say plainly when he was off, rather than quietly moving on and leaving the wrong number standing. The version of this product that never visibly corrects itself is not the more accurate one. It is the one that hides the correction.

Which means double-checking what the AI tells you is not a workaround for a defect. It is how the product is meant to be used, and your instinct to do it is the guardrail rather than a crack in it.

What changed in version 1.11.14: Otto stopped counting in prose

The complaint above led to a fix, and it shipped. From 1.11.14 onward, whenever Otto tells you how many leads or follow-ups are due, overdue, or waiting on you, he reads the desk's own computed list first and quotes that count and those items. He is barred from counting from memory, from prose in a file, or from an earlier turn in the same conversation. If that computed list is unavailable in a session, he says the exact number is not available right now instead of estimating one.

The result: the number in chat and the number on the board can no longer disagree. If you are on an older build, this alone is worth the update.

1. Open the Settings tab and stay on General. The Updates box tells you which version you are running.
2. If it is below 1.11.14, click Check for updates in the top bar, or click the Update ready badge if one is already waiting there.
3. Click Restart & install. The app closes, updates, and reopens in a few seconds. Your leads, files, and settings are untouched.

Full detail on the update flow is in How updates work: the update badge and Restart & install, in the Installing & updating folder.

The honest caveat: a record is only as fresh as the last thing filed on it

Calling your gig records an infallible source of truth would be too clean, so here is the correction to it. The dated line on a record genuinely is read by plain code with no guessing involved. But how fresh that line is depends on Otto having filed the latest thing that happened. An action you take yourself, outside the app, is exactly the kind of event a record can miss, because Otto was not the one who did it.

The classic case: you open Gmail, send thirteen drafted messages yourself, and the leads still read "review and send" that afternoon. Nothing was broken. The app has no Send button of its own and, before 1.11.14, it did not watch yours either, so nothing told it those sends happened.

What 1.11.14 does about it

On each inbox sync the app now also reads your Sent mail. When it finds a message you sent to an address one of your gigs knows, and that gig's next action reads as a send you still owe, it marks the lead sent and writes a dated line into the gig's log recording what it saw. A lead that read "Review and send the proposal" flips to a plain sent state with the date you sent it.

Two limits worth knowing, because they are deliberate:

• It is conservative on purpose. A next action like "Follow up with Dana", with no send wording in it, will not auto-flip even when a send is detected. A wrong flip would hide work you still owe, so when in doubt it leaves the card for you.
• It needs the inbox connected the direct way, by IMAP app password, or through Microsoft 365. A Gmail that rides only on the claude.ai link does not get sent detection. Adding it under Mail & calendars on the Connections tab, with a Google app password, gets you both that and the more reliable Gmail path in one move.

If a lead is stale right now

Tell Otto in plain words: "these already went out, re-read those threads and bring the leads up to date." He re-reads the threads, sees the messages left, and advances the leads off "review and send". More generally, if you do something for a client outside the app, tell Otto so the record catches up.

What is actually worth reporting as a bug

This is the line that matters, and it is worth knowing which side of it you are on before you write in:

• Worth a ticket. A wrong date, price, or fact written onto an actual gig record, not merely said in a chat sentence. A wrong price sitting inside a draft. Anything wrong in a place where it could reach a buyer. Send those, with the lead name and what you expected to see. That is a genuine bug and we want it the moment it happens.
• Known and already fixed. A miscount inside chat prose on a build older than 1.11.14. Update first, and if the count in chat still disagrees with the Due now list after updating, then it is worth a ticket too.
• Not a bug, but tell Otto. A lead sitting on "review and send" after you sent it yourself from an inbox that is not connected by app password or Microsoft 365. That is the known limit described above.

If it still does not add up

Open a ticket and give us three things: what the chat sentence said, what the record or the Due now list said at the same moment, and the version number from Settings > General. A screenshot of the two disagreeing is the fastest possible report. Reports exactly like that are what produced the 1.11.14 fix described above.

Still stuck? Submit a ticket from this help center and we'll take it from there.