Field notes
Conversations more than doubled
This was our strongest window so far. From a smaller pool and fewer invitations, Shannon opened more conversations and scheduled more meetings than in any window we have reported, and we know where it came from.
The window
Between 11th and 18th November, across all accounts, 13,790 prospects were found and Shannon admitted 2,266. Shannon Desktop · 18 Nov 2026
Eight days, starting the day after last week's window ended, so nothing is counted twice.
The numbers
- Found: 13,790 prospects.
- Admitted by Shannon: 2,266, an admission rate of 16.4%.
- Opening messages: 1,873 placed, and operators approved 1,828.
- Invitations: 1,709 sent, and 398 accepted.
- Follow-up messages: 445 sent to people who had accepted.
- Replies: 79 prospects replied, and 101 conversations are under way.
- Meetings: 11 scheduled. None completed yet.
Shannon Desktop · 18 Nov 2026
What we read
- Conversations more than doubled. Last window 66 prospects replied and 41 conversations were open. This window 79 replied and 101 are open. The window before that, 6 were open.
- Meetings scheduled rose again. 1, then 3, and now 11. The first completed meeting is the next number to land.
- It came with fewer invitations. 1,709 were sent, against 2,660 last window, and 398 were accepted, against 476. Follow-ups still rose, from 255 to 445. That is last week's fix at work: follow-ups no longer dry up part-way through a window.
Shannon Desktop · 10 Nov 2026 Shannon Desktop · 18 Nov 2026 Shannon Desktop · 2 Nov 2026
An evaluation week
Shannon learns from what happens next, so this was an evaluation week: we measured as much as we sent. We read three things.
The live numbers above. They are the observed result, and the first that can show version 0.7, which was deployed the day before the window began and has since been accepted. Shannon Desktop · 18 Nov 2026
Where it came from. It was Shannon's own changes, and not luck. The agent that reads each prospect was rebuilt, it works from better context, and our collection controls and enrichment are stronger. Those changes are why this window looks different from the last, and the store shows them. FBED Labs · 18 Nov 2026
- The read of each prospect was finished far more often. Shannon's read was complete for 96.4% of the prospects admitted this window, up from 85.8% the window before. Fewer were left waiting.
- The replies were better. 52 people sent a reply that our reply reader classed as positive, against 22 last window. Counting replies alone would have hidden that.
- The agents did more work. Our agents ran 2,147 prospect reads this window against 1,017, 2,362 message-writing calls against 1,024, and 699 reply-collection runs against 39. Each call carried about a quarter more context, and the agents recorded no errors in either window.
- The dashboard and the store agree. The Desktop store holds 2,260 admitted prospects for the window, within a handful of the dashboard's figure.
Shannon Desktop store · 18 Nov 2026 Spans store · 18 Nov 2026
The admission rate moved. It was 12.0%, then 10.3%, then 11.8%: three windows under 12%. It is now 16.4%. Less was found, and the number admitted fell by less than the number found. That is what tighter collection and better enrichment should do: a smaller pool, and more of it good enough to admit. Shannon is admitting more of a better-collected pool. Shannon Desktop · 26 Oct 2026 Shannon Desktop · 2 Nov 2026 Shannon Desktop · 10 Nov 2026 Shannon Desktop · 18 Nov 2026
The tests we run before anything goes live. We test on a fixed set of leads, with the scorers blind to which message is which, and we write the pass mark down before the run. The first test used 17 frozen leads. On them, blind judges scored the new opening-message agent's drafts 2.68 for would-reply, against 1.65 for the messages we actually sent, and 0.18 for harm, against 1.35. Lower is better for harm. The judges caught 12 of 12 planted messages with an invented fact, so the scoring holds up. FBED gym · 5 Nov 2026
The same discipline works in both directions. A second version of the prospect-reading agent was tested on the same 17 leads. Its insight and usefulness scores fell on every one of them, so it was rejected and the first version stays frozen. We only ship what passes. FBED gym · 5 Nov 2026
What changed for Shannon this week
- Version 0.7 is accepted, along with its first two patch releases, in our release registry. Their changes are in the window above.
- Version 1.0 is in development. The opening message is next: more of what goes into a first message will be varied and measured, one part at a time. We will say more when there is something to read.
Shannon Desktop · 18 Nov 2026 Spans store · 18 Nov 2026 FBED Labs · 18 Nov 2026
What we'll report next Monday
The same table for the next seven days, and the first completed meetings as they land.