Call review — August 2026: why deals were lost and which questions won
Produced by: Run EXAMPLE-QSC-01 · 6 of 8 attached files read · document on SALES-88 · demo values
- Question
- The question you typed, verbatim, so the document can be read a month later without you: "Why did we lose last month, and which questions show up in the deals that advanced?" Leave the question empty and the standing review runs instead.
- Answer
- At most five sentences, and every share in them reads "n of m" — a bare percentage is refused before the document is written. Here: 3 loss reasons, of which 2 were said out loud by a buyer and 1 was inferred. When the transcripts do not support an answer, this section says so and the run is still a success.
- Coverage
- Before any finding, because a reader who learns at the end that files were skipped has already believed the table. 6 of 8 attached files were read across 4 accounts; 2 were skipped, each with the reason — an audio recording Aestus does not transcribe, and a password-protected PDF. 5 of 6 calls matched a deal ticket, so 1 counts toward frequency and nothing else.
- Loss reasons
- One row per reason, each marked said by the buyer or inferred — and the marking is enforced, not requested. "No independent security review to show when the buyer's own review asked for one" carries the buyer's own words: "A controls document is not an independent review. We will revisit next year." (CISO, Northwind Freight). "Implementation effort judged too large for the current quarter" is marked inferred and names the sentence it rests on. A buyer-stated reason with no quote, or an inference that names nothing, is refused rather than softened.
- Questions that appear in the calls that advanced
- 1 pattern survived the threshold of three calls across two accounts: the rep asks who signs off on security before showing the product, in 3 of 6 calls across 2 accounts. It may say the calls that advanced had it only because deal outcomes were read off the board; point it at no deal project and the same section reports how often the question came up and states that it can show no more.
- Objections we did not answer
- 2 objections nobody answered on a call, and 2 tickets filed for them, each linked back to the call library and titled with an [OBJ-…] marker so the next run updates the ticket instead of filing a second one. "We will follow up" does not count as an answer.
- What this does not show
- Never empty, and never optional: 5 lines naming what this evidence cannot reach. First among them — This reads only the files attached to SALES-88. A call nobody wrote down, or wrote down somewhere else, is not in here.
- Memory proposals
- 1 cited candidate claim, quarantined until a person accepts or rejects it. Nothing is written to the workspace's memory on its own, and an agent can never approve its own claim.
- Sources
- 7 rows, one per file or ticket any line above cites, each with the artifact revision and the line range the file reader returned. Every marker in the document resolves to a row here, and a claim citing a source that is not listed is refused before you see it.