Put the deck's churn claim, the data room's cohort number, and the call transcript's quote in one table

What's already on hand

The material is all there. The claims sit in the pitch deck, the numbers that back them up or knock them down sit in data-room files, and the remarks that cut against the numbers sit somewhere in the middle of a call transcript. All that's left is going back and forth between the three, finding the same item in each, and lining them up.

• 40 pitch-deck slides — 6 metrics slides, 3 on market size, 2 on the competitive landscape. The rest is team and vision.
• 130 data-room documents — monthly revenue sheets, a cohort retention spreadsheet, 14 supply contracts with major customers, 9 sets of board minutes, a draft reps-and-warranties schedule.
• 6 reference-call transcripts — 3 current customers, 2 churned customers, 1 former employee. 40 to 60 minutes each; roughly 8,000 words apiece once written out.
• 3 market reports — 2 paid research PDFs, 1 public industry report.
• The investor's own notes — 3 meetings, 2 to 3 pages each.

How it works today

He pages through the deck, pulls out every claim worth checking, and builds a table. To fill the "evidence" column, he opens the data room. He hunts for the cohort sheet among 130 files with names like "v3_final_revised.xlsx," opens it, checks the number, copies it into the table, and writes which file and which tab it came from into the cell next door. Every single claim costs one round trip.

The transcripts don't turn up in search at all. He remembers someone saying something important on a call, but not where inside 6 files × 50 minutes. Hit Ctrl+F for "churn" in the transcripts and nothing comes back, because the founder's word for it isn't the customer's word for it.

Ten of the fourteen days go to copying evidence over next to claims. The judgment gets squeezed into the last three.

How the material goes in

Kick off extraction and go to bed. While the 130 documents come in, the only work left for a human is registering the vocabulary.

How you ask

"put the founder's churn claims next to the data-room numbers"
→ A 3-row table.
   Row 1: deck claim "1.8% monthly churn" — source "Pitch deck · slide 12"
   Row 2: cohort sheet, 2024-09 cohort at 3.1% — source "cohort_retention.xlsx · tab 2"
   Row 3: call quote "basically stopped using it in the back half"
          — source "Reference call 4 · line 412"

"any customer named on a reference call that isn't in the revenue sheets?"
→ Two. Click the note link on a row and the original opens.
   One: mentioned on a call as "contract ended last year", no row in the sheets.
   One: the call and the sheet spell it differently, so it landed as two entities.
        You decide whether to merge them under "Also known as" on the entity page.

"@churn pull together what people said about it"
→ The "In your words" section on the entity page.
   4 verbatim sentences: 2 from the founder, 2 from churned customers.
   Each one carries who said it and the source note and line.

These answers exist because the "Extract sources" step tied 130 documents and 6 transcripts into one knowledge graph (a map that connects concepts as points and lines). A single point — churn — has the deck's claim, the cohort sheet's number, and the call's quote each attached to it by a line. A row in the table is one of those lines, unfolded.

The numbers in the table trace back to the source text. Hover a connection chip on the entity page and the sentence that states the fact is quoted, captioned "{note} · line {n}". Click it and the sheet opens at the spot where 3.1% came from.

What changes

What's left is the judgment. Whether to work from the deck's 1.8% or the cohort's 3.1%, and how much weight the call's contradiction carries, is the investor's call. What this workflow takes off his plate ends at finding the relevant passage in all three sources and gathering them into one table.