No call wasted, no thread wasted, no note wasted
Run every call recording, email thread and rep note through one extraction. Each fact is typed and keeps its sentence, so your CRM becomes queryable.
Jared Hawkins-Young · Founder, Lost Star TechnologyOctober 8, 20265 min readPart 2 of 13Key takeaways
- One extraction reads call recordings, email threads and rep notes the same way, so a fact counts whether it was said on a call, written in a thread or noted after a meeting.
- Every signal gets one of eight types (pain, requirement, spend, competitor, economic buyer, champion, paper process, timeline) and keeps the sentence, the speaker and the moment it came from.
- A new signal is drafted for the rep to approve, a confirming one raises nothing, and a contradicting one becomes an argument for the rep, never an overwrite.
- Typed signals make the book queryable: which accounts named a competitor this quarter, which deals have a paper process and no security contact.
To get call notes into your CRM automatically, run every recording, email thread and rep note on a deal through one extraction that pulls out typed facts, attaches the exact sentence to each, and drafts them onto the record for the rep to approve. Not a summary: a pain, a requirement, a competitor, a paper process, each with who said it and when. Do that for every touchpoint and the book stops being a pile of transcripts. It becomes something you can ask questions of.
What does one deal leave behind?
Take the Northwind deal from part 1. Marcus, one of your account executives, has had one recorded call with Dana Reyes, Northwind's VP Sales. There are three emails in the thread and two notes he wrote afterwards. Today the call gets a summary from the recorder, the thread sits in Gmail, and the notes sit wherever Marcus keeps notes. Each source holds something the other two do not.
| Source | What it catches | What it misses | How it is cited |
|---|---|---|---|
| Call recording | The buyer's own words: scope changes, objections, the paper process, who signs | Anything said after the call ended | REC 31:14, with the speaker |
| Email thread | Forwards, new names on cc, written asks such as a SOC 2 report | Tone, and anything decided on a call | EMAIL 3, with the sender and date |
| Rep notes | What the champion said after the recording stopped, or on a call that was not recorded | Exact words. It is the rep's account | NOTE 1, with the author |
The recorder's summary covers the first row, loosely. The other two rows are where Dana's forward of the one-pager to her CFO lives, and the fact that a rep left Northwind in August and took his account history with him. Neither was said on the recording.
How does one extraction read all three the same way?
- 1Extract the claim
- 2Type it
- 3Attach the sentence
- 4Compare with the record
The extraction runs the same four steps on every source. It pulls out each claim. It assigns one type from a fixed list. It attaches the sentence and the speaker: a timestamp for a recording, a message number for a thread, an author for a note. Then it compares the signal with what is already on the record.
That last step is what makes it useful. A new signal is drafted as a field for the rep to approve. A signal that confirms the record raises nothing. A signal that contradicts the record does not overwrite it. When Dana says "We'll start with the four reps, not six" and the opportunity still prices six seats, the result is an argument for the rep, which is what part 3 is about.
What is a typed signal?
A signal is one fact a buyer stated, with a type, a value, a speaker and a source. The type list is short on purpose. Eight types cover what a B2B software deal turns on: pain, requirement, spend, competitor, economic buyer, champion, paper process and timeline. Most of them will look familiar from MEDDPICC, the qualification checklist many sales teams already use: economic buyer, paper process, pain, champion and competition are letters in it, and requirements and timeline sit under its decision criteria and decision process. Its paper process covers procurement, legal review, security questionnaires and contract logistics, which is exactly where the Northwind deal is stuck.
| Type | Signal | Said by | Source |
|---|---|---|---|
| Pain | Reps update HubSpot from memory, on Fridays | Dana Reyes | REC 06:12 |
| Pain | Two deals slipped last quarter before anyone saw them | Dana Reyes | REC 12:40 |
| Pain | A rep left in August and his account history left with him | Marcus, after the call | NOTE 1 |
| Requirement | The rep approves every record before it is written | Dana Reyes | REC 14:05 |
| Requirement | Works inside HubSpot, no new login for the reps | Dana Reyes | REC 15:30 |
| Requirement | Four seats, not six, onboarding included | Dana Reyes | REC 18:02 |
| Requirement | Writes deals and contacts only, never marketing data | Dana Reyes | REC 22:15 |
| Competitor | "It writes a summary, not the fields" | Dana Reyes | REC 24:40 |
| Competitor | Evaluating the recorder's own CRM sync | Dana Reyes | EMAIL 3 |
| Spend | Budget approved for Q4, under $30K | Dana Reyes | REC 27:51 |
| Spend | One annual invoice, not monthly | Dana Reyes | REC 28:30 |
| Economic buyer | VP Sales signs under $50K, CFO above | Dana Reyes | REC 30:22 |
| Economic buyer | Forwarded the one-pager to the CFO "for visibility" | Dana Reyes | EMAIL 1 |
| Paper process | Security review takes six weeks, not started | Dana Reyes | REC 31:14 |
| Paper process | Security review before legal, kickoff on the 14th | Dana Reyes | REC 38:12 |
| Paper process | Wants the SOC 2 report before the kickoff | Dana Reyes | EMAIL 2 |
| Champion | Bringing the security lead in herself | Dana Reyes | REC 38:40 |
| Timeline | Four reps live before the February sales kickoff | Dana Reyes | REC 33:10 |
Fixed types are what make the record queryable. A summary paragraph cannot be counted. A row typed competitor, on an account, with a date and a speaker, can. The outbound series makes the same move before the first meeting: in outbound part 9, a public signal has to earn its way into an email. Here the signals come from the buyer's own mouth, which makes them the strongest ones you will get.
What can you ask a typed book that you cannot ask a pile of transcripts?
Once every deal is typed the same way, the questions you now answer by going round the reps in the pipeline review answer themselves:
- Which accounts named the recorder's own CRM sync as the alternative this quarter, and what did each rep say back?
- Which deals have a paper process on the record and no security contact?
- Where has the economic buyer never been on a call, while the stage says Commit?
- Which deals had scope cut on a call while the amount stayed the same?
None of these needs anyone to reread a recording. They are filters on a table, and every row in the answer links back to its sentence. That table is what the rest of this series runs on: the arguments in part 3, the forecast flags in part 4, and the promise ledger in part 5.
Salesforce's 2024 State of Sales survey of 5,500 sales professionals found that 80% of reps on teams using AI said it was easy to get the customer insights they need to close deals, against 54% at organizations without it. It is a vendor-run survey and a correlation, not proof of cause. The direction is still the point: the insight is already in the conversations, and the work is getting it out in a shape you can use.
How accurate is it, and who checks?
It will misread things. A buyer thinking out loud can sound like a requirement. A number said as an example can look like a budget. So nothing goes straight into the CRM. The signals arrive as a drafted record with the sentence beside each field, and the rep approves, edits or rejects them, which takes about two minutes per call. Part 7 covers what happens to the edits, and why they are the most useful data the system gets.
Who this is not for: a team that does not record calls. Threads and notes still carry signals, but the recording is where most of them come from. On the Northwind deal it was fourteen of eighteen.
Next in the series
Part 3, It does not move your numbers. It argues with you about them, takes the contradictions from this step and shows what the rep actually sees: one deal, two arguments, each with the sentence behind it. The series page lists every part in order.
Common questions
- What is the difference between a call summary and a typed signal?
- A summary is a paragraph about the call. A typed signal is one fact with a type, a value, a speaker and a timestamp, such as a paper process: security review takes six weeks and has not started, said at 31:14. A summary can be read. Signals can be counted, filtered and compared with the CRM record.
- Do email threads and rep notes add anything if the call is recorded?
- Yes. Threads carry forwards, new people on cc and written requests, and notes carry what happened off the recording. In the Northwind example, four of eighteen signals came from the thread or a rep note, including the forward that showed the CFO was watching.
- Does anything get written to the CRM without the rep approving it?
- No. Signals arrive as a drafted record with the sentence beside each field, and the rep approves, edits or rejects each one. Stage, close date and amount are never written by the system at all. They arrive as recommendations with the quote behind them.
Sources
Written by Jared Hawkins-Young from the system he builds and runs. Diagrams are drawn from the same specs as the software. Published October 8, 2026. Illustrative numbers are labelled as such; every other figure carries its source.