Every meeting ends with a recap. Every thread gets an answer or an owner.
Draft the recap minutes after the call, every claim traced. Answer questions from approved material or route them, and revive quiet deals on their own rhythm.
Key takeaways
- A recap is drafted minutes after the call, says back what the buyer said in their words, and traces every claim to a timestamp. The rep sends it.
- Inbound questions are answered only from approved material, with the source cited, or routed to a named owner. Anything priced, promised or contractual always routes.
- A deal is cold when it breaks its own reply rhythm, not a fixed rule: a buyer who replies every three days triggers an alert after twelve quiet days.
- The follow-up on a quiet deal opens on the issue still open on the ledger, never on "just checking in."
Automate the draft, not the send: have the recap written from the call within minutes, with every claim traced to the moment it was said, and leave it for the rep to send. Treat every inbound question the same way, answering it from approved material with the source cited or routing it to a named person. When a deal goes quiet, measure the silence against that deal's own rhythm and open the follow-up on the issue still outstanding.
What should a recap after a sales call say?
Dana,
You said the summary is not the problem, the fields are: everyone updates HubSpot from memory on Friday, and two deals slipped last quarter before you saw them.
Four reps, not six. The revised quote is attached at four seats with onboarding included. The SOC 2 report is attached too, so it is not waiting until Tuesday.
On what it touches in HubSpot: I will send a written scope note for your security lead by Thursday.
You said the review takes six weeks and has not started. If your security lead can make the 14th, I will bring the questionnaire already filled in.
Marcus
A good recap does four things. It says back what the buyer said, in their words, so they recognise it. It states what changed, here the move from six seats to four. It lists what each side owes, with dates, straight from the ledger in part 5. And it ends on one ask.
Marcus's draft opens on Dana's own complaint: the summary is not the problem, the fields are. It confirms the scope cut and attaches the revised quote. It attaches the SOC 2 report early. It says when the HubSpot scope note will arrive. And it asks for one thing, the security lead on the 14th. Every sentence traces to a timestamp, which is why it reads like Marcus wrote it, not like a summary tool did.
Speed matters for a second reason. Dana will forward this. A recap is the document a champion forwards when selling internally, to a CFO or a security lead the rep has never met, so it has to survive the forward. Drafted minutes after the call, it reaches her while the conversation is still fresh on both sides.
How should an AI answer a buyer's questions by email?
Carefully, and less often than it could. Each inbound question is split into parts, and each part is checked against your approved material: the security FAQ, the SOC 2 report, the published product docs. A part the approved material answers is drafted with the citation, and the rep presses send. Anything priced, promised, contractual or outside that material goes to a named owner with the thread attached. Nothing is guessed.
Dana's next email asks two things at once. How long do you keep the transcripts you read? The security FAQ answers that, so the draft quotes it and cites the section. Could we start at three seats? That is a pricing question, so it goes to Marcus. One email, two outcomes, and neither of them is a send.
| Question | Answerable from approved material? | What happens |
|---|---|---|
| "Can we see your SOC 2 report?" | Yes, the report itself | Drafted with the report attached |
| "How long do you keep transcripts?" | Yes, the security FAQ | Drafted with the section cited |
| "Could we start at three seats?" | No, it is priced | Routed to the rep with the thread |
| "Will you sign our DPA as written?" | No, it is contractual | Routed to whoever owns contracts |
| "Does it work with our sequencer?" | Only if the product docs say so | Drafted with the doc, or routed to the solutions engineer |
The bar is high on purpose. Gartner surveyed 645 B2B buyers in August and September 2025 and reported in May 2026 that 51% say they are more likely to encounter misleading information from generative AI. A vendor whose own answers might be invented is the last thing a buyer in the middle of a security review needs. In the first weeks most questions route. That is the system working, not failing.
The same router handles replies to cold email. Outbound part 14 covers how an interested reply becomes a meeting on the rep's calendar, under the same rule: approved material or a person, never a guess.
When is a quiet deal actually cold?
A fixed rule, such as no reply in seven days, treats every buyer the same. It fires too early on a buyer who always takes a week and too late on one who answers the same afternoon. Measure each deal against its own rhythm instead. Dana replied about every three days, so the alert fires after twelve quiet days, four times her usual gap.
The follow-up does not say "just checking in." It opens on the row still open on the ledger: the introduction to the security lead that Dana promised at 38:40, three weeks earlier. It offers to send the security packet straight to the reviewer and attaches the questionnaire already filled in. A follow-up that remembers exactly what someone promised does not read like automation. It reads like the rep who never forgets anything.
What stays with the rep?
- Pressing send, on the recap, on every drafted answer and on the follow-up.
- Anything priced, promised or contractual, whoever asked and however simple it looks.
- Whether to send the follow-up at all. Some deals are quiet for a reason the rep knows and the thread does not.
Who this is not for: deals that close on one call, where there is no thread to answer and no rhythm to measure. And the router is only as good as the approved material behind it. If your security answers live in one person's head, write them down first.
Next in the series
Every draft in this part works because every claim in it can be checked. Part 7, Press on anything and hear where it came from, is about that check: what a rep sees when they press a field, and what happens when they correct it. The series page lists every part in order.
Common questions
- Should AI send follow-up emails after sales calls automatically?
- It should draft them, not send them. A recap drafted minutes after the call, with every claim traced to the recording, saves the rep the writing. The rep reading it and pressing send is what keeps a wrong claim from reaching the buyer.
- How does the system decide whether it can answer a buyer's question?
- It splits the question into parts and checks each against approved material such as the security FAQ, the SOC 2 report and the product docs. Parts the material answers are drafted with the citation. Anything priced, promised, contractual or not covered goes to a named person with the thread.
- When should a rep follow up on a deal that has gone quiet?
- When it breaks its own rhythm. Measure how often that buyer usually replies and alert at a multiple of the gap, four times in the example here, then open the follow-up on the specific issue still outstanding rather than a generic check-in.
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.