Post-event follow-up that doesn't read like it was written by a machine

The most cited study on lead response time is 15 years old, and it still stings. In a 2011 audit of 2,241 US companies, researchers writing in Harvard Business Review found the average response to a web lead took 42 hours, and 23% of companies never responded at all (Oldroyd, McElheran and Elkington, Harvard Business Review, 2011). Firms that replied within an hour were nearly seven times as likely to qualify the lead as firms that waited even one hour longer.

Treat that as an old baseline. It measured web leads, not dinner guests, and it predates most modern sales tooling. A 2022 test by Chili Piper, a scheduling vendor, found that about 30% of B2B software vendors never answered a demo request, and those that did took 4 hours 50 minutes on average (Chili Piper, 2022). Neither study measures event follow-up directly. Both point the same way: interest fades fast and companies are slow.

Event teams know this. Ninety-one percent of B2B events leaders say improving follow-up is a priority (Forrester via Marketing Week, 2025). AI makes speed easy. It also makes it easy to send 40 emails that sound as if nobody from your company attended. Gartner found 73% of B2B buyers actively avoid suppliers that send irrelevant outreach (Gartner, 2025). Fast and generic is a bad trade.

What follows is a six-step workflow for a hosted event of roughly 15 to 60 guests. To make it concrete, picture a hypothetical dinner for 24 on a Tuesday night.

Step 1: Build the follow-up table before anyone writes a word

Every good follow-up email starts as a row in a table. Build it the morning after, with these columns:

name | company | title (as registered) | attendance (attended / no-show / cancelled)
| CRM status (open opportunity + stage / customer / known contact / not in CRM)
| owner | host notes

Attendance and deal stage together decide the message. A guest with a late-stage opportunity gets a note from the account executive that picks up the actual conversation. A no-show from a target account gets something useful from the evening, with no guilt attached. A customer gets thanks from the person who manages the relationship. Someone not yet in the CRM gets created there before anything is sent.

Step 2: Get the specifics the same night

The one input AI cannot produce is what actually happened. Before the hosts leave the venue, or first thing the next morning at the latest, run a 10-minute debrief. One line per guest each host spoke with: what they asked about, what they are working on, anything that was promised.

"Asked how we handle migration from their current system; wants the reference call we mentioned" is a note. "Great conversation" is not.

Pay special attention to requests for expertise. Freeman found that 58% of attendees want to speak with subject matter experts after events, while only 26% of exhibitors plan for it (Freeman, 2025). Those requests are the highest-value follow-ups on the list and the easiest to lose.

If a guest has no note, that is information too. Nobody talked to them. Flag it now rather than papering over it later.

Step 3: Prompt for specificity, and forbid the rest

The prompt has one job: make the model use the notes and refuse to fill gaps. Here is a drafting prompt for attendees.

You are drafting post-event follow-up emails for [company] after a
dinner for [audience] on [date] in [city].

For each row in the table below, write an email of no more than 90
words from the listed owner.

Rules:
- The first sentence must reference the specific note for that guest.
  If the notes column is empty, do not write an email. Output
  "NEEDS NOTE" and the guest's name instead.
- Do not use these phrases: "great to connect", "hope you enjoyed",
  "touch base", "circle back", "reach out", "exciting".
- Do not state or guess the guest's job title, company news, or
  anything else not in the table.
- Include exactly one next step. Use the one promised in the notes if
  there is one. Otherwise suggest one concrete, low-effort option.
- Plain, direct sentences. No exclamation marks.

Table:
[paste rows]

The NEEDS NOTE rule matters most. It stops the model from inventing a conversation and sends the gap back to a human. The title rule exists because models will cheerfully supply a job title that was accurate two years ago.

For no-shows, a separate prompt:

Draft a three-sentence note to a guest who registered for [event]
but did not attend. Do not mention that they missed it and do not
express disappointment. Share one specific point from the evening,
taken from the summary below, that fits the role listed for them.
Offer to send [asset] or set up a 20-minute call.

Summary of the discussion:
[paste five bullets]

Then run a critic pass over the full set of drafts:

Below are 24 follow-up drafts and the source table. Flag any draft
that (a) could be sent to a different guest with only the name
changed, (b) states a fact that is not in the source table, or
(c) commits us to anything: pricing, timelines, introductions, or a
specific person joining a call. List flagged drafts by guest name
with the reason.

The interchangeability test in point (a) does most of the work. If a draft could go to anyone, it should go to no one.

Step 4: Route by owner and set a clock

Who sends the email matters as much as what it says. Guests with open opportunities hear from the account executive. Customers hear from their account owner. Everyone else hears from the host who met them, or from a named rep if the host is an executive who will not send 15 emails.

Then agree a deadline. A workable default is 24 hours for guests with open opportunities or anyone who asked for something specific, and 48 hours for the rest. The HBR data is old and was about web leads, but nothing since suggests that interest improves with waiting.

Step 5: What to check

A human reads every draft before it goes out. Most marketers already work this way: HubSpot, a vendor, found only 4% of marketers use AI to write whole pieces on its own (HubSpot, 2025). The check is where that habit earns its keep.

• Trace every fact. Each specific claim should map to a note or a table cell. If it doesn't, cut it.
• Check that the note matches the person. Models working through a long table sometimes attach one guest's detail to another's email. Keep batches small, eight to ten rows, so errors are easier to spot.
• Delete added titles. If the model named a role, confirm it or remove it.
• Clear commitments with the owner. Anything that promises pricing, a timeline or an introduction needs the sender's approval.
• Read it aloud. If it sounds like a newsletter, rewrite the first line yourself.

Step 6: Decide what never gets automated

Some messages should always start and finish with a person, even if AI helps with a first draft.

Replies come first. Once a guest writes back, a human answers. The same goes for anyone in active negotiation or procurement, where a stray phrase can create an expectation the deal team did not intend.

A guest who raised a complaint or a problem during the evening should hear from someone senior, in their own words. So should any executive-to-executive note. A draft your CEO rewrites is fine. A note sent in your CEO's name that your CEO never saw is not.

Salesforce, another vendor, reports that 84% of marketers admit they still run generic, one-way campaigns (Salesforce, 2026). Post-event follow-up is one of the few places where specificity costs almost nothing, because the material already exists in the hosts' heads.

The best measure of this workflow is simple. Count how many drafts came back NEEDS NOTE. If it is more than a handful, the fix belongs in the room, and no prompt will cover for it.