Predicting no-shows before the night of the dinner

Basecamp ran the same breakfast twice. The free version drew 84 registrations and 38 attendees, a show rate of about 45%. The second version asked for a refundable $100 deposit. Registrations fell to 55, and 50 people came, a 90% show rate (Skift Meetings, 2026). Fewer sign-ups, more people in the room.

The obvious lesson is "charge a deposit." Sometimes that is right. For an executive dinner, it often is not an option. The more useful lesson is that attendance is predictable from signals you already have, and commitment is one of the strongest.

This workflow builds a no-show score from those signals, uses AI to write reminders by risk tier, and saves human effort for the guests who matter most and are most likely to drift.

Step 1: Know your baseline

Before predicting anything, measure what you have. Pull every hosted event from the last 12 to 24 months with two numbers each: registered and attended.

Then compare against published benchmarks, keeping their sources in mind. Across more than 860 events, PheedLoop found a median no-show rate of about 28% for free events and about 17% for paid events, with an overall median around 20% (PheedLoop, 2026). A vendor dataset from Clutch Events, covering 191 events, found attendance rates of 70% for executive dinners, 61% for private roundtables, 52% for networking dinners, 38% for practitioner-led conferences and 19% for vendor-owned conferences (Clutch Events, 2026). Both are vendor research. Use them as a sanity check, not a target.

If your dinners are running well below those figures, the problem may be the list, not the reminders.

Step 2: Pick signals, and be honest about which are proven

Five signals are worth collecting. The published research supports two of them. The rest are hypotheses your own history can confirm or kill.

Commitment has the strongest evidence behind it. PheedLoop found registrations with zero ticket purchases had a median no-show rate of about 34%, against about 17% for six or more tickets. At the individual level, paying attendees showed up 2 to 4 percentage points more often than non-paying attendees at the same event, a pattern observed in 57 to 65% of the 195 events analyzed (PheedLoop, 2026). For free events, look for proxies: did they answer a dietary question, add a colleague, reply to the confirmation?

Seniority and role come next. Clutch Events found directors and above attend at a 62% rate, and technical practitioners show three times the no-show rate of senior leaders (Clutch Events, 2026). Vendor data again, but specific enough to test against your own.

Past attendance has no external benchmark, and it is the signal most worth building. Did this person attend the last event they registered for? Freeman puts the industry's blended year-over-year attendee retention at only 30 to 35% (Freeman, 2025), so someone who has actually come back to your events is already unusual.

Registration timing and email engagement are untested. The question is whether someone registered months out or last week, and whether they opened the confirmation and the first reminder. No study in our research quantifies either signal for B2B events. Collect them and see whether they separate your attendees from your no-shows.

Travel distance is the fifth: office city versus venue city. Use the work location in the CRM, never a home address. Again, a hypothesis to test.

Step 3: Build a score you can explain

Resist anything clever. A points model in a spreadsheet is enough, and you can defend every number in it. A starting version might look like this:

+3  attended the last event they registered for
+2  director level or above
+2  paid, or completed an optional registration question
+1  opened the confirmation email
+1  office within 30 minutes of the venue
-2  registered more than six weeks out with no engagement since
-2  registered for a past event and did not attend

0 to 2 = high risk, 3 to 5 = medium, 6+ = low

The weights are placeholders. Now check them. Score the registrants from your last three to five events as if you did not know the outcome, then compare with who actually came. If high-risk guests did not no-show more often than low-risk ones, change the weights or drop the signal. This backtest is the only thing that turns the score from a guess into a tool.

AI can help here. Upload the historical sheet and ask:

This table lists registrants for five past events with columns
[list] and an "attended" column (yes/no). Using only this table:
1. For each column, compare the attendance rate of registrants who
   have the attribute with those who do not.
2. Flag any column where the difference is based on fewer than
   15 people.
3. Do not build a predictive model. Show the counts behind every
   percentage.

The last line matters. Ask for counts, not a model. With a few hundred rows, a fitted model will find patterns that are noise, and you will not be able to tell.

Step 4: Draft reminders by tier

Each risk tier gets a different message. Low-risk guests need logistics. Medium-risk guests need a reason. High-risk guests need an easy, guilt-free way to cancel, because a clean decline is worth more than a silent no-show.

That last point has data behind it. Clutch Events found roundtables see about 11 advance cancellations for every day-of no-show (Clutch Events, 2026). In that format, most people who drop out tell you first. A cancellation a week out lets you fill the seat from the waitlist. A no-show on the night does not.

Prompt for the drafts:

Write three reminder emails for a [format] on [date] at [venue],
hosted by [name, title]. Under 90 words each.

Version A, confirmed low-risk guests: logistics only. Time,
address, parking, dress, who to call on the day.
Version B, medium-risk guests: one line on why this conversation
is worth their evening, based on this agenda: [paste agenda].
Version C, high-risk guests: ask them to confirm or release their
seat by [date], and make declining easy and gracious.

Use only the facts in this prompt. Do not name other guests,
invent speakers, or promise anything not listed here.

Step 5: Call the top tier

For the handful of guests who are both high value and high risk, a machine should not send the reminder. A person should call, or the account owner should write a personal note, three to five days before. This is the confirmation step Basecamp got from a deposit, and you are getting it from a conversation instead.

Keep the list short. If your call list is longer than a quarter of the room, the score is too cautious or the invite list needs work.

What to check

• Every merge field in every reminder. Wrong venue, wrong date or wrong host name on a VIP email is worse than no email.
• Invented detail. Models like to add a speaker, a menu item or a "limited seats" line. Delete anything you did not supply.
• The score on small events. On a guest list of 15, one person swings the rate by nearly seven points. Treat the score as a triage tool, not a forecast.
• Privacy. Keep signals to business data the guest gave you or that sits in your CRM. Do not scrape social profiles to guess who is traveling.

Where it goes wrong

The score reflects the past. A new format, city or audience will behave differently, and your weights will be wrong until you have data. A score can also become an excuse to over-invite. Adding a third more guests because the score says a third will vanish is fine for a conference. At a dinner for 16, it can mean turning people away at the door.

After each event, add the outcome to the sheet and run the backtest again. After five events, you will trust the score. After ten, you will know which signal actually predicts your audience.