// INSIGHTS
Smart Matchmaking: Turning an Attendee List Into Meetings
6 min read
An attendee list is not a network. Two thousand names in a searchable directory produces a lot of scrolling and very few meetings. Matchmaking is the engineering that closes that gap: a profile schema worth matching on, a scoring function you can explain in one line, a request-and-accept flow tied to real time slots, and a physical meeting area that actually exists on the floor plan. Skip any one of those four and you are back to a directory.
Matching is scoring, not magic
The schema decides the ceiling. Job title and company name are almost useless for matching, because a Vice President of Business Development at a logistics firm could want six different things. What works is intent, captured as two structured fields: what this delegate is looking for, and what they can offer. Six to eight structured fields is the practical limit. Past that, completion rates collapse and you end up matching on empty profiles, which is worse than not matching at all.
The scoring function should favour complementarity over similarity. Two people who both sell warehouse automation have very little to discuss. One who sells it and one who is procuring it have a meeting. So the weighting puts the highest score on an offer that matches a stated need, a lower score on shared sector or region, and a small score on shared session attendance, which is a decent proxy for shared interest and comes free from your QR scan data.
Explainability is not optional. Every suggested match should carry two visible reasons, in plain language, such as a stated need in cold-chain logistics against a stated offer in the same area. Delegates ignore a ranked list they do not understand, and they trust a short list they do. This is also where the honest limit of AI sits. Weighted rules over structured fields will get you most of the way. Embeddings over free-text bios help when delegates have written something meaningful, and add nothing when the bio field is empty, which it usually is.
What the maths has to respect
What a matchmaking engine needs
Intent-based profile schema
Structured looking-for and can-offer fields drawn from a controlled vocabulary, not free text. Controlled vocabulary is what makes scoring possible and reporting meaningful.
Explainable scoring
Weighted overlap that favours complementary intent, with the top two contributing factors surfaced in the interface. If the reason cannot be shown, the weight should not be in the model.
Slot inventory with hard limits
Every delegate publishes available slots and the system prevents double-booking on either side. A popular exhibitor may receive eighty requests against twelve slots, and the software has to handle that gracefully.
Confirmations on the channel people read
Push notification plus WhatsApp for the meeting confirmation and a reminder fifteen minutes before. Email confirmations for on-site meetings are read after the meeting has finished.
From CSV to a confirmed meeting
Enrich at registration
Make four intent fields mandatory in the registration form, before payment. Chasing profile completion after the fact typically gets you a fifth of the list.
Open profiles two weeks out
Delegates need time to browse, request and negotiate. Matchmaking that opens on the morning of day one produces meetings on day two at best.
Run matching, then let humans override
Generate the ranked suggestions, then give the organiser a console to force or block specific pairings. Organisers know things about their delegate list that no scoring function can see.
Check in at the table
Numbered tables in a staffed meeting zone, with a QR scan at the start of each meeting. That scan is the only honest measure of whether the matchmaking worked.
What usually goes wrong
Empty profiles are the number one killer. If only a fifth of delegates complete their intent fields, the engine has nothing to work with and the suggestions look random, which destroys trust in the feature for the rest of the event. The only reliable fix is to move those fields into the registration flow itself and make them mandatory. Nobody abandons a paid registration over four dropdowns, and everybody ignores a profile-completion reminder email.
The second failure is physical. Matches get made, meetings get accepted, and then there is nowhere to hold them, so two people mill about near the coffee counter and give up. A no-show rate of a third is normal without numbered tables, a staffed zone and a reminder fifteen minutes before. The third failure is over-matching. Sending a delegate thirty suggestions is functionally identical to sending none, because both produce zero action. Cap suggestions at five to eight per day, refresh them daily, and let the delegate search manually if they want more.
Common questions
Do we actually need AI for this?
Not for most events. A weighted rules engine over structured intent fields is transparent, fast, cheap to tune, and easy to explain to a delegate who asks why they were matched. Machine learning earns its place when you have rich free-text profiles or several years of historical meeting outcomes to learn from, and not before.
What is a realistic meeting completion rate?
Of the meetings that get accepted, expect a meaningful share to not happen without reminders and a staffed meeting zone. With numbered tables, a fifteen-minute reminder and a check-in scan, that gap narrows considerably. Report accepted meetings and checked-in meetings as two separate numbers, because conflating them makes the post-event report dishonest.
Does matchmaking work for a 300-person event?
It works better, in relative terms. At 300 delegates the whole list is browsable, so matchmaking competes with simple search, but the slot and reminder infrastructure still adds real value. Below about 150 delegates, a well-designed directory with good filters usually delivers the same outcome for less money, and we will say so.
Turn your delegate list into a meeting schedule
Share your registration form and we will show you which fields make matching possible.
