Why most matchmaking fails

Tradeshows run on matches. I keep coming back to this. Everything else — the badges, the booths, the coffee — is logistics in service of getting the right two people in front of each other. So when matchmaking falls flat, the whole show underdelivers. Exhibitors go home with a stack of business cards and no pipeline. Attendees sit through meetings that had nothing to do with why they came. And you're the one who gets blamed for both.

Here's the uncomfortable part. Most event matchmaking doesn't fail because the algorithm is bad. It fails because the data feeding it is thin.

I've spent a lot of the last few years staring at that problem, and here's what I've watched happen again and again. Organizers invest in a matchmaking engine, wire it into their app, and expect relevance to show up on the floor. Then the meetings happen, and half of them miss. The instinct is to blame the algorithm — or the attendees, for "not engaging." But the algorithm did exactly what it was told. It was told very little.

Before I pull that apart, let's agree on the thing itself. Event matchmaking is the process of pairing attendees and exhibitors at a tradeshow or conference based on what each side actually wants — buyers with relevant suppliers, partners with partners — using registration data, enriched company data, and an algorithm to schedule meetings that are worth both people's time.

You're matching on a form nobody wanted to fill out

Think about where your matchmaking data actually comes from. A registration form. A few dropdowns, a "select your interests" grid, a job title field, some optional questions almost nobody answers.

That's declarative data — what people tell you about themselves when they sign up. And I want to be honest about what really happens on those forms.

People are in a rush. They're registering from their phone, in a meeting, three weeks before the show, thinking about anything but your taxonomy. So they tick two boxes out of twenty, and they pick the most generic ones. They skip every optional field. They type "Manager" or "CEO" into the title box whether it's precise or not. The custom fields you fought your team to add — the ones that were supposed to power great matching — come back half empty, or filled with noise.

None of this is because attendees don't care. It's because a form is a tax, and people pay the smallest amount they can get away with. Every extra field you add lowers completion and raises the odds the answers are junk. So you're stuck: ask for more and lose people, ask for less and fly blind.

Then your matchmaking engine goes to work — on that.

It matches two companies who both selected "software," and calls it a lead. One sells CRM to retailers. The other builds firmware for factory robots. On paper, a perfect match. In the meeting room, two people wondering who scheduled this. It matches a "Manager" who's actually a junior coordinator with no budget to a VP who came to close deals. Both walk away thinking the event wasted their time.

Good matchmaking on bad data isn't matchmaking. It's a random number generator with a nice interface.

Why event matchmaking fails: your data is broken in three ways

When I look closely, thin data isn't one problem. It's three, and each one breaks matching in its own way.

It's missing. Fields left blank, questions skipped, whole sections ignored. The engine can't match on what isn't there, so it falls back to the little it has — usually the most generic signal in the profile.

It's vague. "Software," "consulting," "services," "innovation." Words that are technically true and completely useless for matching. Two people can share every tag on your form and have nothing to sell each other.

It's wrong. Inflated titles, outdated companies, a personal email with no company attached, the interest someone ticked by accident. This is the dangerous one, because the engine trusts it completely and matches with confidence in the wrong direction.

Missing data gives you no match. Vague data gives you a shallow one. Wrong data gives you a confident, specific, completely misfired one. And most registration data is some mix of all three.

Declarative data is the seed, not the answer

I'm not saying declarative data is useless. It's the seed, and some of it you can't get any other way. When someone tells you they're here to source suppliers, not to sell — that's intent data, and it's gold. No enrichment tool on earth can infer that reliably. The one or two things a person actually bothers to tell you about why they came are the most valuable inputs you have. Keep asking for them, and design your form to protect them by not burying them under twenty fields nobody finishes.

But you can't stop there. The gap between what someone types into a form in fifteen seconds and who they actually are is exactly where matchmaking dies. Match on the form, and you match on adjectives. Match on who they are, and you match on business.

Closing that gap has a name: enrichment. Filling the blanks the form left behind.

From a thin profile to a real one

Start with what you've got. Often a registration gives you three things: a name, a company, an email. That's the seed.

From that seed you can pull a lot more than people expect. What the company actually does. Its size, sector, and where it operates. The person's real role and seniority. Recent signals about what the company is in market for — hiring, funding, a new product line, expansion into a region. This is firmographic and contextual data — facts about the business — not self-reported adjectives about it.

That's the raw material. On its own, though, it's still messy: different sources describe the same company three different ways, titles don't line up across regions and languages, and half the signals are noise. Raw enrichment data can be as unusable as the blank form was.

This is where AI earns its place. Not as magic — as the layer that turns messy inputs into something your matchmaking can actually read. It does three concrete jobs.

It normalizes. "Sr. Mgr, Bizdev," "Business Development Manager," and "Responsable du développement" all become one clean, comparable role. Your engine can finally treat like as like.

It infers. From what the company does, it fills the interests the attendee never ticked. A procurement lead at a mid-market grocery chain almost certainly cares about cold-chain logistics and packaging suppliers, whether or not they said so. The AI writes that in — grounded in the company, not invented.

It reads intent. Combining the declarative signal ("I'm here to source") with the firmographic picture, it flags whether this person is here to buy, sell, or partner, and gives your matching a direction, not just a topic.

The result: "Software" becomes "B2B CRM for mid-market retailers." "Manager" becomes "Head of Procurement." The blank interest field fills from what the company visibly does, not from the two boxes someone had patience to tick. Same attendee. Same five-minute registration. A profile you can finally trust.

The prompt that does the work

Here's the shape of it. You hand the model the sparse profile plus the enriched firmographic data, and you ask it to complete and sharpen — with one rule that matters more than the rest: no guessing dressed up as fact.

You're enriching an attendee profile for tradeshow matchmaking.

Here's what we know:
- Name: {name}
- Company: {company} — {what the company does, from enrichment}
- Stated title: {raw title, e.g. "Manager"}
- Stated interests: {raw interests — often blank or generic}
- Stated goal: {why they came, if given}

Do three things:
1. Normalize the title into a real role and seniority level.
2. Infer 3–5 specific interests this person likely has, grounded in
   the company's actual business — not the generic boxes they ticked.
3. Flag whether they're here to buy, sell, or partner, and why.

Rules:
- Be concrete. "Software" is not an answer; "B2B CRM for retail" is.
- Ground every inference in the company data.
- When you're guessing, say so. Never present a guess as a fact.

That last rule is the one people skip, and it's the one that keeps the whole system honest. An enrichment that quietly invents facts is worse than a blank field, because your engine believes it. A good prompt makes the model show its work: here's what I know, here's what I'm inferring, here's where I'm unsure. You keep the confident parts and flag the rest for a human — or for the attendee to confirm in one tap.

"Isn't this creepy?"

Fair question, and I want to answer it straight. There's a line between enrichment and surveillance, and it's not blurry. You're working from business facts — what a company does, what a role means, what a firm is publicly in market for. You're not profiling private lives. The test I use is simple: would this feel reasonable if the attendee saw it? A match explained by "you're both in cold-chain logistics" passes. Anything you'd be uncomfortable showing them doesn't belong in the profile. Enrichment done right doesn't hide from attendees — it gives them a more personalized event experience and can show its reasoning when asked.

What this means for your next show

I've stood in enough half-empty halls to know the organizers who win the next few years won't be the ones with the cleverest matchmaking algorithm. Good algorithms are becoming table stakes; everyone will have one soon. The edge is the data underneath it — declarative where it counts, enriched everywhere else, and always live.

That's the bet I'm making, and it's why we built Mytradeshow.ai: unify the data into one event data graph, put AI agents on the busywork of completing and sharpening it, and let the matches run on something real. From data to contracts.

If you want to go deeper, a few threads I've pulled on separately: why a tradeshow is really a two-sided marketplace and what that changes about how you run one, how to engineer serendipity at your event instead of hoping for it, and how to run experiments on your event so next year's decisions rest on evidence, not gut.

Matches are the whole point. Feed them well.