Buyer interest capture at registration without adding twelve more questions

The exhibitor side of your matching engine arrives free. Categories are on the contract, the sales team collects them because the floor plan needs them, and by the time the profile deadline passes you know what most of the floor sells.

Buyer interest capture has no equivalent. Nothing in the commercial process makes a visitor tell you what they came to source, so the only route is asking, and every question you add to the registration form costs completions. That trade is the whole subject, and most shows settle it by argument in a meeting six weeks before registration opens, with the audience acquisition lead and the matchmaking lead each certain the other is wrong.

It can be settled with an experiment on your own form instead, in one week, with two numbers.

What you already have without asking

Before adding anything, count what the file already contains.

Job title and company name are on every registration, and both carry weak category signal. A company called Midlands Bakery Supplies tells you something. A job title of packaging technologist tells you something. Neither is reliable enough to route a meeting on its own, and both are free.

Prior editions are the stronger source. A returning buyer who took four meetings last year has a category history nobody had to ask for, and a returning visitor who scanned eleven stands has a better interest profile than any picker will produce. On most shows, returning registrants are somewhere between a third and half of the file, so half your buyer side interest problem is a matter of joining to last year rather than asking a new question.

That leaves the first-time registrants, who are the ones you know nothing about and the ones the matching programme most needs to place. The picker exists for them.

What does one more question cost?

Something, and the number is knowable rather than a matter of belief.

Galesic and Bosnjak, writing in Public Opinion Quarterly in 2009, reported that a longer stated questionnaire length lowered the share of people who took part, and that questions placed later drew shorter and more uniform answers. Registration forms behave the same way, with an extra mechanism working against you: the registrant is not a survey participant doing you a favour, they are a buyer trying to get a badge, and anything between them and the badge is friction they did not ask for.

The mistake is treating that cost as a reason to ask nothing. A show that captures no interest data has a matching programme running on job titles, which produces exactly the generic proposals that make buyers stop opening the app. The cost is real and it has to be weighed against something, so measure both sides.

Running the picker test properly

Split registrations randomly, not by day and not by channel. Two arms, 2,000 registrations each, running for however many days it takes to accumulate them.

Arm A gets a five item picker drawn from the top of your category tree. Arm B gets fifteen items from the second level. Both are optional, both use the same wording, both sit in the same position in the flow. Everything else stays identical, which is harder than it sounds, because somebody will want to run a promotional email in the middle of it.

Measure four things. Completion rate from the page carrying the picker. Mean categories selected per completer. Share of completers selecting exactly one option. And the position of the options chosen, which needs the display order recorded per registrant.

Two thousand per arm is not an arbitrary round number. The completion rates you are comparing sit near 90 per cent, where the standard error on a single arm is the square root of 0.9 times 0.1 divided by 2,000, which is 0.0067, so the difference between two arms carries an error of about one point. That resolves a five point effect comfortably and a one point effect not at all. If your show takes 3,000 registrations in total, run the test across two editions or accept that only a large effect will be visible.

Suppose the result comes back like this. Arm A completes at 92.4 per cent, which is 1,848 of 2,000, with a mean of 1.7 categories selected. Arm B completes at 86.1 per cent, which is 1,722, with a mean of 2.9.

Reading the two numbers together

The completion difference is 126 registrations on 2,000, which is 6.3 points. Scaled to a show taking 14,000 registrations through that page, the long picker costs about 882 registrations.

The data difference runs the other way. Arm A produced 1,848 times 1.7, which is 3,142 category records. Arm B produced 1,722 times 2.9, which is 4,994. The long picker captured 1,852 more category records from the same 2,000 registrants, and at show scale that is roughly 13,000 more.

So the trade is 882 registrations against 13,000 category records, and the answer depends on which one your show is short of. A show fighting for audience volume in a soft year takes the registrations. A show with strong volume and an exhibitor base complaining about lead quality takes the categories.

My own preference is the five item picker in the flow, with the longer list moved to a confirmation step, and the reason is that the two costs are not equally recoverable. A registration you lost is gone, and you will never know which buyer it was. A category you did not capture can be asked for again next Tuesday, in an email, from somebody who has already committed to attending.

There is a third number worth putting next to those two. In Arm B, 31 per cent of completers selected exactly one category against 18 per cent in Arm A. A single selection from a fifteen item list is often a person who has stopped reading, which means part of the extra 1,852 records is depth and part is people ticking the first plausible thing to get past the page.

Option order changes what you capture

Krosnick and Alwin found in Public Opinion Quarterly in 1987 that the order in which options are presented changes which ones respondents choose, and that the effect follows from how people process a list rather than from what they want.

This is easy to test and almost nobody does. Randomise the display order per registrant, store the order shown, and compare the selection share of each category by the position it appeared in. If your top category is picked by 34 per cent of registrants when it appears first and 21 per cent when it appears eighth, then roughly a third of its apparent popularity is position.

The consequence is not academic. Exhibitors in the category that happens to sit at the top of an alphabetical list get more buyer demand attributed to them, your demand-side counts by category are distorted in a fixed direction every edition, and the taxonomy work in the tree design from I7 gets judged against numbers that are partly an artefact of the form.

Randomising fixes the measurement. It also removes an argument with the exhibitor whose category always appeared eleventh.

Where should the long list live?

After the confirmation, in a separate step, framed as something the registrant gets value from.

The confirmation page is the best real estate in registration and it is usually wasted on a calendar link. The badge is secured, the friction cost is gone, and the request can be honest: pick the categories you want meetings in and we will use them to build your schedule. A buyer who has already registered and is being offered meetings has a reason to answer that a buyer trying to get a badge does not.

Ask at level two of the tree, never at leaf level. A buyer will not scroll 400 leaves, and the matching engine can climb or descend from whatever they give you, which is the depth question in I9. Capture the coarse declaration from the buyer and let the engine do the rest against the exhibitor side, where profile text has been mapped to leaves as I8 describes.

Two more places worth using. The session agenda, where a saved session in a category is a declaration nobody had to type. And the meeting request itself, where a buyer asking for a specific exhibitor tells you more than a picker ever will, which is I13's subject.

Where interest capture stops

Stated interest is a claim about a future that has not happened. A buyer ticking industrial automation in March is describing a project that may be cancelled by September, and the matching engine will still be routing meetings on that tick in October.

The second limit is the one to watch on any show with a large international audience. Category vocabulary translates badly, and a picker that reads clearly in English produces near-random selections from registrants working in a second language. Compare mean selections and single-selection share by country before concluding anything about demand from a market.

The third is that the buyers you most want are the ones least likely to answer. Senior people registering in ninety seconds from a phone skip optional questions at a higher rate than junior staff completing a form at a desk. Any interest capture strategy that relies only on the registration form therefore under-represents exactly the buyers your exhibitors are paying for, and the fix is the returning-buyer history and the behavioural signals rather than a longer form.

Start by adding one column to your registration export: the number of interest categories selected. Group it by first-time and returning, and by registration channel. If first-time registrants average under one category and returning ones average three, your picker is not the problem, and the matchmaking gap you have been trying to close with better scoring is a gap in what you asked for at the front door.