Onboarding questions for matchmaking that people will actually answer

The form review meeting runs the same way every year. Audience acquisition wants three fields, because every extra question costs registrations and the target is a number on a slide. Matchmaking wants nine, because product fit needs categories and authority fit needs a purchasing role. Both sides have a point and neither has a measurement, so the outcome depends on who is more senior.

Onboarding questions for matchmaking are a trade with two quantities in it, and both are measurable in the same experiment. Somebody has to run it once, and after that the argument has a number in it.

What does each question actually buy you?

Every question on the form should map to a term in the score, and any question that maps to nothing comes off.

Categories of interest feed product fit, which carries the largest weight in a standard configuration. Sector feeds industry fit. A purchasing role question feeds authority fit. Those three are the load-bearing ones.

Now the questions that survive by tradition. Company size can be derived from your own account records or a company data source for most registrants, so asking is spending a completion on something you already have. Country arrives with the billing or the IP data. Job title is already there, and a seniority mapper can read it, so a separate seniority question is a duplicate. A free text box asking what the buyer hopes to achieve produces sentences a human enjoys reading and no scoring code can use.

The test to apply to any proposed question is blunt: name the sub-score it moves and the arithmetic it enters. If nobody can, the question is research, and research belongs in a post-show survey where it costs no registrations.

What should the test actually measure?

Completion rate is the wrong metric on its own, and it is the metric everyone reaches for because it is the one the registration platform reports.

Kohavi, Tang and Xu set out the discipline in Trustworthy Online Controlled Experiments in 2020, and the piece that matters here is the overall evaluation criterion: a single metric agreed before the experiment runs, capturing the trade you actually care about, computed over everyone randomised. Two rules follow directly.

First, the metric has to be joint. Count buyers who reach a usable matching state, which for most shows means at least one proposal above the threshold your concierge team uses, say 0.6. A short form that produces high completion and low match quality loses on that metric, which is the correct outcome.

Second, the denominator is arrivals. Computing your metric over people who finished the form conditions on something the treatment itself changed, and it will flatter the longer form, since the people willing to answer seven questions were more engaged before they saw either version.

Running the test on your own registration flow

Randomise on arrival at the registration page. Assign half to the three question version and half to the seven question one, and hold everything else constant, including the email sequence.

Take 6,400 arrivals, 3,200 in each arm. The short form completes 2,944, which is 92.0 per cent. The long form completes 2,720, which is 85.0 per cent, so the extra four questions cost 224 completions and seven points of completion rate. That number alone would end the argument in favour of the short form in most organisations.

Now score both cohorts and count buyers with at least one proposal above 0.6. In the short arm, 41 per cent of completions clear it, which is 1,207 buyers. In the long arm, 52 per cent clear it, which is 1,414 buyers.

Per arrival, the short form delivers 1,207 divided by 3,200, which is 37.7 per cent. The long form delivers 1,414 divided by 3,200, which is 44.2 per cent. The gap is 6.5 points in favour of the longer form.

Check it against noise. The standard error on a difference of two proportions near 0.40 with 3,200 in each arm is the square root of 0.40 times 0.60 times the sum of 1 divided by 3,200 twice, which is the square root of 0.00015, which is 0.0122. So 1.2 points, and a 6.5 point difference is more than five standard errors. That is a result you can take to the form review meeting.

Two secondary numbers are worth reporting alongside. The absolute registration loss, because a 7 point completion drop on a show targeting 40,000 registrations is 2,800 people and somebody owns that target. And the effect on the registration pacing curve, since a longer form can shift completions later without losing them, which looks like a loss until the final week.

Decide the run length before you start, and let the registration curve set it. Registration arrivals are seasonal in a way web traffic is not: the people who register in the first fortnight after launch are returning loyalists, and the people who register in the last ten days are a different population with different form patience. A test that runs for two weeks at launch measures the loyalists and will not replicate in September. Run the assignment across the whole campaign, or across at least one full early bird cycle, and report the effect for the early and late cohorts separately when the two disagree.

Not every show gets this answer. A large open-registration expo where most visitors never open the meeting programme will find the extra questions buy very little, because the denominator of interested buyers is small. Running the test is what tells you which kind of show you have.

Where the questions sit in the flow

Hosted buyers and general visitors are different populations with different tolerance, and one form for both is a compromise that serves neither.

Converve's 2026 hosted buyer guide sets out the sequence as application and qualification, then matchmaking, then preference selection, then scheduling, then execution and follow up, and the ordering is the point: qualification comes before any matching happens. A hosted buyer is applying for flights and hotel nights against a commitment to attend meetings, and in that context a longer form is proportionate. Ask the full set there, including the purchasing role question and a category picker with enough depth to discriminate.

General visitors are buying a badge, sometimes for free, and their patience is shorter. Give them the three questions that feed product fit and industry fit, and get them through.

The consequence for scoring is that two populations arrive with different attribute completeness, and the score has to handle that without punishing the visitors. A visitor missing the authority answer should sit at the population default for that term, which keeps them rankable, and the same principle applies to any term with no data behind it at a first edition in I17.

Asking the second half later

The questions that lose completions at registration cost almost nothing once somebody is inside the platform looking for meetings.

Progressive profiling works well at events because the buyer has a reason to answer. A prompt in the agenda builder saying which categories to prioritise arrives when the buyer is already thinking about their two days. A question inside the meeting request flow asking what they want from the meeting is answered because it improves the meeting they are already requesting.

Set a rule for what may be asked where. Registration carries only what the score needs to produce a first proposal. Everything else moves to a profile completion prompt with a visible benefit attached, and the prompt fires at most twice.

Measure the recovery rate. If 38 per cent of visitors who skipped the category question at registration answer it inside the agenda builder, you have recovered most of what the short form gave away, and the completion argument dissolves. If the recovery rate is 4 per cent, the prompt is in the wrong place and moving it is cheaper than changing the registration form.

How many options that category picker should hold is a related question with a different answer, and it is what I10 owns, including what happens to selection behaviour as the list grows.

Where this stops

An experiment on form length answers a question about your show, this year, with this audience mix. A sourcing show where buyers travel to meet suppliers and a technology conference where the meeting programme is a side attraction will get different results, and neither result transfers.

The deeper limit is that declared answers degrade. A buyer picks four categories in February, and by October two of them have stopped being priorities, but the score is still ranking on the February answer. Behavioural data corrects for that drift, which is why the questions matter most at a first edition and least at a show with five years of history behind the signals you already collect in I12.

There is also a limit on how much you should trust your own threshold. Picking 0.6 as the definition of a matched buyer is a convention, and if the score is not calibrated then 0.6 means whatever the last configuration change made it mean. Report the count at 0.5 and 0.7 alongside, so that the conclusion does not depend on one arbitrary cut.

This week, take your current registration form and write next to each question the sub-score it feeds. Any question with a blank next to it is a candidate for removal, and the fastest useful experiment is dropping those and adding one category question, which is a change your registration platform can ship in a day and your matchmaking score will notice immediately.