Why EventIntro asks the same five questions every time

Why five questions, not ten?

Five is the band that balances completion rate with signal density. Longer surveys see steep drop-off; shorter ones don't give the LLM enough material to extract reliable seek/offer pairs. Five questions land in the sweet spot — most attendees finish, and the matching pipeline has enough text to work with.

Survey-design folklore is broadly consistent on this: every additional required question past about seven causes a measurable bump in abandonment, and below about three the platform doesn't have enough to differentiate matches. We landed on five after iterating in private testing; the choice is intentional rather than arbitrary, and shrinking or expanding it would change the matching surface materially.

What does each question extract?

Each of the five questions targets a specific layer of the attendee profile that the matching pipeline needs. In order: current projects (so the LLM can place you in time), offerings (what you can give to others), needs (what you're looking for), professional interests (broader topical signal), and a free-form prompt for anything else. Together they yield seek/offer keyword pairs the embedding layer can search.

1. What are you working on? — Anchors the profile in the present. The LLM extracts current focus, recent moves, and project context that ages quickly and would be wrong if the profile were six months stale.
2. What could you help others with? — The offer side of the seek/offer pair. We deliberately ask "could you help with" rather than "what skills do you have" — it produces specific, actionable phrases instead of resume keywords.
3. What would you like help with? — The seek side. The matching pipeline's complementary-fit search depends on this being honest; the most common failure mode is members leaving it vague to avoid looking inexperienced.
4. What interests you professionally? — Broader topical signal that catches near-future relevance the first three questions miss. Useful for spotting members who'd connect well even when their offers and needs don't directly match.
5. Anything else to share? — Free-form. People use it for personality, unique experiences, or specific connection goals the structured questions don't surface.

Why are the questions fixed?

Match quality depends on the consistency of the signal the pipeline operates on. The LLM enrichment, the seek/offer keyword extraction, and the embedding-space layout are all tuned for the specific shape of these five questions. Letting each host write their own would introduce signal drift, degrade cross-cohort match quality, and lose the transferability that lets a member's profile work in more than one cohort.

The reasoning, in plain terms:

• Signal consistency drives match quality. The pipeline learns what good seek/offer pairs look like from the canonical questions. Ad-hoc questions vary in framing, specificity, and emotional load — and the variance shows up as worse matches.
• Profile transferability requires shape stability. A member who joins an alumni cohort and then later joins a mastermind cohort keeps their profile. That only works if both cohorts collected the same shape of signal. Custom questions break the carry-forward.
• Question design is harder than it looks. "Ask people what they offer" sounds easy. Writing the question so introverts and extroverts answer with comparable density, so the LLM can reliably extract keywords from either, and so the answer survives translation between industries — that's the design work the canonical five questions encode. It isn't work hosts should have to redo every cohort.

We believe there's a future version of EventIntro where hosts can append optional structured fields — industry tags, time zone, employer size — without disturbing the matching-signal questions. But host-rewritten replacements for the seek/offer questions themselves are not on the roadmap; the quality cost is too high.

What can hosts customise?

Hosts customise the things that don't affect match quality: cohort name and description, banner image, invitation email template, RSVP deadlines and reminder cadence, breakout-group size, matching priority (complementary vs. similar), and dynamic-regrouping settings. The survey questions themselves are the same across every cohort by design.

For the full host-side configuration walk-through, see The 15-Minute Event Setup. For the FAQ-form summary of the customisation question, see the FAQ.

How does EventIntro use the answers?

EventIntro processes each set of answers through an LLM that expands them into a richer profile and extracts seek/offer keyword pairs. Those pairs are converted to vector embeddings and stored in a pgvector index. Matchmaking then queries the index for each attendee's complementary fits — people whose offerings answer this attendee's needs, and vice versa.

For a fuller pipeline overview, see How EventIntro Works. The fixed-question shape is what makes that pipeline reliable across diverse cohorts; custom questions would force the pipeline to work harder for less consistent signal.