A national entrepreneur conference

The annual national conference of a global peer network for business owners, held at a resort abroad for a couple of hundred attendees. The chair wanted every attendee to leave having met the one person they most needed to meet, rather than leaving it to the alphabetical seating chart. North Group built a custom AI model that read every attendee's survey and paired them, from the small-group sessions to the breakfast tables, and the personal readout that got attendees to go deep on that survey.

The challenge: A long survey, every question required, often skipped

Curating a conference for a couple of hundred people means knowing each of them well: what they build, what they are carrying, what they can give and what they need. That meant asking busy entrepreneurs to complete a long, personal survey with every question required, the kind that usually gets a few minutes of attention and a lot of skipped answers.

The survey had to be worth the time. A generic thank-you would not do it. The answer was a personal readout, written from each person's own words within minutes of submitting, as the reward for going deep, and the raw material the curation engine would work from.

The solution: Eighty questions, a custom model, readouts within minutes

A survey of about eighty questions in thirteen sections: business, stage of life, what you give, what you need, personality and archetype, interests, values, and why you came. Every question required. Two consent questions at the end decide whether a readout is generated and whether open-ended answers can be used to write it.

A custom AI model, built for this conference, read every submission: what each person gives, what they need, their stage of life, their values, and how they see themselves against how others see them. Weighted by three pairing criteria the chair set, it assigned the small-group sessions, named each attendee's one person to meet, set the breakfast tables, and matched mentors to the next generation of members, with guardrails to keep existing groups and couples apart and to mix chapters and tenure.

The same data produced a personal readout for every attendee who asked for one, written by an AI agent from a craft guide: quote the person's own language, lead with the gap between self-image and reputation, pull what they are working through into one thread, and say what the pairing model was reading on them. Around eight hundred words, delivered as an email with a designed PDF within minutes of submitting.

Every pairing was revealed on site: sealed envelopes at the small-group session, a card with one name on it, a table assignment at breakfast. Nobody was placed by the alphabet.

The outcomes: Nine in ten finished, a hundred readouts sent

Nine in ten attendees completed a survey that took them thirty-seven minutes on average, against the four to seven a typical automated survey gets, and a hundred personal readouts went out within about fifteen minutes of submission.

Every pairing at the conference came out of one model reading one dataset, and attendees opened an envelope to find a name chosen for them from their own answers. The chair's follow-up emails used the readouts as the reason to finish the survey. The work led directly to a program with one of the network's city chapters and to conversations with the team planning the following year's conference.