Matching

PreviousNext

Three scores for any brand and talent pair, and why a deal needs both sides to say yes

What it is

Matching gives every brand and talent pair three numbers from 0 to 100:

  • Brand wants talent: how much the brand is likely to want this person for its brief.
  • Talent wants brand: how much the talent is likely to want to work with this brand.
  • Match: the two combined. A deal needs both sides to say yes, so the weaker side holds the match back. A brand that wants a talent at 95, and a talent who wants the brand at 20, gets a match of 33, not 57.

Every score comes with a short list of plain reasons, such as "Uses the brand (they told us)" or "Asks $2,500, inside the $3,000 budget". Each score also has a confidence. Confidence shows how much of the score rests on facts we actually know. A low confidence means the score is based on thin information, and the team should read it with care.

How a brief becomes a shortlist

  1. The brand describes what it wants. Its words are read once and turned into search criteria. Each criterion is marked must, should, or must not. The criteria come from research into 89 search filters that 34 competitor platforms offer. Most things a brand asks for are "should". Only what the brand says is required becomes "must".
  2. The closest talent are found quickly. Talent profiles are compared with the brief, and the closest ones form a short list. Nobody outside that list is scored.
  3. Each talent is scored for both sides, using the reasons and confidence described above. See How the scores are built.
  4. Pairs that break a hard rule are dropped. A talent who blocked the brand is never shown, however well the rest fits.
  5. The best matches are proposed. The top few are proposed. About one slot in ten goes to a talent who probably fits but ranks lower, so newer or thinly profiled talent still get a chance.

Groups and rules that matter

  • Categories and competitors. Talent and brands are grouped into 24 categories and 30 competitor sets. Saying "I use Nike" lifts Nike strongly and the sportswear category a little. It never gives Adidas a point by name.
  • Exclusives. A live exclusive with one brand blocks its competitors, not that brand.
  • Unclaimed accounts are included. An account Zooly built from research, or one a God View simulation added, can be matched like any other. Its public page stays hidden until someone claims it, and nothing is sent to it until an admin approves or it is claimed. Internal test accounts stay out.
  • Protected traits are never used. Race, religion, disability, sexual orientation, and similar traits are never used to filter or score anyone.

Where a brand meets the matches

For every deal type except podcast, the suggestions a brand sees in its deal chat come from matching. Each one carries its score, a few plain reasons why it fits, and a price line when the talent has a minimum on file. Podcast search keeps the older path, because the matcher does not yet know how many downloads an episode gets. See Brief discovery.

Learning from deals

Every score is saved, whether or not the talent was proposed. When a deal opens for a pair the matcher proposed in the last 30 days, that match score is copied onto the deal record. The team can then compare the scores with how deals ended, to check whether higher matches really led to paid deals. See Deal record.

The team's matching page

The admin area has a Matching page with three tabs:

  • Run: pick a brand, type a brief, and optionally pick the deal type and a budget in dollars. The page shows the shortlist: each talent's match, brand wants, talent wants, confidence and reasons, with exploration picks marked. Talent a hard rule removed are listed with the rule in plain words, such as "Talent blocked this brand". The run's cost is shown, and a link opens the run in full, with every talent it scored.
  • Pair: pick one brand and one talent to see their scores and reasons. No brief is needed and it costs nothing.
  • Learning: the replay check (how often paid deals scored above the ones that died) and the report of why matched deals died. Both say plainly when there are too few deals to trust the numbers.

What it does not do yet

  • It does not change offers, payments, or deal status. Those still come from the offer itself.
  • Paid research is not switched on. Matching can ask for extra research on likely matches, but nothing is spent on it until the research step is connected. A monthly cap and an off switch are in place for when it is.
  • Brands have no page of matches to browse. A brand meets matches inside its deal chat. The team has the Matching page in admin, but there is no list page for brands or talent yet.
  • The starting weights are not tuned. They come from research. They will be adjusted by hand after about 100 matched deals, and later fitted to the results.
  • Audience details are often empty. Location, age, and interests stay blank until research or platform data fills them.

What comes next

  • Wire the research step so paid lookups fill in thin profiles.
  • A screen for brands to see their matches, and for the team to see match results and deal outcomes.
  • Podcast matching, once downloads per episode are known.

Related Documentation