How We Built the Matchmaking Algorithm Behind KreatorBoard

KreatorBoard is a swipe-based matchmaking platform for Creators and Brands. It works like a modern dating app UX, but the goal is business collaboration: campaign deals, partnerships, and long-term brand-creator relationships.

The Problem We Wanted to Solve

Most creator-brand discovery is still broken:

• Brands waste time searching random profiles.
• Creators get low-quality outreach.
• Good matches are buried under noise.

So we built a system where both sides can quickly discover high-fit profiles with a simple right-swipe and left-swipe flow.

Design Principles for Our Algorithm

We built the matching engine around four principles:

• Relevant first: show the most compatible profiles early.
• No dead ends: users should never see an empty swipe screen for too long.
• Safe and clean: blocked and unmatched users should never reappear.
• Mutual intent: a match is created only when both users express interest.

Step 1: Candidate Generation (Who can be shown)

When a user opens swipe, we fetch candidates using hard filters first:

• Not self
• Not already swiped recently
• Not already matched (for active feed)
• Not blocked or unmatched
• Match objective filters (Brand to Creator, Creator to Brand, Brand to Brand, Creator to Creator)
• Optional category filters (fashion, saas, food, gaming, etc.)

This gives us a strict pool of candidates.

Step 2: Strict to Relaxed Fallback (to avoid empty feeds)

Early-stage marketplaces face a cold-start problem. If we use only strict filters, users with niche preferences may get zero cards.

So we use a two-mode strategy:

• Strict mode: full filter conditions
• Relaxed mode: if strict results are too low, gradually widen constraints

This keeps relevance high while ensuring users always have profiles to explore.

Step 3: Ranking (Who appears first)

From the candidate pool, we compute a compatibility score. Current ranking combines profile-fit signals like:

• Role and objective alignment
• Category overlap
• Audience and influence type fit
• Reach and follower compatibility
• Profile quality and completeness
• Recency and freshness

A simple conceptual scoring form:

Score = w1(role_fit)
      + w2(category_overlap)
      + w3(audience_fit)
      + w4(scale_fit)
      + w5(freshness)

Then we show top-ranked profiles first.

Step 4: Match State Machine (What happens after swipe)

• Right swipe means "I'm interested".
• Left swipe means skip.
• If both users right-swipe each other, a match is created.
• If only one side likes, it appears as a pending like.
• Blocks and unmatches immediately remove visibility and message access.

This keeps the experience intent-driven and low-noise.

Step 5: Why This Works for Creators and Brands

For Creators:

• Better brand discovery
• Higher quality inbound opportunities
• Less random outreach

For Brands:

• Faster creator shortlisting
• Better campaign fit by category and audience
• Cleaner collaboration pipeline

Real-World Product Layer

The algorithm is tightly connected to product behavior:

• Swipe deck updates by objective filters in real time
• Pending likes and mutual matches are surfaced clearly
• Matches move to chat and collaboration flow
• Safety controls (block, report, unmatch) feed back into candidate filtering

What's Next

As user volume grows, we'll move toward a learning system that adapts ranking from live behavior (swipes, replies, conversions), while keeping hard safety and objective filters in place.

Our goal is simple: help creators and brands find the right partner faster, with less noise and more trust.