Creator Discovery Engine (RocketAPI)

Overview

Creator sourcing and content direction at Dansu were being driven by manual Instagram browsing and intuition. That created two failure modes: we didn't reliably know what formats were winning right now, and we had no scalable way to build a creator distribution network. I built a system with two distinct scored outputs to fix both.

The problem

Raw views are a weak signal for decisions. What mattered was:

• Momentum — what's rising now, not what went viral six months ago
• Repeatability — formats and creators that perform consistently, not single spikes
• Practicality — whether a creator is contactable and open to collabs

Before building, I validated the concept manually: took a small set of creators and reels, tracked basic signals by hand, and produced a first shortlist. When that output drove better briefs and better outreach outcomes than pure intuition, I automated it.

What I built

I built a single pipeline with two scored outputs, both feeding into a dashboard and CRM.

Step 1 — Creator discovery

Seeded niche clusters (festival/rave, dance, gym/running-adjacent) and expanded via similar-accounts logic. This generated a steady inflow of candidates without manual scrolling.

Step 2 — Data ingestion via RocketAPI

Pulled reel-level data (views, engagement, recency, captions, hashtags where available) and creator-level data (follower count, posting volume, bio and category, contact route signals) at scale.

Step 3 — Dual scoring

Content intelligence scoring (pattern detection)

Goal: surface formats with momentum I could copy and brief against.

Signals: engagement velocity (traction relative to age), recency weighting (prioritise what's happening now), repeatability cues (patterns appearing across multiple creators, not one-off virals).

Output: a ranked feed of reels grouped into format buckets — hooks, editing rhythms, themes, tropes — segmented by niche.

Creator network scoring (partner prioritisation)

Goal: prioritise creators who are both effective and actionable.

Signals: consistency (median and repeat performance, not just peak), engagement quality proxies (comments, saves/shares where available), brand fit (niche alignment, content style, location), practicality (contactability, signs of collab openness, speed to activate).

Output: a ranked creator shortlist with score band and documented "why them" reasons. A-tier got personal outreach, lower tiers were deprioritised.

Step 4 — Dashboard and CRM integration

Dashboard for filtering by niche, score band, recency, and location. CRM sync for execution — score, tags, contact route, "why them" rationale, outreach status, and next action. This created a closed loop: insight → shortlist → outreach → outcomes → refine scoring.

How I iterated

I adjusted signal weights when the ranking didn't correlate with outcomes. I added volatility controls to stop single-spike creators dominating, exclusion rules for bot accounts and repost farms, and niche-specific tagging logic for better segmentation. Manual spot-checks on top-ranked entries caught edge cases before they wasted outreach budget.

Systems I built

• RocketAPI integration for reel and creator data ingestion
• n8n scoring and enrichment pipeline with deduplication and error handling
• Supabase for all creator records, scores, and outreach history
• CRM sync with structured reason codes and next-action fields
• Dashboard for filtering and shortlist review

Impact

• Creator sourcing went from manual browsing to an automated, ranked shortlist
• Content briefs became grounded in observed winning formats rather than intuition
• Effort allocation improved — more time spent on high-quality, high-fit creators
• The system produced a competitive edge in organic reach through better creator selection and better-briefed content