The Business Challenge
The business situation the organisation faced before the project began.
Brands and operations teams faced significant challenges in creator profiling and collaboration suitability. Manual discovery and evaluation were highly time-consuming and subjective, frequently leading to mismatched creator-brand collaborations. These suboptimal matches resulted in poor campaign reach, low audience engagement, and wasted ad spend. The team needed an automated, data-driven way to deep-profile creators based on their actual content and evaluate campaign fitment at scale before shortlisting.
Why AI Was the Right Solution
Why AI was the right approach — and what alternatives were considered.
Traditional keyword matching and static rules engines failed to capture the nuanced content philosophy, visual tone, and contextual alignment of creator content. AI—specifically Large Language Models—was essential to reason over unstructured multi-modal content (Reel video content, captions, metadata) and translate implicit brand fit into objective, quantified alignment scores at scale.
How the Solution Was Delivered
Discovery, design, development, testing and rollout — the journey, not the tooling.
- 1Discovery & stakeholder alignment with brand operations teams to define evaluation criteria
- 2Data ingestion setup to process and analyze recent Instagram Reels and metadata
- 3LLM profiling engine development for content philosophy, visual tone, and niche classification
- 4Vector search integration on Google Cloud Storage for multi-parameter semantic discovery
- 5Automated campaign fitment scoring module build and rollout
Key Technical & Architecture Decisions
Architecture, model selection, workflow and trade-offs.
• LLM & Infrastructure Selection: Chose Google Gemini and Vertex AI on Google Cloud for scalable processing of unstructured text/metadata and robust multi-parameter reasoning.
• Storage & Retrieval Strategy: Ingested enriched creator profiles into Google Cloud Storage (GCS) and enabled full-text and semantic vector search, allowing brand teams to query via flexible natural-language persona descriptors.
• Fitment Module Design: Separated baseline content classification from dynamic campaign evaluation, allowing creators to be profiled once and scored dynamically against multiple active campaign briefs.
Challenges & How They Were Solved
Obstacles hit along the way and how they were overcome.
• Nuance in Creator Style: Subjective metrics like "visual tone" or "content philosophy" produced inconsistent manual reviews. Solution: Developed a structured multi-parameter prompt schema with calibrated evaluation rubrics to ensure consistent LLM outputs across creator profiles.
• Search Precision at Scale: Raw keyword search yielded irrelevantly matched creators for complex brand briefs. Solution: Combined full-text and semantic vector search over enriched profile metadata to allow deep conceptual discovery.
User Adoption & Change Management
How users responded — training, change management and feedback loops.
The pipeline was introduced alongside existing campaign ops teams to validate LLM alignment scores against historical high-performing campaigns. By demonstrating strong correlation between automated fitment scores and actual campaign performance, brand teams gained confidence in the system, transitioning from full manual search to AI-driven shortlisting.
Business Outcomes & Impact
The measurable outcomes from the underlying AI Deployment — with the story behind them.
Significantly reduced ops creator discovery time from hours to minutes per campaign, while driving higher average reach and engagement rates on collaboration Reels through automated LLM-based fitment scoring.
Business Outcome Categories
The story behind the numbers
By replacing manual, subjective reviewer evaluations with an automated multi-parameter profiling pipeline, brand operations teams drastically accelerated their campaign setup workflows. Creator discovery and fitment checking shifted from hours per campaign down to minutes. Beyond time savings, automated alignment scoring led to higher brand-creator relevance, driving measurable improvements in average campaign reach and viewer engagement across collaboration Reels.
Lessons Learned
What surprised the team, what worked well, and what would be done differently.
- Invest early in clear taxonomy for content philosophy and niche classification.
- Combining structured metadata with semantic embeddings yields far better retrieval accuracy than semantic search alone.
- Transparent alignment scoring builds trust faster with operations teams than opaque "black-box" recommendations.
Future Opportunities
Where this solution could go next.
• Dynamic feedback loop integration to re-calibrate alignment scoring using post-campaign performance metrics automatically.
• Real-time automated negotiation and outreach integration for top-matched creators.
Professional Reflections
Guidance for another builder tackling a similar problem.
Prioritize structured profile schemas early in the development process to make LLM outputs deterministic enough for scoring and retrieval. Decoupling profile enrichment from campaign-specific scoring saved significant compute while maintaining high evaluation precision.