From CSV to Decisions: How the Evaluation Pipeline Works

Understanding how FounderScan processes your applications helps you get the most out of the platform. Let's walk through the complete pipeline from CSV upload to scored results.

The Five-Stage Pipeline

FounderScan processes applications through five distinct stages, each designed to add intelligence and context to your evaluation process.

Stage 1: Data Ingestion

When you upload a CSV file, FounderScan's parser goes to work:

Automatic field detection: The system identifies common fields like company name, website, founder names, LinkedIn URLs, industry, stage, and custom application questions.

Data normalization: Names are standardized, URLs are validated, and duplicate entries are flagged.

Startup creation: Each row becomes a startup record with structured metadata and the full application text preserved.

Pro tip: Include LinkedIn URLs for founders whenever possible. This dramatically improves the enrichment quality in the next stage.

Stage 2: Founder Enrichment

For each startup, FounderScan enriches founder profiles using our data enrichment layer:

Professional history: Previous companies, roles, and tenure
Education background: Degrees, institutions, and fields of study
Network indicators: Company connections, shared experiences
Public presence: Articles, speaking engagements, open source contributions

This enrichment happens automatically and typically completes within seconds per founder.

Stage 3: News & Context Gathering

FounderScan searches for recent, relevant news about each startup and its founders:

Company news: Funding announcements, product launches, partnerships
Founder mentions: Interviews, awards, industry recognition
Industry context: Market trends affecting the startup's space

This real-time intelligence ensures evaluations are based on current information, not just what founders chose to share in their application.

Stage 4: AI Evaluation

Here's where the core analysis happens. For each startup, FounderScan's AI:

Analyzes against each criterion:

Examines all available data (application, enrichment, news)
Scores the criterion on a 1–10 scale
Provides detailed reasoning citing specific evidence
Flags confidence level and any data gaps

Generates overall assessment:

Weighted score based on required vs. nice-to-have criteria
Highlight reel of strengths
Areas of concern or uncertainty
Recommended next steps

The AI uses chain-of-thought reasoning to ensure transparent, explainable outputs.

Stage 5: Results & Ranking

Finally, FounderScan organizes results for easy review:

Ranked list: Startups sorted by overall score
Filtering: By score ranges, specific criteria, or tags
Comparison view: Side-by-side criterion breakdowns
Export options: PDF reports, CSV data, or API access

Real-Time Progress Tracking

The entire pipeline runs with live progress updates:

[2/5] Enriching founders... 
      ✓ Sarah Chen - CTO (LinkedIn enriched)
      ✓ Michael Torres - CEO (LinkedIn enriched)
[3/5] Gathering news...
      Found 3 recent articles
[4/5] Running AI evaluation...
      Criterion 1/5: Technical Team ✓
      Criterion 2/5: Market Size ✓
      ...

You can watch evaluations complete in real-time or come back later to review finished results.

Pipeline Customization

Several aspects of the pipeline can be tuned:

Criteria weighting: Required criteria count 2x toward the overall score by default. You can adjust this multiplier.

News recency: By default, we fetch news from the last 6 months. For fast-moving industries, you might prefer 3 months.

Enrichment depth: Basic enrichment runs quickly; deep enrichment pulls more data but takes longer.

Evaluation focus: Tell the AI what matters most to your program (technical depth, market validation, team dynamics) to shape its analysis.

Handling Edge Cases

The pipeline is designed to handle imperfect data gracefully:

Missing LinkedIn URLs: Enrichment proceeds with available data; the AI notes the gap
Non-English applications: Support for major languages with translation where needed
Incomplete responses: The AI identifies what's missing and adjusts confidence accordingly
Duplicate submissions: Flagged for manual review rather than processed twice

What Happens to Your Data

Security and privacy matter, especially when handling competitive startup information:

Encryption: All data encrypted in transit and at rest
Isolation: Each organization's data is logically separated
Retention: You control how long data is retained
No training: Your application data is never used to train AI models

Example: A Real Evaluation

Let's trace a single startup through the pipeline:

Input: "TechFlow AI" application with two founders

After Enrichment:

CEO: 8 years enterprise SaaS, Stanford CS, previously at Stripe
CTO: 12 years ML/AI, MIT PhD, previously at Google Brain

News Found:

"TechFlow AI raises $2M pre-seed from Sequoia scouts" (2 weeks ago)
"TechFlow wins TechCrunch Disrupt SF" (1 month ago)

Criterion Scores:

Technical Team: 9/10 ("Exceptional engineering background with direct domain expertise")
Market Opportunity: 8/10 ("Large TAM in enterprise AI, early but growing fast")
Traction: 7/10 ("Strong signals but limited revenue data provided")

Overall: 8.5/10 with recommendation for partner interview.

Ready to see this in action? Schedule a demo and we'll walk you through your first batch.