Paramount

GLOBAL CONTENTS DISTRIBUTION RESEARCH INTERNSHIP

May 2025 - Aug 2025

AI prompting & Agent design · Statistical analysis · Data visualization · Workflow automation

TL;DR

Paramount / Global Content Distribution Research team. An AI agent and audience-profiling pipeline for content distribution across 25 territories.

65% time reduction on pre-release research, 150+ sales decks produced, 25 territories analyzed weekly, 20+ global offices supported.

THE PROBLEM

Global Content Distribution Research (GCDR) at Paramount sits at the intersection of data and dealmaking. The team supports international and domestic content licensing across 20+ global sales offices, translating viewership data from linear TV and streaming platforms into insights that inform licensing negotiations, marketing campaigns, and Home Entertainment strategy.

GCDR operates on compressed timelines. When a show gains momentum, sales teams need performance data immediately to capitalize on international buyer interest. But research workflows at GCDR were manual, repetitive, and slow: pre-release performance sheets required 2-3 hours of manual research per title. The team needed infrastructure to surface insights faster without sacrificing quality.

Goals

  • Automate repetitive research workflows to reduce time spent on manual data retrieval and formatting.
  • Analyze weekly viewership trends across 150+ titles in 25 territories to support licensing negotiations.
  • Build data-backed audience profiles for Home Entertainment to inform targeting and marketing strategy.
  • Translate complex performance metrics into compelling visual narratives for sales teams in 20+ global offices.

THE SOLUTION

AI Automation of Pre-Release Performance Sheets

Before a title goes to market, the research team creates “pre-release performance sheets” (PRPS): standardized documents with cast details, IP background, comparable titles, and historical performance data. Creating one sheet manually took 2-3 hours of research, formatting, and quality control.

I built a Microsoft CoPilot Agent to automate the most time-consuming parts:

  1. 01

    Scoping the Automation

    Rather than trying to automate everything, I identified what AI could reliably handle:

    • Retrieving actor filmographies from internal databases + IMDb
    • Pulling IP history (book series, franchise timelines)
    • Formatting data according to PRPC template
  2. 02

    Agent Design

    • Configured CoPilot Agent to query internal Paramount databases first (higher accuracy)
    • Built fallback logic: if internal data missing → search credible external sources (IMDb, Wikipedia)
    • Created formatting rules matching existing PRPC standards (avoiding team retraining)
  3. 03

    Iteration & Refinement

    • Tested on 10 recent titles, compared AI output to manually created sheets
    • Identified hallucination patterns (AI inventing award nominations that didn’t exist)
    • Added verification layer: Agent flags any claim it’s <80% confident about for human review

Tools: Microsoft CoPilot Agent, internal Paramount API, prompting strategies.

Key Skill: Understanding AI limitations: knowing when to automate vs. when human judgment is essential.

Iteration cycles: 15+ prompt refinements to reduce hallucinations and improve formatting consistency.

Home Entertainment Audience Profiling

Paramount’s Home Entertainment division needed to understand who buys digital content (movies/shows for download or rental). Marketing assumptions existed, but no data-backed audience profile to guide targeting strategy or promotional spend.

  1. 01

    Research Design

    Using GWI (Global Web Index), I designed queries to compare digital transactors against general population across:

    • Demographics (age, income, education, location)
    • Media consumption habits (streaming platform usage, genre preferences)
    • Psychographics (early adopter tendencies, media spending patterns)
  2. 02

    Data Analysis

    For each variable, I ran:

    • Correlation analysis: Which traits most strongly predicted digital transaction behavior?
    • Statistical significance testing: Which patterns were real vs. noise?
    • Hypothesis testing: Validated or challenged existing marketing assumptions
  3. 03

    Synthesis & Presentation

    • Built database organizing all findings by demographic category
    • Created presentation deck translating statistical insights into actionable marketing recommendations
    • Presented findings to Home Entertainment team (my first stakeholder presentation at Paramount)

Tools: GWI platform, Excel (correlation matrices, significance testing), PowerPoint (data visualization).

Statistical Methods: Correlation analysis, chi-square tests for categorical variables, confidence interval calculations.

THE RESULT