Shark Tank Research Project · Status & Plan

Every pitch in the Tank, funded or not.

Building the first complete, searchable dataset of Shark Tank pitches, each one matched to its later update segment, to model what actually makes a pitch get funded.

Phase 2 in progress: completing the transcript set tonight
130 / 359
Episode transcripts secured today (36%)
16
Seasons in scope. Every pitch, not a sample
End of July
Student deadline for all-season data entry
~2 hrs
To transcribe the remaining gap locally
01

The idea

Dr. Barnes's research concept. Jay is bootstrapping the technical setup, then handing off.

The question

What makes a pitch get funded?

Estimate the probability of funding from measurable factors in each pitch, grounded in an established persuasion framework, the elaboration likelihood model.

The novel part

Match the updates to the pitches

Existing research studies the pitches. Nobody has linked the later update segments back to the original pitch. That connection is this project's contribution.

The mandate

Every season, no exceptions

The student agreed on one condition: cover the entire run of the show, not a sample. That condition sets the scope for everything below.

Faculty lead
Dr. James Barnes
Designs the research and the model
Mississippi State University · owns the study
Researcher
The student
Enters pitch data for every season
4 credit hours · ~10 weeks · full-time job
Technical setup
Jay
Bootstraps transcripts, data flow, model
Hands off once the pipeline runs
What gets delivered

1 · Statistical dataset

Every pitch coded into the variables the model needs, ready for analysis.

2 · Per-pitch graphics

A visual summary for each pitch, so any single deal is easy to read at a glance.

3 · Searchable database

The whole set, browsable by people learning how startups get funded.

02

Transcript coverage

Full episode transcripts are hard to source. Here is exactly what we have and how we close the rest.

130 of 359 episodes

Secured today from a transcript source the research agents surfaced, organized by season with a manifest.

229 episodes remain. The bulk is one continuous block, seasons 4 through 11, which no public transcript source covers.

In hand (130) To transcribe (229)
Coverage by season
Complete Partial Missing
The gap is seasons 4 through 11 (2012 to 2020, roughly 208 episodes). Six independent source checks confirmed every public transcript and subtitle library draws from the same small pool, which skips these seasons entirely. Seasons 12 through 16 are complete, seasons 1 through 3 are partial.

Path to a complete set, tonight

Verify first
A

Check netraptor

One source lists subtitle packs for the exact gap seasons (5, 6, 8, 10, 11). It was offline today, so verify one episode is real before trusting it.

Reliable
B

Transcribe locally

Run Whisper on Jay's machine for whatever stays missing. It is the only method that guarantees the gap closes, and it can add speaker labels. About two hours of local compute.

Then
C

Index and match

Line every transcript up against the master episode list so each one maps cleanly to its season, episode, and pitches.

03

Data and model pipeline

Simple and free by design, so the student can move fast and James can maintain it.

Built with Claude
The instrument
Defines every variable to capture from a pitch.
→
Free, simple
Google Form
The student's entry point, one submission per pitch.
→
Auto-collected
Response sheet
Every entry lands in one growing spreadsheet.
→
Elaboration likelihood
Statistical model
Estimates probability of funding. Structure already built from a sample.
Next capability
Model updates itself
Each new pitch the student adds recalculates the model automatically, no manual rerun.
Next capability
Outlier alerts
Flag any entry that falls far outside the model, to catch data errors and surprising deals early.
04

Roadmap

Where we are today and the sequence from here to a living, searchable dataset.

Source and secure transcripts

TodayDone

130 of 359 episodes in hand, organized by season with a manifest. A full map of what exists, what is partial, and what is missing.

Owner: Jay

Complete the transcript set

TonightIn progress

Verify the netraptor source, then transcribe the seasons 4 through 11 gap locally with Whisper. Goal is a complete set of all seasons by tonight.

Owner: Jay

Enter every pitch

By end of JulyNext

The student captures the data for every pitch across all sixteen seasons through the Google Form. A sample is already in and was used to build the model structure.

Owner: The student

Match updates to pitches

Research phaseNext

Link each update segment back to its original pitch. This is the connection no prior research has made, and the heart of the study.

Owner: Dr. Barnes

Live model and conversational search

Longer termLater

The model updates as entries arrive and flags outliers. Then load the pitches into a vector database, the same approach as the sermon-search build, so learners can explore and ask questions in plain language.

Owner: Jay sets up, Dr. Barnes runs

Where this goes

A searchable library of every Shark Tank pitch, so anyone learning to build a startup can explore what gets funded, and why.