Prepared by Scott Johnson:ResumeLinkedInsfjohnson3@gmail.com
How I Built It

One Strategist.
The Output of a Team.

I built an AI-enabled system that lets me capture, analyze, and communicate information at a massive scale, all at low cost.

The Intelligence Gap

Executives need reliable intelligence on competitors’ AI strategies

Competitive intelligence firms can do this, but at a five-figure annual cost.

AI-generated research is cheap, but it hallucinates, has limits on data volumes, and lacks access to valuable data sets.

So I built something better: a system that processes far more material, uses better data sources, keeps human judgment in control, and is designed to reduce errors.

My out-of-pocket costs were only about $30 per company. Not five figures. Thirty dollars.

Why I built this

This project demonstrates practical AI fluency by solving a real business problem at low cost. It is a work sample, not a product for sale.

Visit my portfolio page to see more of my AI work and learn about my background.

The Challenge

Millions of words. Fragmented and difficult to interpret

The process begins with a sprawling mix of hiring activity, web coverage, corporate boilerplate, professional histories, and technology data.

Raw external record

Millions of words entering the system

Job listings, web pages, public announcements, tool mentions, profile histories, and repeated fragments

Different formats
Different dates
No shared schema

AI Strategy Readout Portfolio Project by Scott Johnson
Published output

A decision-ready AI strategy readout

A strategic view built from verified evidence

How the company organizes AI
What it is building and buying
The data and infrastructure beneath it
Where AI reaches the business
Technology stack

One document
Every claim sourced
Analyst controlled

The Work

I handled the work from end to end

This was not a prompt wrapped in a website. I designed the research method, built the system, operated the AI workflow, made the analytical calls, and created the final product.

One Project, Five Roles
01

Researcher

Defined the questions and found the inexpensive data sources

02

Systems builder

Designed the low-cost operation that captures, cleans, joins, and routes the material

03

AI operator

Built the layered LLM workflow and tested its findings against primary sources

04

Analyst

Made the calls, killed weak claims, and defined the narrative

05

Product designer

Designed and built the published brief where the research becomes useful

The System

3 million words, distilled into strategic insight

I designed an end-to-end research architecture: staged processing to handle a massive amount of data, cost-efficient LLM extraction, and systems to minimize errors.

AI handles the large-scale processing, freeing me to focus on higher-value strategic work.

The Comparison

I gave ChatGPT Deep Research the exact same assignment

ChatGPT missed critical evidence and got major conclusions wrong
My system produced a materially more accurate analysis
Yet ChatGPT’s flawed report still looked credible enough for executives to act on

Advantages of My System

The Approach
Why the readout holds up under scrutiny
Proprietary research data
Uses private and public data ChatGPT cannot access
Staged processing
Handles far more information than a single Deep Research run can process
Full dataset, not a sample
Uses the full evidence base, not only what ChatGPT happens to find
Claim validation
Requires source support instead of accepting plausible LLM conclusions
Human judgment
Adds critical analysis where LLM reasoning is incomplete or misleading

What ChatGPT Got Wrong

Analysis Review
Findings from a line-by-line review of the ChatGPT report (GPT-5.6 Sol)
Missed critical evidence
Missed a foundational part of Fanatics’ technology architecture: its internally built lakehouse
Missed the Head of AI
Missed the company-wide AI/ML platform and team
Missed Fanatics’ strategic choice to remain vendor-agnostic across AI models
Missed AWS Bedrock as a core part of Fanatics’ AI infrastructure
Missed Claude Code as Fanatics’ main developer productivity tool
Missed numerous use cases across manufacturing, marketing, finance, customer service, engineering, security, Collectibles, and HR
Misinterpreted the evidence
Built a faulty tech-stack graphic from available vendor lists rather than Fanatics’ actual architecture
Mistook Fanatics-built systems for vendor capabilities, making the company look more vendor-dependent
Botched its data and infrastructure analysis by lumping vendors, products, and infrastructure together
Turned unknowns into findings
Built an incorrect org chart that did not include the actual Head of AI
Called governance a major weakness simply because it could not find public information about it
Treated the lack of visible AI patents as evidence that Fanatics lacked proprietary technology
Invented workforce estimates despite having no reliable employee count
Used confidence scores to make unsupported conclusions look credible
Gave misguided recommendations I didn’t ask for
Recommended extensive KPIs and measurement frameworks
Debated whether Fanatics should build a proprietary sports-domain LLM
Suggested a new AI council and target operating model
Built to Repeat

$30 per company

This can run again.
And again.

The asset is not only the BCG and Fanatics brief. It is a durable, low-cost way to create a verified strategic readout for any company — with the human analyst still responsible for the final word.

Read the BCG readout Read the Fanatics readout
System Powered By Python Docker LightRAG Claude Cowork Google Colab Replit Exa