The Playbook: Case Studies from an AI-Run Business
Most AI advice is theory. This page is receipts. BigPlans Media runs on the systems documented below — built with Claude, in production, doing real work every day. Each case study follows the same honest format: the problem, what we built, what happened, and what we’d tell a client attempting the same thing. Written by Dr. Eva Wolf, DBA — Claude Certified Architect (Anthropic), marketing professor, and founder of BigPlans Media.
Case Study 01 — The Research Radar: An Automated Research-to-Podcast Pipeline
The problem. Hundreds of AI and marketing research papers publish every day. Marketers need the handful that matter, translated into plain business language — but no solo operator can read, screen, script, record, and publish daily. Content teams solve this with headcount. We wanted to solve it with architecture.
What we built. An automated research-to-publishing pipeline built with Claude Code and GitHub Actions. Each day it ingests newly published research, screens the pool against a version-controlled editorial policy (EDITORIAL.md — selection rules like “empirical strength first” and “verify before featuring” live in the repo, and Claude applies them), scripts the episode with per-paper structure (business question → finding → evidence check → practical action), generates the voice track, and publishes the episode, show notes, and transcript across Apple Podcasts, Spotify, Amazon Music, and BigPlans.media.
The result. 37+ published episodes and counting, produced daily with a fraction of the hands-on time a traditional production would need. A recent episode screened 249 papers and featured the three that cleared the evidence bar. The show now runs a hybrid format: Dr. Wolf’s recorded framing around segments voiced by “Evita,” an AI research analyst trained on her screening methodology — with every episode disclosing exactly that.
What we’d tell a client. The screening prompt is not the hard part — the editorial policy is. Codify your judgment in a file the AI must obey, version it like code, and update it from your own review notes weekly. Automation without codified taste produces content nobody should publish.
Built with: Claude Code · GitHub Actions · MCP · ElevenLabs
Case Study 02 — AI-Assisted Grading: A Custom Claude Skill for Canvas
The problem. College professors spend a large share of grading time on mechanical work — navigating the LMS, applying the rubric consistently, and drafting individualized feedback — which competes directly with the part that matters: coaching students. Dr. Wolf teaches marketing at three institutions and needed rubric-consistent, personally voiced feedback at scale, without handing grading decisions to an AI.
What we built. Two connected tools. A Chrome extension that assists grading workflows inside Canvas, and a custom Claude Skill — a reusable instruction package — that operates in Canvas SpeedGrader: it reads a submission, applies Dr. Wolf’s rubric, and drafts warm, specific coaching comments in her voice. By design, it never submits: every rubric line and comment is reviewed and finalized by the professor. The AI drafts; the human decides.
The result. A repeatable grading assistant that keeps feedback consistent across large sections while preserving full human judgment on every grade — and a template pattern (one Skill per course module) now being replicated across courses.
What we’d tell a client. The design principle transfers to any high-stakes review workflow: let AI do the reading, structuring, and drafting; reserve submission authority for the human. “Draft everything, decide nothing” is the adoption pattern that survives compliance review — in education and everywhere else.
Built with: Claude Skills · Claude in Chrome · Canvas LMS
Case Study 03 — Notes That Teach the Machine: An MCP Editorial Feedback Loop
The problem. Every content operator gives their AI the same feedback twice — because reactions captured in the moment (in a notes app, on a walk, after listening to an episode) never make it back into the system that generates the content. The feedback evaporates; the AI never improves.
What we built. A closed editorial loop using the Model Context Protocol (MCP). Dr. Wolf reviews episodes and captures reactions in Granola, her AI meeting-notes app. Granola’s MCP server connects those notes directly to Claude — no exporting, no pasting. On a weekly cycle, Claude pulls the notes, sorts them by a simple convention (dated notes = newsworthy findings queued for the Friday live show; undated notes = workflow feedback), proposes edits to the pipeline’s editorial policy file, and files technical improvements as GitHub issues. Dr. Wolf approves the diffs; the pipeline inherits her taste.
The result. Listening notes now update the production system that makes the next episode — the podcast’s editorial policy is written by its own review process. The same loop surfaced the show’s format redesign and its current model-upgrade evaluation.
What we’d tell a client. MCP is the difference between “AI you talk to” and “AI that’s plugged into your tools.” Start with one read-only connection to a system you already use daily — the leverage arrives when your existing habits start feeding your automations for free.
Built with: MCP · Granola · Claude Code · GitHub
More builds in progress
This page grows as the work ships. Currently in build: a comment-to-DM automation app (official Meta Graph API), a hybrid-voice production upgrade to the Radar pipeline, and a model A/B evaluation framework. Each becomes a case study here when it’s live and measured — tested, not assumed.
Work with us. BigPlans Media designs, builds, and teaches Claude-powered systems like these for businesses and teams — AI adoption strategy, managed content engines, AI visibility, and hands-on training. Get in touch.
About the author. Dr. Eva Wolf, DBA (Florida International University) is a Claude Certified Architect (Anthropic), marketing professor at three colleges, and founder of BigPlans Media. Her doctoral research tested AI-generated personality-targeted advertising with 394 B2B decision-makers — read the research.
Ready to apply these methods to how buyers find you? See SEO + AI Search services.