Video analysis for a healthcare systems firm

This is a worked example, not a named client. The firm is a composite: a regional healthcare systems analysis practice that records workflow walkthroughs on video and turns them into specifications for developers. The workflow and the failure points are real patterns from this kind of practice. The company and the numbers are illustrative.

The problem

A senior project manager watches each client video and builds the field specification by hand: pause the video, screenshot the screen, paste into a document, annotate, repeat. Eight to ten hours per project. The practice completes 20 to 30 projects a year this way, and its growth target needs 75. The bottleneck cannot be delegated, because knowing which moments matter is the healthcare expertise the firm sells.

Two more failure points follow from the manual method. Client commentary spoken during the video gets missed when the PM's eyes are on the screenshot tool. And the developer handoff runs on the PM's prose, so developers call with clarification questions the spec never answered.

The build

A video analysis pipeline with the PM in the loop:

• It detects commentary moments: stretches of the recording where the client is talking about what is on screen, and captures them.
• It flags the screen changes and pulls contextual screenshots, so the PM reviews candidate frames instead of hunting for them.
• The PM validates and edits the output. The expertise stays in the loop; the pause-paste-annotate labor does not.
• It generates the PDF handoff in the format the developers already take.

Technical shape: AWS serverless, processing on Bedrock Claude 3.5 Sonnet, WebSocket progress updates so the PM can watch a two-hour video process in under two minutes. Deployment inside the firm's existing HIPAA-managed infrastructure.

The numbers it would have to hit

• Analysis time: 8 to 10 hours per project down to about 2. That is the whole case. At 2 hours, the 75-project target fits the same staff.
• Annual PM time on documentation: 150 to 750 hours, most of it returned.
• Client commentary: captured by the pipeline rather than missed by the screenshot method.
• Investment: $25,000 against those hours. At even the low end of the PM time range, it pays for itself inside a year.