AI Showcase: Rebuilding a Flipped Course with an AI Avatar

Instructor
Brian Mayhew, Ph.D., Associate Dean for Online Innovation, Wisconsin School of Business, and Professor, Accounting and Information Systems
Course
Accounting and Information Systems 630: Foundations of Auditing
Application
I teach this as a “flipped” course, where core lecture material is delivered through video. The videos carry the foundational material, roughly 70 to 80 minutes of lecture a week, so class time can shift toward application: Socratic questioning, working problems, and discussion. We can get in and get actively learning, and I can hear directly from students about what’s landing and what needs more work.
I recorded the original videos at home in 2021 and they were due for an update. I don’t look like that anymore, the background isn’t great, the sound isn’t great. More significantly, the textbook and some accounting standards have changed. I decided this year to rebuild all of the video content – roughly 12 modules with seven to eight lectures each. That was looking like a heavy lift – more than 80 hours of video!
Working with Billy Kardasz and Derek Schneider on the WSB Digital Media team, we created a new process:
- Starting with an original 2021 recording, I converted the video to captions and ran those through Google Gemini to generate a transcript
- I reviewed and updated the transcript, correcting errors, tightening language,and adding current examples and revised accounting standards. This becomes the new script.
- We created an AI avatar, built from my likeness and voice, using a subscription version of HeyGen. The avatar uses the script to deliver each video’s intro and outro; the slide content in between runs on AI-generated voiceover.
- Kardasz and Schneider helped assemble the pieces in Camtasia and published them through Kaltura into Canvas.
Results
The pedagogical benefits of this flipped approach have been clear since 2021 – we are able to do much more application. The video also allows students to review concepts at their own pace – on discussion boards, I see students referencing specific lecture timestamps to answer each other’s questions. And Kaltura’s quiz integration lets me embed low-stakes checkpoints directly in the video, so students can engage right then and there.
The most concrete win with this new process is turnaround time. While the time to update the course from the original video to the AI version was significant, it was much less than re-shooting everything. And because the whole pipeline runs on an editable transcript rather than a studio recording, updates are fast. Swapping in a current financial statement or reflecting a new accounting standard no longer requires re-shooting anything, just updating the transcript and regenerating. It’s a relatively light lift for me. I’ve also cleaned up my delivery, trimming filler language like my habit of saying “so.”
It’s so easy to change, you’re tempted to change it all the time – kind of like when we shifted decades ago from typewriters to word processes. My rule now is to know when a video is good enough and move on.
However, dividing up the technical tasks is still a work in progress. Producing at this pace is taking up the bulk of the A/V team’s bandwidth, and that’s with me doing some of the technical assembly. Scaling this beyond one course will depend on additional support, likely student labor, for the Camtasia and transcript-cleanup work that doesn’t require instructor-level judgment.
I didn’t set out to become a guinea pig or a trailblazer or whatever it is I am, but I’m happy to do it.