Blog
Teaching as an Act of Love
By Teasha Cable
How one university cut accreditation reporting from weeks to hours, and built a decision intelligence system that tells faculty which of their students need a call today.
Dr. Bernadette Howlett cried the first time she saw the results of her own work. That's not a sentence you expect in a conversation about predictive models and dashboards.
Howlett, (known to everyone at Western Governors University as Dr. B), is the Operations Analytics and Experiential Product Director.
Dr. B helped the university cut accreditation reporting from weeks to hours, and build a decision intelligence system that tells faculty, in real time, which of their hundreds of students need a call today.
When she talked about the data showing the system had actually closed outcome gaps for students, many who started with more obstacles than other students, her voice changed.
I've never seen anything else do that before.
It was the first hint that this conversation was going to be less about dashboards and more about why she builds them at all.
Twelve Million Decisions, One Person at a Time
Howlett came to this work with an engineer's brain, and she borrowed a framework most people would never think to apply to a university: the decision engineering used on factory floors (don't think widgets).
WGU was making essentially the same decision, whether to reach out to a struggling student, and how, 12 million times a year.
Any decision made that often deserves the same rigor a manufacturer gives a production line. But she drew a hard line where it mattered. "Humans aren't widgets," she said, and she meant it.
What she borrowed from manufacturing wasn't a way of thinking about people. It was a way of thinking about decisions:
- Know the levers
- Know the settings
- Make the process faster and clearer for the human pulling them
That human is the instructor staring down 500 students, trying to figure out who needs help and who's fine on their own...
The Model Writes the Message. The Human Sends It.
You've heard of FOMO...FOM, is the Faculty Outreach Message Engine Howlett's team built. FOM drafts a personalized outreach message using social and emotional learning and something called social norming, essentially telling a student that "others just like them have succeeded from this exact spot," then pointing them toward what to do next.
Here's the part that matters: that message never goes straight to a student. It lands on the faculty member's dashboard first. They can send it as is, rewrite it in their own voice, or scrap it. "A machine can't do that," Howlett said about the human warmth teaching requires.
I think teaching is an act of love.
She wasn't being sentimental for effect. It came up again and again, unprompted, as the actual reason any of this exists.
Getting it Wrong on Purpose, in the Right Direction
Every predictive model fails somewhere, and the real work is choosing which failure hurts less.
Howlett's team leans toward avoiding false positives, meaning they'd rather a model undersell a student than oversell one.
She traced that choice back to a blunt moment at a meeting with employers earlier this year, when one stood up and told a room full of higher ed leaders, "we don't trust you," because graduates were showing up without the writing, math, and teamwork skills their degrees were supposed to guarantee.
"I didn't hear one employer complain the other direction," Howlett said.
Nobody's ever upset that a new hire turned out to be more capable than expected The Tool Nobody Wanted, and What It Taught Her Before the current system existed, WGU tried a simpler version: a service level agreement dashboard meant to gently encourage faculty to act quickly on recommendations. It came from good intentions and a real data-backed reason.
It took three days to become "the enemy," in Howlett's words, no matter what disclaimer sat at the top of the screen.
None of us want to be micromanaged. It's like someone hovering over you while you load the dishwasher.
So when the team rebuilt the system around decision intelligence, they left that scoreboard out entirely. It's a small choice that says a lot:
A tool built to help people has to account for how it feels to use, not just what it measures.
The Takeaway
Howlett's closing advice cuts against the current: resist the pressure to adopt AI just because everyone else is doing it.
Fall in love with the actual problem first. Sometimes the answer is a new policy, not a new model. Decision intelligence, done well, isn't about replacing the people closest to the work. It's about giving them better information faster, so the part only a human can do, the caring part, has room to happen.
Listen to my conversation with Dr. B wherever you get your podcasts.