Three AI systems in production: data agent, course recs, acceptance predictor.
A publicly traded EdTech enterprise was slowed by manual reporting, static rankings and hand-reviewed admissions. Decisions that should have taken minutes took weeks, and the data needed to run the business sat behind analysts and dashboards. Plexe forward-deployed alongside their team and shipped three production AI systems — owned by the company — that changed how it runs day to day.
The challenge
- Execs waited weeks for analyst-built reports on enrollment and outcomes
- Universities and courses ranked by static rules and recency
- Admissions teams reviewed every application by hand
- Inconsistent applicant scoring across reviewers
What we built
- A talk-to-your-data agent: the C-suite asks questions in plain English and gets answers in seconds
- Personalized university and course recommendations for every learner
- An ML model that scores acceptance probability the moment an application lands
- Admissions teams freed to focus on edge cases, not triage
Results
- 4x faster decisions for the C-suite
- +20% enrollment uplift
Why they came to us
This is a public company operating at real scale: hundreds of thousands of learners, thousands of courses, and a constant flow of applications across multiple programs. The leadership team had no shortage of data — what they lacked was a way to act on it quickly. Every strategic question routed through a small analytics team, every ranking ran on hand-tuned rules, and every application was read by a human before a decision could be made.
The cost wasn't only speed. Static rankings buried good courses, inconsistent reviewer judgement meant similar applicants got different outcomes, and the executive team was steering a fast-moving business on reports that were already weeks old. They didn't want a dashboard or a one-off consulting engagement — they wanted production AI systems embedded in their own product, run by their own people.
Talk to your data
The first system replaced the reporting bottleneck with a conversational data agent. Executives ask questions in plain English — enrollment by region, course completion trends, the funnel from application to acceptance to enrollment — and get answers, with the underlying numbers, in seconds. What used to be a multi-week request to an analyst is now a sentence typed into a chat box.
Because the agent is grounded in the company's own data model, answers are auditable rather than guessed. The C-suite reports making decisions roughly four times faster, and the analytics team was freed from ad-hoc report requests to work on higher-value modelling.
Recommendations and acceptance scoring
The second system personalizes university and course recommendations for every learner, replacing static, recency-based rankings with a model that adapts to each person's goals and history. Better matches surface the right programs to the right learners and contributed to a roughly 20% uplift in enrollment.
The third system scores acceptance probability the moment an application lands. Admissions reviewers no longer triage every file by hand; the model handles consistent, high-volume scoring and flags the cases that genuinely need human judgement. Decisions are faster and more consistent across reviewers, and the team spends its time on edge cases instead of routine sorting.
What they own now
All three systems run in production inside the company's own stack — not as a black box, but as software the team understands and operates. Plexe's role was to build it with them and hand over the keys, so the capability compounds long after the engagement instead of creating a new dependency.