Last January, we launched an internal AI agent at illuxi that we call Leroy.
The goal was simple: respond faster, better, and at scale to our learners’ requests.
With close to 290,000 learners on the platform over the past year, one thing is clear: support requests are not decreasing—they are structurally increasing.
Industry benchmarks are well known:
A high-performing SaaS platform typically keeps support volume below 10% of active users per month (Fullview)
Around 8% of ARR is spent on support and customer success (SaaS Capital)
The real question wasn’t “should we automate?” The real question was: how do we do it without degrading the customer experience?
The Real Challenge: It’s Not AI, It’s Data
An AI agent is not intelligent by default.
It’s only as good as what you feed it.
Since November, the real work has been: structuring our answers, documenting frequent cases, standardizing processes, and continuously enriching our knowledge base.
Leroy isn’t “smart.” He’s well trained.
The True Cost of Support
Take a simple example: “I forgot my password.”
It takes only a few minutes to answer.
But the real cost lies in the interruption.
A study by Gloria Mark (University of California) shows it takes on average 23 minutes to regain focus after an interruption.
Internally, we use a conservative assumption of 15 minutes per request.
At a fully loaded cost of $75/hour, that’s $18.75 per 15-minute block.
Our Reality After One Month of Optimization (February 2026)
We handled 810 support requests.
Without AI (100% human model): 810 × 15 minutes = 202.5 hours, 810 × $18.75 = $15,187, for an estimated total of ~$15,200.
And that’s theoretical. In reality, no employee is 100% productive—interruptions, meetings, and cognitive load increase the real cost.
Cost with Leroy + 1 Human
Breakdown: 421 requests handled by Leroy, 389 escalated to a human.
AI portion: $345 USD (support software) + $370 USD (AI usage) ≈ $950 CAD.
Human portion: 97.25 hours at $25/hour (outsourced) ≈ $2,500 CAD.
Total Cost with AI
≈ $3,450 CAD
Results
Cost without AI: ~ $15,200
Cost with AI: ~ $3,450
Savings: over $11,700 in a single month (February 2026)
And that doesn’t include 24/7 availability, faster response times, reduced friction, and freed-up time for the team.
Leroy, in Practice
This wasn’t a complex project.
No massive budget, no oversized team, no endless timeline.
Just a real need, existing data, and disciplined execution.
The Real Question
It’s not “how do we do AI?”
It’s: where are your simplest operational frictions—and how do you eliminate them?
What’s Next
Leroy is just the beginning.
Next steps: enrich the knowledge base, improve resolution rates, optimize human escalation, and expand use cases.
The Goal
Not to replace humans, but to free up time where it creates the most value.
Conclusion
Applied AI is not a promise.
It’s a concrete, measurable, and accessible lever.
Leroy is a simple example—and most importantly, a repeatable one.









