WORKLOAD DISTRIBUTION SYSTEM
A weighted case load scoring system — borrowed from RPG combat mechanics — that eliminated queue ambiguity and drove Average Speed of Answer from 20 minutes to 5.
The problem was not effort. It was ambiguity.
Frontline reps are operationally focused on resolving their existing cases and meeting Mean Time to Resolution targets. That focus is appropriate — MTTR is a core performance metric. But it meant that when a new case arrived, it landed in a queue that no one was actively watching. Not because reps were neglecting it, but because watching the queue was no one’s explicit job in that moment.
The result was a structural gap between case arrival and case pickup. That gap showed up directly in response time and SLA metrics. Initial Response SLA attainment was at 35%. Average Speed of Answer was running 20–30 minutes.
Borrow the right algorithm.
Solving the ambiguity problem required a distribution mechanism — a clear, dynamic answer to the question: whose turn is it? The inspiration came from turn-based RPGs. In combat systems of that genre, turn order is not determined by a flat rotation. It is calculated from character attributes — speed, status effects, equipped gear — that are summed into a weighted score. The character with the lightest load moves first.
Simple case count was not viable. Two reps can each own five cases while carrying dramatically different workloads. Counting only high-severity cases was not viable either — dismissing lower-severity cases would turn them into a quiet vulnerability for escalations.
Complexity in the model. Simplicity in the execution.
With the scoring model in place, the operational layer was deliberately kept simple. The Saved Search surfaces the current point totals for each rep at any given moment. At the start of each shift window, the first rep to arrive runs the search, reads the current scores, and maps the turn order from lightest to heaviest load. That sequence is posted to the dedicated Slack group chat in a single line. When a rep picks up an incoming case, they tag the next person in the chain. The tag sends a Slack notification. No manager intervention required. No ambiguity about whose turn it is.
THE TURN ORDER BOARD
Adjust a case’s weight. Watch the queue reorder in real time. Activate auto-mode to see what the system became.
What this system carried forward.
Borrow the right algorithm, not the familiar one.
Simple case count was the obvious metric. It was also wrong. The insight that unlocked the system was recognizing that case load is a multi-dimensional variable — and that a better model for calculating it already existed in an entirely different domain. Generative AI made it practical to cross that domain boundary.
Use AI to make itself unnecessary.
The scoring model required ongoing monthly recalibration as case data evolved. Rather than creating a recurring dependency on the model, GenAI was used once to produce a self-sufficient offline tool. The intelligence was distilled into an instrument. The instrument runs without the model.
Simplicity in execution protects complexity in design.
The scoring model is sophisticated. The floor mechanic — a single line in Slack, a tag, a notification — is as simple as a system can be. That asymmetry is intentional. The complexity lives where the manager can manage it. The simplicity lives where the rep has to execute under load, mid-shift, without thinking twice.