The Capacity Forecasting Agent: A Weekly Workflow

An agent that runs every Monday, forecasts the week, and files tickets for the bottlenecks. The forecasting model, the ticketing integration, and the false-alarm rate.

Weekly cadence

Every Monday at 9 AM, the agent runs. Forecasts for the week: traffic, storage, compute. Files tickets for projected bottlenecks.

Weekly is the right cadence for most capacity planning. Daily is too noisy; monthly is too late.

The fixed cadence builds the team's habit. "Capacity tickets show up Monday" becomes part of the routine.

Forecasting model

Time-series forecast on each metric: traffic, storage growth rate, CPU utilisation. Seasonal adjustment for weekly and monthly cycles.

Confidence intervals: the forecast is a band, not a line. The width of the band is part of the output.

Out-of-distribution flags: when the recent data deviates from the historical pattern, the forecast is flagged as uncertain.

Ticket integration

When projected utilisation crosses a threshold (e.g., 80% of provisioned capacity within the week), file a ticket.

Ticket includes: the metric, the projection, the recommended action (scale up by N units), the deadline.

Tickets close when the action is taken or the projection improves. The agent does not auto-close; humans verify.

False alarm rate

Forecasts are noisy; some tickets will close without action because the projection improved.

The false-alarm rate target is < 30%. Higher means the model is too sensitive; tune the threshold.

Lower than 10% is suspicious: the model is probably missing real bottlenecks. Calibrate by checking what the team actually scaled in the last quarter.

Escalation paths

Standard ticket: filed in the team's queue.

Urgent ticket (projected breach within 48 hours): also notify on-call.

Critical projection (already breached): page immediately; a forecast that arrived too late is still useful.