Appearance
Capacity forecasting
Stop being told the disk is full. Be told when it will fill.
Every check on every asset gets a projection. Disk usage at 28% today, growing 1.1% a day, reaches the warn threshold in 27 days. Ticket the customer this morning. Schedule the upgrade on a Tuesday in three weeks. Don't drive out at 11pm on a Friday because someone hit save once too often.
- Per-asset, per-metric
One projection per agent per check. Disk, memory, CPU, and anything else with a numeric warn threshold. No fleet-level averages that hide the one box about to fall over.
- Confidence-scored
Every projection carries a high/medium/low confidence label and the sample window it was computed over. Low-confidence noise is hidden from the operator view by default.
- Settle-in aware
A newly imaged Windows box grows fast because Windows Update is catching up. We label those projections as settling, and tighten them as the machine matures. Honest about its blind spots.
What we project
Anything with a numeric value and a threshold.
- Disk usage approaching the configured warn threshold
- Memory pressure trending up over weeks
- CPU load creeping during business hours
- Any custom check with a numeric value and warn_threshold set
How the prediction is made
A regression line over recent steady-state telemetry, not a guess.
- Daily aggregates roll up the raw check stream into stable per-day numbers
- A regression line is fit over the most recent window of daily values
- The threshold-crossing date is computed analytically from slope and intercept
- Confidence reflects how tightly the data hugs the line. Noise drops to "low" and gets hidden
What we are honest about
Forecasts have failure modes. We name them.
- Brand-new machines (under fourteen days) get no forecast at all. The data is install-settling, not steady-state, and projecting from it would be misleading
- Machines fourteen to twenty-nine days old get a "settling" label so the operator knows to interpret the projection loosely
- Mature machines (thirty days and up) regress only over the latest two weeks, picking up "now-trend" rather than mixing in settle-in
- Step-changes (disk replaced, large app uninstalled) and seasonal cycles are not yet handled. Coming.
Where it shows up
On the asset detail page today, alerts and reports next.
- Capacity Outlook panel on every asset detail page
- Forecast-threshold alert rule type (coming): auto-ticket when a projection crosses below N days at acceptable confidence
- Fleet outlook report (coming): per-client PDF showing every asset projected to need intervention in the next quarter
- Dashboard widget (coming): "N assets approaching capacity in the next thirty days"
Why this matters
The difference between Managed Services and managed reactivity
The industry calls itself Managed Services and then runs reactively. Disk fills. Ticket opens. Customer is unhappy. Tech drives out. Same story, every quarter.
The reason mainstream RMMs don't ship forecasting isn't that the maths is hard. It is not. It is that storing a year of per-asset telemetry cheaply is hard, and most RMMs prune the data after ninety days because keeping it would melt their database. OpsMerge keeps the data, rolls it up, and projects forward. The customer-visible difference is one sentence in the next quarterly review: "we caught and replaced two disks before they filled, with planned downtime, on a Tuesday."
It is also a difference in the conversation you have when you renew the customer's contract.
Pairs well with