About this article: This article is reviewed against the cited sources. This is source research, not lab validation. Check the stated versions and your own requirements before making changes.
Take twelve months of figures for each resource (the monthly peak for processor and memory, month-end used space for storage), draw a straight line from the first month to the last, and extend it until it crosses the level at which you want to upgrade. That gives a likely month. Where usage has a yearly pattern, such as an end-of-financial-year rush, compare each month with the same month last year as well. Then subtract the time it takes to approve and carry out the upgrade, and that is when the order needs to go in.
Why it matters
Upgrades that arrive as emergencies cost more: rushed approvals, after-hours work and users waiting on a full disk or a slow server. A simple forecast turns "we will need more space at some point" into a budget line with a month attached. It does not need a statistician; it needs a year of consistent figures and some care about what they mean.
How it works
Step 1: collect the right figures
For each server, record monthly:
- Storage: used space on each volume at month end.
- Memory: the peak in-use figure during business hours.
- Processor: the peak sustained use during business hours, not a momentary spike.
Windows can collect these itself. Microsoft's logman tool "Creates a counter data collector" that records performance counters on a schedule. Your hosting provider may also hold this history; ask for it.
Use peaks, not averages, for memory and processor. An average can look comfortable while month-end is already saturated.
Step 2: look for trend and seasonality
The forecasting textbook Forecasting: Principles and Practice describes a time series as "a trend-cycle component, a seasonal component, and a remainder component". In plain terms: is it growing, does it repeat a yearly pattern, and how much bounces around for no obvious reason? Plot the twelve figures before choosing a method.
Step 3: apply a simple method
Two simple methods from the same textbook fit small-business capacity figures:
- Drift method: the forecast grows by the average monthly change in the history. The authors explain this "is equivalent to drawing a line between the first and last observations, and extrapolating it into the future". Use it for storage that grows at a steady pace, not for figures with a strong yearly cycle, such as processor peaks at end of financial year.
- Seasonal naïve method: "we set each forecast to be equal to the last observed value from the same season (e.g., the same month of the previous year)". Good for processor and memory peaks that follow the business calendar.
Where both trend and season are present, a rough approach we suggest (our own, not the textbook's) is to take last year's same-month figure and add the year's growth. Treat it as a first estimate: with only one year of history you have seen each season once, so you cannot yet tell trend and season apart with confidence; two or more years make it far more reliable.
What to choose and why
Worked example: storage
A hypothetical file server's data volume:
- November last year: 1,200 GB used.
- October this year: 1,620 GB used.
- That is 420 GB of growth over 11 monthly steps, about 38 GB a month.
- The volume is 2,500 GB, and the business has chosen to act at 80 per cent, which is 2,000 GB. (The threshold is the business's own choice; our article on headroom uses 85 per cent in its example. Neither figure is a rule.)
- The gap is 380 GB. At 38 GB a month, that is about 10 months away: roughly August next year.
- If approval and the change take two months, the order should be raised around June.
Worked example: processor peak
The application server peaked at 55 per cent in June 2026 (end-of-financial-year reporting) and runs around 35 per cent in other months. For illustration, the business assumes the comparable June peak will grow by five percentage points over the following year. Its preliminary June 2027 forecast is therefore 60 per cent. This is an assumption to validate, not a measured twelve-month change. A straight-line view on the average would have missed June entirely.
Both examples are illustrative.
Turn the forecast into a decision
- Record the forecast month and the assumption behind it (method, threshold, lead time).
- Put the cost in the next budget.
- Re-check quarterly; replace the forecast if the line bends.
Read next
Related articles you may want next.
- HostingHow much headroom to keep: thresholds and lead time for hosted resourcesHow a business sets written trigger levels for storage, memory and processor on hosted servers, so upgrades are ordered early enough, peaks are absorbed and nobody argues about when to buy more.Read article
- HostingMonth-end, EOFY and seasonal peaks: planning temporary capacityHow an Australian business maps its payroll, BAS and end-of-financial-year deadlines onto its hosted servers, and decides whether to cover a predictable peak with a temporary increase or year-round headroom.Read article
- HostingSizing a business application server from measurements, not the vendor's minimumWhich Windows performance counters to collect on an existing server, for how long and at what interval, and how to turn the results into processor, memory and disk sizes for a hosted replacement.Read article
Risks and limits
- A forecast is a range, not a date. The textbook explains that "A prediction interval gives an interval within which we expect y_{t} to lie with a specified probability". Treat the month as a middle estimate and act early if being late is costly.
- Business changes break trends. A new branch, new software or a document-retention change will move the line. Re-forecast when the business changes.
- One year shows one season. With twelve months you have seen each season once; a single unusual June can mislead.
- Clean-ups distort history. A big archive move shows as a drop; note it so it is not read as a trend.
- Peaks need consistent measurement. Changing how peaks are measured mid-year makes the series incomparable.
Sources
Opened 15 November 2026; the decomposition chapter and the published sizing article opened 16 November 2026.
- Hyndman, R.J. and Athanasopoulos, G., Forecasting: Principles and Practice, 3rd edition, OTexts: Time series decomposition, Some simple forecasting methods and Prediction intervals.
- Microsoft Learn: logman create counter, updated 22 October 2025.
Related article: Sizing a business application server from measurements, not the vendor's minimum covers how to collect the figures across a full business cycle.
Related service: Managed support.
Learning notes
- Use monthly peaks for processor and memory, month-end used space for storage.
- The drift method draws a line from the first to the last month and extends it.
- The seasonal naïve method repeats last year's same month.
- Subtract approval and change lead time from the forecast month.
- Re-forecast quarterly and whenever the business changes.
This article was originally published on .