Croston Intermittent Demand Forecast Calculator
Spare parts and slow movers sell in occasional lumps, and ordinary exponential smoothing chases each lump and then decays. Croston's method smooths demand size and demand interval separately, which gives a steadier rate. This calculator performs the update and shows the bias corrected SBA rate.
Your numbers
Smoothed size of non-zero demands before this update.
Smoothed number of periods between demands.
Interval that ended with the new demand.
Typical range 0.05 to 0.2 for intermittent items.
Your results
One update step only. In practice the estimates are initialised from history and updated only in periods with demand; periods with zero demand leave both estimates unchanged.
Forecast rate vs smoothing constant
Sweeps Smoothing constant alpha from 0.05 to 0.5, holding your other inputs fixed.
Forecast spare parts from your ERP
Optional: ask for a Croston and SBA run over your service parts history.
Optional. The tool and every export work without this. No spam.
How to use it
Enter the last stored size estimate and interval estimate for the item. When a new demand arrives, enter its quantity and the number of periods since the previous demand. Pick alpha, usually between 0.05 and 0.2, and a horizon. Store the updated size and interval as the new starting estimates, and repeat on the next demand. In periods with no demand nothing changes.
- Initialise from the first few demands or the history average
- Keep alpha low because lumps are noisy
- Track forecast at the lead time total, not per period
- Check ADI first with the CV classification tool
The formulas
When demand d occurs after q periods: z becomes z plus alpha times (d minus z), p becomes p plus alpha times (q minus p), and the forecast per period is z divided by p. Croston is known to be biased high because the expectation of a ratio is not the ratio of expectations. The Syntetos-Boylan Approximation multiplies by one minus alpha over two, which removes most of that bias and is generally preferred.
Worked example
Size 12, interval 5, a demand of 20 arrives after 3 periods, alpha 0.1. New size is 12.8, new interval is 4.8, Croston rate is 2.67 per period and the SBA rate is 2.53. Over 12 periods that is about 30 units. Compare that with a plain moving average, which would jump when the lump of 20 enters and drop again a few periods later.
Common mistakes
Applying Croston to smooth items wastes accuracy, so classify first. Measuring accuracy with MAPE fails on zeros, so use MASE, scaled error or service level outcomes instead. Setting alpha too high makes it behave like naive forecasting. And remember that the rate is an average: safety stock for intermittent items needs a lead time demand distribution, not a normal approximation.
Frequently Asked Questions
When should I use Croston instead of exponential smoothing?
When many periods are zero, roughly ADI above 1.32. For smooth items with demand nearly every period, plain exponential smoothing is simpler and usually just as accurate.
What is the SBA correction?
The Syntetos-Boylan Approximation multiplies the Croston rate by one minus alpha over two. It reduces the upward bias in Croston's estimate and is widely used for spare parts.
Does a zero period update the forecast?
No. In the classical method both estimates are only updated when a non-zero demand occurs. The interval count simply keeps growing until the next demand.
How do I measure accuracy for intermittent items?
Avoid MAPE because actuals are often zero. Use scaled errors such as MASE, bias, or judge by the stock outcomes: fill rate and inventory held.
Netray can compute Croston or SBA forecasts for your spare parts directly from ERP transaction history and write suggestions back for planner approval.
Related Tools
Demand Variability CV and XYZ Classification Calculator
Turn mean demand, standard deviation and average demand interval into a CV, an XYZ class and a Syntetos-Boylan demand pattern so you can pick the right forecast method.
Supply Chain & InventoryMAD to Safety Stock Forecast Error Calculator
Use your measured mean absolute deviation of forecast error to set safety stock and reorder point instead of a guessed buffer.
Supply Chain & InventoryLinear Regression Trend Forecast Calculator
Enter six periods of demand and get the least squares trend, how well it fits and the forecast for the period you choose.
Supply Chain & InventoryNew Product Launch Demand Ramp Calculator
Plan a launch with no history: ramp to peak, total demand over the horizon, channel pipeline fill and a suggested initial build quantity.
More Supply Chain & Inventory tools
See all- Oracle Fusion Inventory Turns and Carrying Cost Calculator
- Distribution ERP Fill Rate Calculator
- Distributor Rebate Tracking Worksheet
- Branch Transfer Planning Worksheet
- Distributor Slow-Moving Inventory Review Worksheet
- Distributor Vendor Performance Scorecard Worksheet
- Distributor GMROI Calculator
- Distributor Backorder Cost and Fill Rate Calculator
- EOQ with Vendor Minimum Order Calculator
Keep exploring
Free tools
- Forecast Exception Review TemplateDocument which SKUs breached your forecast tolerance, why, and who fixes it before the demand review. Exports to PDF or Word.
- Demand Review Meeting Minutes TemplateCapture forecast changes, assumptions, risks and owned actions from the monthly demand review in one clean document.
- Moving Average Forecast CalculatorEnter the last six periods of demand, choose a 3, 4 or 6 period window and get the simple moving average forecast with a comparison to the naive forecast.
ERP fixes and how-tos
- SyteLine forecasting, planning horizon and planning calendarHow forecasting, the planning horizon and the planning calendar work together in SyteLine MRP and APS, with parameter guidance and common mistakes to avoid.
- Parts Not Showing Demand in Epicor Kinetic: Forecast, Planning BOM and Time Fence ChecksWhy demand is missing in Epicor Kinetic MRP: forecast entry, planning BOM, time fence settings, release dates, and capable to promise, with a check list.
- SYSPRO Forecasting: How Forecasts Feed MRPHow SYSPRO forecasting feeds MRP: setting up forecasts, choosing a method, how sales orders consume forecast, and how to check the MRP result before releasing.
Guides
- Demand Sensing for Manufacturers: Shortening the Forecast HorizonDemand sensing for manufacturers: how short-horizon signals, POS and channel data, and ML models cut forecast error 20-40% over statistical baselines.
- Demand Forecasting Configuration in Infor SyteLineConfigure SyteLine demand forecasting with statistical models, forecast consumption, demand management parameters, and accuracy measurement for production planning.
- Infor M3 Demand Planning and Forecasting ConfigurationConfigure Infor M3 demand planning with statistical forecasting, forecast consumption, MPS integration, and demand sensing for accurate production scheduling.
AI on your ERP
- AI Demand Forecasting for Manufacturers: Better Numbers and an Explanation Planners Can TrustAI demand forecasting explains ERP forecast changes in plain language, flags exceptions for planners, and handles short-history SKUs without Excel overrides.
- NetSuite Inventory and Demand Planning With AI That Checks Your Reorder LogicAI for NetSuite inventory and demand planning: review reorder points, safety stock, item demand plans and stockouts using SuiteQL on private infrastructure.
Troubleshooting
- Why PyTorch GPU memory fragmentation causes intermittent CUDA out of memory errorsPyTorch GPU memory fragmentation causes intermittent CUDA OOM despite free VRAM. Fix it with expandable_segments and shape bucketing, not a bigger GPU.
- Why you get CUDA out of memory when nvidia-smi shows free VRAMCUDA out of memory despite free VRAM in nvidia-smi is memory fragmentation, not exhaustion. Fix it with PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True.