Anticipating a stockout: what signals does an organisation really need?
A stockout is not primarily an AI problem. It is a problem of anticipation, data and decision-making. This Insight shows when rules, forecasting and predictive models become useful.
Stockouts are often noticed too late. Yet several signals exist before inventory reaches zero: available stock, consumption rate, inbound orders, supplier lead time, seasonality and forecast demand.
Before choosing an algorithm, identify the operational signals that actually describe the problem.
01 · THE BUSINESS PROBLEM
Before predicting, understand the stockout mechanism
Available inventory alone is not enough. The same stock level can be comfortable or critical depending on demand and replenishment lead time.
A 2025 study using more than 1.6 million SKUs found current inventory, short-term demand forecasts and recent sales among the most influential stockout predictors.
02 · START SIMPLE
A first alert does not necessarily require AI
A first measure compares stock coverage with replenishment time.
CoverageCt = St / D̄
Available stock and inbound orders can then be compared with expected lead-time demand plus safety stock.
Alert ruleSt + Ot < DLTt + SS
This deterministic rule is transparent and can already be operationally useful for a limited number of relatively stable items.
PEDAGOGICAL DEMONSTRATOR
Change the signals
Buffer availableProjected position remains above safety stock.
EXPLAINABLE DIAGNOSIS
Current coverage exceeds supplier lead time and the inbound order maintains a buffer above safety stock.
PROJECTED INVENTORY TRAJECTORY
Forecast ≠ stockout
A rise in forecast demand does not automatically mean a stockout. Risk depends on the combination of demand, available stock, inbound orders, replenishment lead time and safety buffer.
This demonstrator illustrates a decision logic using simulated data. It is not a performance validation on real company data or an inventory policy recommendation.
03 · WHEN FORECASTING HELPS
Demand is not always regular
Trend, seasonality, promotions, one-off orders and long zero-demand periods can make averages insufficient. Spare parts are a classic intermittent-demand case.
Intermittent-demand research shows why standard methods are not always appropriate. Croston-type methods separately model demand intervals and positive demand sizes, but no method is universally best.
Rules & indicators
Stock, consumption, lead time, inbound orders and thresholds.
Forecasting
Estimate future demand when history supports learning trend, seasonality or intermittence.
Predictive model / AI
Combine many signals to estimate and prioritise stockout risk.
04 · WHEN AND HOW TO USE AI
AI becomes useful when rules are no longer enough
The goal is not to replace inventory management, but to improve anticipation and prioritisation. AI is worth testing when many items, heterogeneous demand patterns or multiple frequently updated signals make simple rules insufficient.
In practiceStock + sales + orders + lead times + context → model → forecast or risk score → dashboard / alert → human decision.
When are rules enough? If a few explainable thresholds already identify risky situations well, adding AI may not create additional value.
05 · MEASURE VALUE
An accurate model is not necessarily a good inventory system
Forecast accuracy alone is insufficient. Operational consequences also matter: stockout rate, service level, fill rate, average stock, overstock, holding cost, stockout cost and actionable alerts.
For classification models, precision, recall, F1, PR-AUC or calibration may complement evaluation. The best system improves inventory decisions under the organisation's actual objectives and constraints.
06 · WHERE TO START
A path proportionate to organisational maturity
- Identify the decision to improve.Which stockouts actually matter and how early must action be taken?
- Check available data.Stock, sales, orders, suppliers, lead times, quality and history.
- Build a first rule.Test an explainable alert before adding complexity.
- Measure what it catches and misses.Link evaluation to business consequences.
- Add forecasting or AI if value is demonstrated.Increase sophistication only when the problem and data justify it.
The OECD highlights data maturity as a fundamental barrier to AI adoption and notes that managers can struggle to connect AI with concrete workplace problems. This progression therefore starts from the need rather than the technology.
MINI-DIAGNOSTIC
What level of solution does your situation justify?
Select what is already available in your organisation. This educational orientation does not replace a full business and data assessment.
Start by structuring the essential data.A simple, explainable rule can be the first solution level.
Choose a response proportionate to the situation
The appropriate solution depends on data quality, process reliability, the number of items and demand variability. Depending on the context, an explainable business rule, targeted automation or a forecasting system can address the need without unnecessary complexity.
SOURCES & LIMITATIONS
Main references
Croston, J. D. (1972). Forecasting and stock control for intermittent demands. Operational Research Quarterly, 23(3), 289–303.
Pinçe, Ç., Turrini, L., & Meissner, J. (2021). Intermittent demand forecasting for spare parts: A critical review. Omega, 105, 102513. https://doi.org/10.1016/j.omega.2021.102513
Teunter, R. H., & Duncan, L. (2009). Forecasting intermittent demand: A comparative study. Journal of the Operational Research Society, 60, 321–329. https://doi.org/10.1057/palgrave.jors.2602569
Teunter, R. H., Syntetos, A. A., & Babai, M. Z. (2011). Intermittent demand: Linking forecasting to inventory obsolescence. European Journal of Operational Research, 214(3), 606–615. https://doi.org/10.1016/j.ejor.2011.05.018
OECD. (2025). The adoption of artificial intelligence in firms: New evidence for policymaking. OECD Publishing. https://doi.org/10.1787/f9ef33c3-en
Stockout prediction study (2025). Study based on more than 1.6 million SKUs. ScienceDirect.
About the demonstrator
The demonstrator parameters and results are simulated for illustration only. They are not an empirical validation of a SolDataSynth model on client data.