Predictable and unpredictable demand: when to use forecasts and when to use buffers

Predictable and unpredictable demand should not be managed in the same way. One of the most common mistakes in planning is trying to improve the forecast for every product equally, when not all SKUs behave in the same way, have the same stability or create the same operational impact.

In many cases, the problem is not the algorithm, but how demand is classified. Some products deserve analytical effort, recurring review and more sophisticated forecasting models. Others, however, should be managed with buffers, exception rules or specific decisions within the S&OP process.

Distinguishing between predictable and unpredictable demand allows planners to use their time more effectively, protect service levels and avoid unnecessary inventory. It also helps improve alignment between sales, operations, purchasing and finance, because it avoids demanding the same level of accuracy from products that behave in completely different ways.

What is forecastability in demand?

Forecastability is the extent to which demand can be predicted with a reasonable level of reliability. It does not simply measure whether a forecast was accurate in a specific period, but whether the behaviour of an SKU, family or channel makes it possible to build a useful forecast for decision-making.

Demand with high forecastability usually shows repeatable patterns, sufficient volume, low relative variability and a degree of stability over time. By contrast, demand with low forecastability tends to be intermittent, show irregular peaks, change abruptly or lack reliable historical data. In these cases, trying to improve accuracy can consume significant effort with limited operational return.

Understanding forecastability changes the way planning is approached. The aim is no longer to “forecast everything better”, but to decide what can be forecast, what should be protected with buffers and what requires exception management. This distinction is especially important in environments with large portfolios, slow-moving products, promotions, launches or highly variable demand.

Why not all demand should be forecast in the same way

Demand forecasting cannot be treated as a uniform exercise. A stable, recurring product with sufficient volume should not be managed in the same way as an intermittent SKU, a recent launch or a product affected by one-off commercial events. Each type of demand requires a different policy.

When all products are reviewed using the same metrics, criteria and frequency, the process becomes less efficient. Planning teams spend time on low-impact SKUs, while critical or genuinely manageable products do not receive the attention they need. It also creates unnecessary discussions around errors that cannot actually be corrected with a better model.

The mistake of treating every product the same

Treating every product the same often leads to unrealistic decisions. The same level of accuracy is expected from stable SKUs and intermittent products. Low-value SKUs are reviewed in the same detail as strategic families. Forecasts are penalised for behaviours that were not predictable in the first place.

This approach also distorts performance analysis. A high percentage error in a slow-moving SKU may look serious, but have little economic impact. By contrast, a small deviation in a critical family can affect inventory, production and service levels.

When improving the forecast is not enough

There are situations where improving the forecast does not solve the operational problem. If demand depends on sporadic orders, decisions from key customers, non-recurring events or market changes that are difficult to anticipate, the model will have a natural limit.

In these cases, the answer should not be to keep adjusting the forecast indefinitely. It may be more effective to define buffers, review rules, specific replenishment policies or alert mechanisms that allow the business to react when demand is triggered.

How to identify predictable demand

Predictable demand is demand that allows an actionable forecast to be built. This does not mean it will always be accurate, but that there is enough signal to anticipate expected behaviour and make reasonable decisions about purchasing, production, inventory or capacity.

Identifying it correctly allows much of the process to be automated and focuses human effort where it adds real value. To do this, it is useful to analyse historical stability, volume, variability, seasonality, behaviour by channel and sensitivity to external events.

Stable history and repeatable patterns

A stable history does not mean demand is flat. It means demand responds to recognisable patterns. There may be seasonality, weekly cycles, growth trends or recurring variations, provided those signals are repeated with a certain level of consistency.

When patterns are repeatable, the forecast can add value because it anticipates future needs on a solid basis. This makes it possible to plan purchasing, adjust inventory, prepare capacity and reduce reactive decisions.

Enough volume to detect signals

Volume is key to separating signal from noise. In products with very low demand, a one-off sale can completely distort the statistical reading. When there is sufficient volume, it becomes easier to identify trends, seasonality or real changes in behaviour.

For this reason, a fast-moving SKU is usually more suitable for forecasting than one with sporadic consumption. Not necessarily because it is more important, but because it generates enough data to build a more reliable forecast.

Low variability against the expected pattern

Variability does not always prevent forecasting. What matters is whether variability can be explained by an expected pattern or whether it is driven by erratic movements. Seasonal demand can be predictable if the seasonality repeats consistently.

By contrast, if deviations do not follow any clear logic, the forecast loses explanatory power. In those cases, it is worth reviewing whether the product should be managed through buffers, exception rules or a differentiated service policy.

How to detect unpredictable demand

Unpredictable demand is demand where the forecast has limited ability to anticipate actual behaviour. This may be due to intermittency, lack of historical data, dependency on a small number of customers, non-repeatable events, poorly documented promotions or sudden market changes.

Detecting it early avoids spending resources on an unrealistic objective. Instead of demanding accuracy where there is not enough signal, the business can define alternative policies: safety stock, exception-based review, supplier agreements, flexible lead times or specific decisions within S&OP.

Intermittent or irregular demand

Intermittent demand appears when there are long periods with no consumption and orders concentrated at specific moments. This makes historical data difficult to interpret, because the absence of demand does not always mean a loss of interest or structural decline.

In these cases, the question is not only how much will be sold, but when demand will appear and what impact it will have if availability is lacking. That is why it is often necessary to combine a limited forecast with specific coverage, replenishment or service policies.

Peaks with no clear pattern

Peaks with no clear pattern create a false sense of opportunity. After a one-off increase, some companies adjust the forecast upwards even when there is no evidence that the behaviour will repeat. This can end up inflating inventory or capacity.

To manage these cases, it is important to separate explainable events from noise. If the peak is caused by a promotion, an exceptional order or a previous stockout, it should be treated as an exception, not as a new trend.

New products or unreliable history

New products present an obvious challenge: there is not enough historical data to build a robust forecast. In these cases, the forecast should be supported by analogies, launch curves, commercial information, initial orders or data from similar products.

Even so, uncertainty needs to be acknowledged. A launch should not be managed with the same confidence as a mature SKU. It is better to define scenarios, review thresholds and adjustment rules as real information becomes available.

When to use forecasts

Forecasts should be used when they provide a sufficiently reliable signal for decision-making. It is not just about generating a number, but about having a forecast that helps plan inventory, purchasing, production, capacity or distribution more effectively.

Using forecasts where they truly add value makes it possible to automate recurring decisions, reduce urgent interventions and improve coordination between departments. It also helps build a common language around expected demand and its operational implications.

Families with recurring demand

Families with recurring demand are natural candidates for forecasting. Their behaviour makes it possible to identify patterns and anticipate needs with a reasonable margin. This helps create more stable plans and reduce unnecessary manual adjustments.

In these families, the forecast can serve as a basis for purchasing, production, replenishment and financial planning. The key is to review relevant deviations without turning every minor variation into an operational discussion.

Products with high operational impact

It is also worth applying forecasts to products with high operational impact, even if their behaviour is not perfect. If a deviation affects critical capacity, suppliers, service or margin, it deserves more attention than average.

In these cases, the forecast is not only about accuracy. It also helps anticipate risks, prepare scenarios and assess decisions before the impact reaches the operation.

Horizons where the forecast adds value

A forecast does not have the same value across all horizons. In the short term, it can help adjust availability, replenishment or immediate production. In the medium term, it supports capacity, purchasing and resource planning. Over the long term, it helps guide strategic decisions.

That is why forecastability should also be assessed by horizon. Demand may be unreliable week by week, but still useful for planning monthly trends or aggregated capacity decisions.

When to use buffers

Buffers are protection mechanisms against uncertainty. They may take the form of safety stock, flexible capacity, supplier agreements, reaction times or prioritisation rules. Their role is not to replace the forecast, but to protect the operation when the forecast has limits.

Using buffers intelligently avoids two extremes: placing too much trust in unreliable forecasts or oversizing resources out of fear of uncertainty. The key is to define which risk needs to be covered, how much it costs to cover it and what the impact would be of not doing so.

SKUs with high uncertainty

SKUs with high uncertainty usually need buffers because the forecast does not offer enough reliability. This happens in products with low rotation, erratic demand, dependency on a small number of customers or high sensitivity to external events.

The buffer should be sized according to criticality, margin, lead time, stockholding cost and expected service level. Not all uncertain SKUs deserve the same level of protection.

Products that are critical for service

Some products need to be protected even if their demand is difficult to forecast. These may be SKUs that are critical for strategic customers, components that block production or items where service failure generates significant penalties.

In these cases, the buffer is not justified by statistical accuracy, but by business impact. The right decision may be to maintain additional coverage, secure availability with suppliers or define allocation rules.

Demand that is difficult to model

When demand is difficult to model, a forecast can still exist, but it should not be the only basis for decision-making. It should be complemented with coverage rules, scenarios, exception-based reviews and action criteria.

This allows the business to accept that uncertainty exists without paralysing planning. Instead of waiting for a perfect forecast, the company defines how it will respond when actual behaviour moves away from the expected scenario.

When to apply exception rules

Exception rules are used to manage situations that should not be mixed with baseline demand. Their purpose is to prevent one-off events, launches, promotions or extraordinary changes from contaminating the recurring forecast.

Managing by exception does not mean improvising. It means defining clear criteria to detect when an SKU should leave the standard planning flow and move into a specific review process. This improves forecast quality and reduces noise in the process.

One-off commercial events

Promotions, campaigns, special orders or commercial agreements can significantly alter demand. If these events are treated as normal demand, the model may interpret a one-off peak as a trend.

That is why they need to be recorded, measured and separated from the recurring baseline. This keeps the forecast cleaner, while promotional impact is managed through its own rules.

Launches and end of life

Launches require specific monitoring because they combine lack of history, commercial uncertainty and operational risk. In these cases, adoption curves, initial orders and analogies with similar products can help, but they need to be reviewed frequently.

End of life also requires differentiated rules. If it is not managed correctly, it can create excess inventory, obsolescence or unnecessary stockouts in the final stage of the product.

Sudden market changes

Market changes can break historical patterns. A price change, new regulation, competitor entry or shift in customer behaviour can reduce the usefulness of historical data.

When this happens, the forecast needs to be revalidated. It is not enough to continue projecting the past. Business judgement, external signals and alternative scenarios need to be incorporated.

Which demand should be reviewed in S&OP?

S&OP should not review all demand with the same level of detail. Its role is to make decisions that matter to the business, not to become a meeting for SKU-level micro-adjustments. Forecastability can therefore help decide which topics should be escalated.

The demand brought into S&OP should be demand that requires cross-functional alignment: strategic products, deviations with financial impact, capacity constraints, supply risks or decisions affecting service, inventory and margin.

Products that are strategic for the business

Strategic products should be reviewed even when they are not always the highest-volume items. They may be relevant because of margin, key customers, commercial positioning, plant impact or dependency on critical suppliers.

In these cases, the forecast should be analysed alongside operational risks. The decision is not only how much is expected to sell, but what the organisation needs to do to ensure availability and profitability.

Deviations with financial impact

A relevant deviation in demand can affect cash, margin, inventory or capacity. When the financial impact is significant, the decision should be escalated to S&OP to avoid isolated responses.

This makes it possible to assess alternatives: adjusting production, reviewing purchasing, changing stock policies, prioritising customers or modifying commercial commitments. Demand stops being a number and becomes a business decision.

Decisions that require consensus

Some decisions cannot be made by planning alone. Prioritising capacity, limiting demand, accepting additional inventory or activating alternative suppliers requires consensus between sales, operations, purchasing and finance.

S&OP provides the framework for making these decisions with shared data. Forecastability helps prepare the conversation by separating what is predictable from what is uncertain, and what is operational from what is strategic.

How to measure each type of demand

Measuring all products with the same metric creates unfair readings and unhelpful decisions. A stable SKU can be evaluated using forecast accuracy, while intermittent demand may need to be measured by availability, coverage or compliance with rules.

Measurement should be adapted to the type of demand and the planning objective. This is not about abandoning traditional metrics, but using them where they make sense and complementing them when they do not explain the operational reality well enough.

Metrics for predictable demand

For predictable demand, it makes sense to use metrics such as forecast accuracy, absolute error, bias, forecast stability and deviation by horizon. These metrics help improve models, detect deviations and adjust the planning process.

It is also useful to measure the operational impact of forecast error. An apparently accurate forecast may still not be enough if it fails for critical products, key periods or families with high sensitivity to capacity.

Metrics for unpredictable demand

For unpredictable demand, evaluating accuracy alone may be of limited use. It is more relevant to measure availability, coverage, stockout frequency, cost of protection, service level and response to demand events.

These metrics recognise that the objective is not always to hit the exact number. Sometimes, the goal is to ensure the organisation can respond at a reasonable cost when demand appears.

Indicators for exception management

Exception management needs indicators that trigger review. These may include deviations from the expected pattern, sudden changes in demand, atypical orders, lead time variations or the expected impact on inventory and service.

The key is to define clear thresholds. If everything generates an alert, the system loses its value. If alerts are well calibrated, the planner can dedicate time to decisions that genuinely change the outcome.

How to reduce the planner’s workload

One of the biggest benefits of classifying demand is reducing the planner’s operational workload. Not every SKU requires manual review, and not every deviation deserves a meeting. Segmentation makes it possible to work more intelligently.

Instead of reviewing the entire portfolio, the planner can focus on exceptions, risks and decisions with impact. This improves team productivity and increases decision quality, because human judgement is applied where it truly adds value.

Automating what is predictable

Predictable demand can be automated to a large extent. If behaviour is stable, the model works reasonably well and deviations remain within acceptable thresholds, there is no need for constant intervention.

Automation does not mean losing control. It means defining rules, monitoring exceptions and allowing the system to maintain the standard flow while the team focuses on cases that require analysis.

Prioritising what requires judgement

The planner’s judgement is most valuable when there is uncertainty, impact or conflict between objectives. For example, deciding whether to protect a critical product, review a promotion, adjust coverage or escalate a deviation to S&OP.

Segmentation by forecastability enables better prioritisation. The planner stops acting as a permanent forecast corrector and becomes a manager of operational decisions.

Avoiding reviews with no real impact

Many organisations spend too much time reviewing deviations that do not change any decision. This creates administrative workload, unproductive meetings and a sense of control with no real impact.

A good process should always ask what decision will be made as a result of each review. If the answer is none, that review can probably be automated, simplified or removed.

Software for classifying demand

Manually classifying predictable and unpredictable demand may be viable in small portfolios, but it becomes complex when there are thousands of SKUs, multiple channels, several warehouses and different planning horizons. In this context, demand forecasting software becomes a key enabler.

Demand forecasting software makes it possible to segment products, detect patterns, measure forecastability, activate alerts and connect the forecast with inventory, purchasing, production and S&OP. This prevents classification from becoming a static exercise and turns it into a recurring planning capability.

Dynamic product segmentation

Dynamic segmentation makes it possible to classify SKUs according to their real behaviour: stability, variability, volume, intermittency, criticality or economic impact. This classification can also be updated when market or product conditions change.

This matters because an SKU does not always belong to the same category. A new product can build history, a stable SKU can become irregular, and intermittent demand can become strategic if its impact on service changes.

Exception alerts

Exception alerts help filter out noise. Instead of manually reviewing the entire portfolio, the system can highlight relevant deviations, pattern changes, stockout risks, excess coverage or products that require intervention.

When configured correctly, these alerts reduce the planner’s workload and improve response speed. The goal is not to generate more information, but to highlight what requires a decision.

Scenarios connected with inventory

Demand classification becomes more valuable when it is connected with inventory. It is not enough to know whether an SKU is predictable or not. The business needs to understand which stock, coverage or replenishment policy it requires.

Advanced software makes it possible to simulate scenarios and assess the impact of different decisions: increasing buffers, reducing coverage, changing replenishment frequency, prioritising suppliers or reviewing service levels. In this way, forecastability is translated into concrete decisions.

Predictable and unpredictable demand for better planning

Distinguishing between predictable and unpredictable demand makes planning more realistic, because it recognises that not all products should be managed with the same logic. Some need forecasts, others require buffers, and others should be handled through exception rules or decisions within S&OP.

This approach helps improve service levels, reduce unnecessary inventory, prioritise the planner’s work and support more consistent cross-functional decisions. Instead of pursuing uniform accuracy across the whole portfolio, the company learns to apply the right level of effort to each type of demand, based on its predictability, criticality and operational impact.

To do this consistently, it is not enough to classify products once a year or manually review thousands of SKUs. Demand forecasting software makes it possible to automate segmentation, detect pattern changes, activate exception alerts and connect each decision with inventory, purchasing, production and S&OP.

At Imperia, we work to ensure demand forecasting is not just a number, but a real decision-making tool. With SCP Studio, we help classify demand, detect exceptions, connect forecasts with inventory and prepare scenarios that enable planning with better judgement. To see how this methodology can be applied in your business, request a demo with our experts.