Forecast and Manage Demand — 10%

How this domain shows up on the exam

Forecast and Manage Demand is Domain I of the CSCP Exam Content Manual v5.0, carrying 10% of the exam^[1]. The manual frames it as one of the first activities in supply chain management: assemble and evaluate the various types of demand for products and services — historical information and future predictions — use that to determine a forecast of future needs, and balance supply with demand^[1].

Conceptually, this domain is the starting gun of supply chain planning. Capacity plans, inventory targets, procurement schedules, and staffing all key off the demand forecast; when the forecast is wrong, every downstream plan inherits the error. That is why it matters to you as a working professional, not just an exam candidate: forecast error is the quiet root cause of both stockouts and excess inventory. The four sections below follow the manual's order — collect and analyze data, understand how demand can be influenced, build the forecast, then align supply with it.

How the exam tests this domain: scenario questions, not definitions. You will be handed a distorted demand history, an error metric with a sign, a tracking-signal value, or a method applied to the wrong demand pattern — and asked what to do. The trap running through the whole domain is confusing the map for the territory: treating the forecast as a target to be met rather than an estimate of demand to be served, and treating a clean-looking metric — MFE near zero, a tracking signal inside its limits — as proof of accuracy.

For reference on the exam itself: 150 questions total — 130 scored plus 20 pretest — taken over 3.5 hours, reported on a 200–350 scaled score with 300+ required to pass^{[2, 3]}.

A. Collect and Analyze Historical and Environmental Demand Data

Concept. Forecasting starts before any math, with two kinds of inputs. Historical demand data is the record of what customers actually wanted — and the discipline here is that demand is not the same as shipments. Shipments are capped by what supply could deliver: in a stockout month the sales record understates true demand, while a bulk-order month may overstate the ongoing run rate. Environmental demand data is the context around the numbers — economic conditions, market trends, the product's life-cycle stage, seasonality, competitor actions, regulatory shifts, and planned events. Analysis means examining both together: cleaning one-time distortions out of the history and noting the forces that will shape the future.

Why it matters. Every forecast inherits its data. Fit a model to supply-constrained shipment history and you bake a permanent under-forecast bias into the plan before a single smoothing constant is chosen. Miss that a product is entering decline, and the trend your model extrapolates is already dead. Data work is unglamorous, and it is where most real-world forecast failures begin.

How it is tested. You will get a history with a poisoned month — a factory shutdown, a one-time bulk order, a stockout — and be asked how to handle it before forecasting. The right move is to flag or adjust the one-time event so it does not distort the baseline: neither leaving it in untouched nor declaring the product unforecastable. A second classic: distinguishing true demand from the sales record when supply was constrained.

The trap. Two opposite mistakes. One is feeding raw sales or shipment data straight into the model as if it were demand. The other is over-scrubbing — adjusting the history until it tells the story you wanted to hear. The exam also tests whether environmental data is treated as a real input: a forecaster who ignores a known market shift is not being "objective," just blind.

B. Influence Demand Through Marketing Activities

Concept. Demand is not only observed; it is shaped. Promotions, pricing changes, advertising campaigns, product launches, and channel incentives all move demand deliberately — and the demand plan has to see them coming. This section exists because the forecast and the marketing calendar are describing the same future from opposite ends.

Why it matters. Two practical reasons. First, planned demand-shaping events must be visible to the forecaster: a promotion week that lifts volume 40% is a promotional event, not a new baseline — fold it into the baseline and you will over-forecast every period after it. Second, demand shaping is itself a supply-balancing tool. When capacity is tight in one period and slack in another, pricing and promotion timing can shift demand toward the slack — managing demand is part of aligning supply with demand, not a distortion of it.

How it is tested. Expect a scenario with a promotion spike in the history and a question about how to treat it in the baseline forecast (isolate the uplift; do not bake it in). Or the reverse: a capacity crunch where the right answer is a demand-side lever — shifting demand with pricing or promotion timing — rather than waiting for the forecast to fix itself.

The trap. Baking promotional lift into the baseline is the big one: it feels like "using all the data," but it manufactures bias. The subtler trap is treating the forecast as a constraint on marketing — "the forecast says 10,000 units, so we cannot run the promotion" — instead of recognizing that marketing activities are legitimate inputs that change the very demand the forecast must describe.

C. Build the Forecast

Concept. This is the technical core of the domain, and it rests on a small set of ideas the exam tests relentlessly: how error is defined, how it is summarized, and which method fits the demand pattern.

Start with the error itself. Forecast error is actual demand minus forecast^[4]. The sign carries meaning: a positive error means demand exceeded the forecast (an under-forecast); a negative error means the forecast exceeded demand (an over-forecast).

Summaries of error answer two different questions. Mean forecast error (MFE) — the average of the signed errors — measures bias: whether the model systematically leans high or low. Its ideal value is zero; a persistently positive MFE means the model tends to under-forecast, a persistently negative MFE means it tends to over-forecast^[4]. Mean absolute deviation (MAD) — the average of the absolute errors — measures the absolute size of the errors regardless of direction^[4]. Both are used to compare alternative forecasting models and to flag, by exception, the models that need attention.

The tracking signal operationalizes that flagging. It is monitored to pinpoint forecasting models that need adjustment, and the rule of thumb is: as long as the tracking signal stays between −4 and +4, assume the model is working correctly^[4]. Outside that band, investigate.

For the forecasting method itself, the exam's weight falls on exponential smoothing. The idea: weight past observations with exponentially decreasing weights, so recent history matters most — unlike a simple moving average, which weights past observations equally^[5]. Single exponential smoothing updates with the formula S(t+1) = α·y(t) + (1−α)·S(t), where the smoothing constant α sits between 0 and 1^[6]. Read it in plain language: the new forecast equals the old forecast plus a fraction α of the last forecast error^[6]. Push α toward 1 and the forecast reacts strongly to the latest actual demand; pull it toward 0 and the forecast barely moves.

How are those constants chosen? Objectively: try candidate values and select the ones that minimize an error-size criterion — mean squared error (MSE), mean absolute error (MAE), or mean absolute percent error (MAPE)^[7]. MAPE, being a percentage, is the natural choice when accuracy must be compared across products at very different volume scales.

Method must match pattern. Single exponential smoothing is not very good when the data has a trend — one smoothing constant is not enough to track both level and slope^[6]. When trend is present, double exponential smoothing — Holt's method — is the tool: it smoothes the data with a second equation for the trend component^[7].

Why it matters. The bias-versus-magnitude distinction drives action. Bias — MFE drifting from zero, tracking signal outside its limits — means something systematic is wrong: fix the model or the process. Pure noise — MFE near zero but large MAD — means the world is volatile, not that the model is broken; chasing noise with constant method changes makes forecasts worse. Forecast accuracy flows straight into safety stock, service levels, and capacity plans — the rest of this book assumes this chapter works.

How it is tested. Compute a single-smoothing update from a prior forecast, an actual, and α. Interpret an MFE or MAD value. Decide what a tracking signal of +5 means. Choose Holt's method over single smoothing for trending data. Pick the better of two models using error measures. Select a smoothing constant by minimizing MSE across trials.

The trap. Four classics. (1) MFE near zero does not mean accurate — large positive and negative errors cancel; always check MAD alongside it. (2) A tracking signal inside ±4 does not mean errors are small — it means there is no strong bias signal; noise can still be large. (3) Lowering α for "safety" when demand is actually shifting: stability is not accuracy, and an unresponsive forecast on moving demand is just a confident lag. (4) Running single smoothing on trending data and concluding the method "needs a smaller α" — the problem is structural, not parametric; trend needs Holt's.

D. Align Supply with Demand

Concept. The forecast describes demand; the supply plan must then be shaped to meet it — or demand shaped, within policy, to fit achievable supply. Alignment is the reconciliation step where the demand plan meets capacity, production schedules, inventory buffers, sourcing, and staffing. A forecast nobody aligns to is decoration.

Why it matters. Imbalance is where forecast error turns into money lost: too little supply against the forecast means stockouts, expediting, and lost sales; too much means excess inventory, obsolescence, and tied-up working capital. Alignment is also recurring — demand, capacity, and supply conditions move, so the balance has to be re-struck, not set once.

How it is tested. The signature scenario: the agreed demand forecast exceeds confirmed capacity, and you must choose the response. The correct move is supply-side — overtime, outsourcing, rescheduling, inventory draws, expedited sourcing — while the demand forecast stays an honest read of demand. Demand-side levers (pricing, promotion timing) are legitimate too; editing the forecast to match capacity is not.

The trap. "Reconciling" by revising the forecast down to what the plant can make. That hides the gap instead of closing it, and it corrupts the one honest number the whole planning process depends on. The forecast is a statement about demand, not a capacity plan — never let the two be confused.

Key numbers

Key takeaways

• Demand data comes in two kinds — historical (what was truly demanded, not merely shipped) and environmental (market, life-cycle, and economic context) — and both must be analyzed before any forecast is built.
• Demand can be shaped: promotions, pricing, and launches move demand deliberately, so promotional uplift must be isolated from the baseline — and demand shaping is a legitimate balancing tool.
• Forecast error is actual demand minus forecast; the sign of the error tells you the direction of the miss.
• MFE measures bias (ideal zero; positive means under-forecasting, negative means over-forecasting) while MAD measures the absolute size of errors — check both, because cancelling errors can hide behind an MFE of zero.
• A tracking signal outside ±4 is the rule-of-thumb trigger to investigate and adjust the model.
• Exponential smoothing weights recent history most heavily; single smoothing lags on trending data, where Holt's double exponential smoothing is the right tool; smoothing constants are chosen by minimizing MSE, MAE, or MAPE.
• Alignment means balancing supply with demand through supply-side and demand-side actions — never by editing the forecast to match capacity.

Chapter 1 quiz — 12 questions

1. A planner replaces a simple moving average with exponential smoothing on stable, level demand. The main change in how past observations are weighted is that

• A. every past observation continues to carry exactly equal weight
• B. older observations receive exponentially decreasing weights over time
• C. only the latest observation counts; all earlier ones are dropped
• D. the oldest observations receive the largest weights in the average

2. Quarterly demand for a new industrial component shows a clear, steady upward trend. A planner using single exponential smoothing sees the forecast lagging below actuals every period. The appropriate change is to

• A. keep single smoothing but halve α to shrink the forecast lag
• B. move to double exponential smoothing (Holt's method) for the trend
• C. switch to a simple moving average that weights all history equally
• D. keep single smoothing and treat the persistent lag as random noise

3. A planner finds that one month in the demand history includes a one-time factory shutdown that cut shipments to near zero. Before building the baseline forecast, she should

• A. leave the month in place, since models need completely unaltered history
• B. drop the whole product line and forecast it qualitatively instead
• C. flag or adjust the one-time event before building the baseline forecast
• D. treat the shutdown month as the start of a permanent demand decline

4. A planner's weekly forecast for a SKU was 800 units, but actual demand came in at 820 units. Using the standard convention, what is the forecast error, and what does its sign tell her?

• A. −20 units; demand was over-forecast by 20 units
• B. −20 units; demand was under-forecast by 20 units
• C. +20 units; demand was under-forecast by 20 units
• D. +20 units; demand was over-forecast by 20 units

5. The agreed demand forecast for next quarter is 10,000 units, but confirmed plant capacity is only 8,000 units. The planner's correct move in aligning supply with demand is to

• A. revise the forecast down to 8,000 so it matches plant capacity
• B. hold the 10,000-unit forecast and hope demand softens on its own
• C. publish a compromise forecast of 9,000 units for the quarter
• D. keep the demand forecast intact and close the gap on the supply side

6. A planner tests smoothing constants of 0.1, 0.3, and 0.5 on the same demand history, computing the mean squared error for each trial. The defensible way to pick the constant is to

• A. pick the middle constant, since extremes tend to overfit history
• B. pick the largest constant, since it adapts to change fastest
• C. pick the constant whose trial produced the smallest MSE
• D. pick the constant that drives the tracking signal to exactly zero

7. A category manager must compare forecast accuracy across two product lines — one selling thousands of low-cost units, the other dozens of high-cost units. The most suitable error measure is

• A. MAPE, because a percentage error is comparable across product scales
• B. MAD, because it expresses errors in the same units as demand
• C. MFE, because it cleanly separates bias from random variation
• D. total summed error, because raw totals are simplest to explain

8. A planner computes the tracking signal for her forecast model and gets +5.3. Following the rule of thumb, she should conclude that

• A. the model is likely off track and should be investigated or adjusted
• B. the model is working correctly and needs no further attention
• C. the forecast should be raised, since demand is biased upward
• D. a new forecasting method must be adopted immediately

9. Two forecasting models ran on the same SKU. Model X has an MFE near zero and a MAD of 18; Model Y has an MFE of −4 and a MAD of 9. Which model is performing better, and why?

• A. Model X, because an MFE near zero proves its forecasts are accurate
• B. Model X, because its larger MAD shows it tracks demand swings better
• C. Model Y, because its negative MFE cancels out Model X's errors
• D. Model Y, because its smaller MAD means smaller average absolute errors

10. Over the last four weeks, a product's mean forecast error (MFE) averaged −15 units per week. The demand planner's best first step is to

• A. raise the baseline forecast, since the model is under-forecasting demand
• B. look for systematic bias, since the model tends to over-forecast demand
• C. widen the tracking-signal limits, since the forecast errors are large
• D. change methods at once, since the MAD must be climbing as well

11. Demand for a product has started shifting rapidly, and the planner wants the exponential-smoothing forecast to react more strongly to the most recent actual demand. She should

• A. raise α toward 1, giving the latest actual demand more weight
• B. lower α toward zero, which makes the forecast more stable
• C. freeze α and feed a longer demand history into the model
• D. switch to a simple moving average with a longer window

12. After a period in which actual demand exceeded the forecast, a colleague argues the next single-exponential-smoothing forecast should be set below the old forecast to "compensate" for the overshoot. What is wrong with this reasoning?

• A. The smoothing constant should be raised instead of lowered
• B. SES ignores the most recent error by design
• C. The forecast should be frozen until the bias disappears
• D. The forecast adjusts toward the error, moving up

Sources cited in this excerpt

1. APICS CSCP Exam: CSCP ECM Version 5.0 Preview (official, ascm.org). ASCM (APICS dba ASCM), effective 2022-03-31. https://www.ascm.org/globalassets/documents--files/learning--development/cscp-v5.0-ecm-preview-.pdf
2. APICS Certifications and ASCM Certificates comparison chart (official, ascm.org). retrieved 2026-09-22. https://www.ascm.org/globalassets/ascm_website_assets/docs/credentials/comparison-chart.pdf
3. ASCM South Central Texas Chapter — CSCP. current as of 2026-09-22. https://sctx.ascm.org/content.php?page=CSCP
4. Measuring Forecast Accuracy: Approaches to Forecasting - A Tutorial. NC State Supply Chain Resource Cooperative, n.d. https://scm.ncsu.edu/scm-articles/article/measuring-forecast-accuracy-approaches-to-forecasting-a-tutorial
5. 6.4.3 What is Exponential Smoothing? (NIST/SEMATECH e-Handbook of Statistical Methods). National Institute of Standards and Technology / SEMATECH, n.d. https://www.itl.nist.gov/div898/handbook/pmc/section4/pmc43.htm
6. 6.4.3.2 Forecasting with Single Exponential Smoothing (NIST/SEMATECH e-Handbook of Statistical Methods). National Institute of Standards and Technology / SEMATECH, n.d. https://www.itl.nist.gov/div898/handbook/pmc/section4/pmc432.htm
7. T.2.5.2 - Exponential Smoothing (STAT 501 Applied Time Series Analysis). Penn State Eberly College of Science, n.d. https://online.stat.psu.edu/stat501/book/export/html/1001