If you’ve ever walked into your favorite coffee shop craving an iced latte in a hot summer day, only to hear, “Sorry, we’re out of ice and oat milk”, you’ve experienced a bad forecast from the consumer side.
From the company perspective, that same bad forecast means:
- Lost sales and frustrated customers
- Extra rush orders to suppliers (higher costs)
- Idle staff waiting for ingredients to arrive
- Damaged brand reputation when customers stop coming back
Forecasting mistakes can set off a domino effect: wasted fresh products, angry distributors, and costly production shifts.
Choosing the right forecasting technique for the right situation can dramatically cut these risks.
The infographic highlights 7 demand forecasting techniques to guide you.
Here, I expand on each method with practical explanations and Excel tips so you can put them into action right away.
.
1. Moving Average
Concept: Averages past sales over a set period to smooth out fluctuations.
Data Needed:
- Historical sales data (weekly or monthly)
- At least 3–12 periods of stable data
Example: A bakery chain predicting weekly croissant sales based on the last 4 weeks.
When to Use:
- Demand is stable
- Minimal trend or seasonality (e.g., bottled water sales in a consistent climate)
Pros:
- Simple to calculate and understand
- Good for short-term stability
Cons:
- Lag effect; slow to react to sudden changes
Difficulty: Low/ Beginner-friendly
Excel Tip:
=AVERAGE(B2:B5)
Replace B2:B5 with the last 4 weeks’ sales. Drag down to forecast.
2. Exponential Smoothing
Concept: Weighs recent data more heavily to respond faster to changes.
Data Needed:
- Historical sales data
- Stable SKU definitions
Example: A coffee chain adjusting latte demand after introducing oat milk.
When to Use:
- Recent data reflects demand better than older data
- Minor trends, little seasonality
Pros:
- Responds faster than Moving Average
- Easy to implement
Cons:
- Requires selecting a smoothing constant (α)
- Still not great for strong seasonality
Difficulty: Intermediate
Excel Tip: Use Data → Data Analysis → Exponential Smoothing.
3. Triple Exponential Smoothing (Holt-Winters)
Concept: Adds trend and seasonality components to exponential smoothing
Data Needed:
- At least 2 years of historical data with clear seasonal cycles
- Consistent time intervals
Example: A smoothie truck forecasting high demand for strawberry smoothies in summer and pumpkin spice smoothies in the fall.
When to Use:
Data has both trend and seasonality (e.g., strawberry smoothies in summer, pumpkin spice smoothies in the fall
Pros:
- Handles both trend & seasonality well.
Cons:
- Requires parameter tuning (trend α, seasonality β, γ).
Difficulty: Advanced in Excel
Excel Tip:
=FORECAST.ETS(Target_Date, Sales_Range, Date_Range, 1, 1)
4. Linear Regression
Concept: Models the relationship between a dependent variable (sales) and an independent variable (e.g., temperature)
Data Needed:
- Sales history
- Related external variable(s)
Example: Predicting lemonade sales based on monthly average temperature.
When to Use:
- Clear relationship between demand and another factor
Pros:
- Easy to interpret
- Can incorporate external drivers
Cons:
- Assumes linearity
- Poor fit for complex patterns
Difficulty: Intermediate
Excel Tip: Data → Data Analysis → Regression
5. ARIMA
Concept: Combines autoregression, differencing, and moving averages for complex time series
Data Needed:
- Large, clean dataset
- No missing periods
- Stationary or transformed-to-stationary data
Example: Cheese export forecasts with shifting demand patterns
When to Use:
- Data shows autocorrelation
This means that in your time series, current values are statistically related to past values.
In other words, sales this month tend to be similar to or influenced by sales from previous months.
For example, if our strawberry smoothie truck sold 500 smoothies last July and 520 this July, it’s likely next July’s sales will also be in the same range.
The sales pattern is not random. It “remembers” past behavior.
- Trend and seasonality may shift over time
Pros:
- Highly flexible for many time series types.
Cons:
- Requires statistical expertise
- Not native to Excel
Difficulty: Expert, often needs R/Python
Excel Tip: Export to CSV → run ARIMA in R/Python → import results
6. Delphi Method
Concept: Forecasting via expert consensus, refined through multiple feedback rounds
Data Needed:
- Expert panel with relevant market knowledge
- Structured questionnaires
Example: Estimating demand for a new plant-based yogurt with no sales history
When to Use:
- New products
- Little or no historical data
Pros:
- Leverages deep market insight
Cons:
- Time-consuming
- Risk of bias
Difficulty: Moderate; more about coordination than math
Excel Tip: Use a table to collect expert forecasts each round and calculate averages
7. Neural Networks
Concept: AI models complex, non-linear relationships in data
Data Needed:
- Large, detailed datasets (sales, promotions, weather, social trends)
- Multiple influencing factors.
Example: Snack manufacturer predicting sales across SKUs, promotions, and countries
When to Use:
- Complex, multi-variable demand patterns
Pros:
- Captures non-linear relationships
- Adaptive to changing data
Cons:
- Requires big data and tech resources
- Can be a “black box”
Difficulty: Expert; needs AI tools
Excel Tip: Use Excel as a front-end for data prep, then model in Python/Azure ML
Actions to Take For Beginners
Start simple with the following actions:
1. Pick one easy method like Moving Average
2. Run it in Excel
3. Track your results
4. Keep your data clean
5. Watch your accuracy
6. Make small tweaks as you go
Quick wins now will build your forecasting muscle fast.
Conclusion
A wrong forecast has painful consequences.
Learn the methods.
Match them to the situation.
Don’t delay. Take action.
Better forecasts start today.
