Demand Forecasting Metrics Every D2C Brand in India Needs to Know

Imagine you're a fashion brand preparing for the upcoming festive season in India. You need to decide how much inventory to order for each of your products – from kurtas and sarees to shirts and shoes. Order too little and you risk stockouts and lost sales. Order too much and you're left with excess inventory that ties up working capital and requires costly markdowns.

The key to finding the right balance? Demand forecasting.

Demand forecasting is the process of predicting future customer demand for a product or service. By analyzing historical sales data, market trends, and other variables, businesses can estimate how much of each item they need to order to meet expected demand without over or under-stocking.

But demand forecasting is both an art and a science. Relying on gut feel alone is risky, especially as your product catalog grows. That's where demand forecasting metrics come in. These key performance indicators (KPIs) provide an objective, data-driven way to measure the accuracy and effectiveness of your forecasts.

As a direct-to-consumer (D2C) brand in India, understanding and tracking these metrics is crucial for optimizing your inventory, minimizing stockouts and excess stock, and ultimately boosting your bottom line. In this post, we'll explore the most important demand forecasting metrics you need to know.

1. Forecast Accuracy

Forecast Accuracy measures how close your forecasted demand was to the actual demand for a given period. It's usually expressed as a percentage, with 100% being a perfect forecast.

The formula for Forecast Accuracy is:

Forecast Accuracy = (1 - |Actual Demand - Forecasted Demand| / Actual Demand) x 100

For example, let's say you forecasted selling 1,000 units of a particular saree design this month, but you actually sold 1,200. Your Forecast Accuracy would be:

Forecast Accuracy = (1 - |1,200 - 1,000| / 1,200) x 100 
                  = (1 - 200 / 1,200) x 100
                  = 83.33%

The higher your Forecast Accuracy, the better you're able to predict demand and align your inventory accordingly. A lower accuracy suggests your forecasting method needs improvement.

To put this into perspective, consider two scenarios:

Scenario A (High Accuracy): You forecasted selling 1,000 kurtas and you actually sold 980. You had enough stock to meet demand without much excess.

Scenario B (Low Accuracy): You forecasted selling 1,000 kurtas but you actually sold 1,500. You likely ran out of stock and missed out on sales.

Clearly, Scenario A is preferable. By tracking Forecast Accuracy over time, you can assess the reliability of your forecasting methods and make adjustments as needed.

2. Mean Absolute Percent Error (MAPE)

While Forecast Accuracy measures your prediction's closeness to actual demand for a single item or period, MAPE measures the average accuracy across multiple forecasts. It's particularly useful for comparing the accuracy of different forecasting models.

The formula for MAPE is:

MAPE = (Sum of Absolute Percent Errors / Number of Forecasts) x 100

Where Absolute Percent Error is calculated as:

Absolute Percent Error = |Actual Demand - Forecasted Demand| / Actual Demand

Let's walk through an example. Say you sell three products (A, B, and C) and you want to compare the accuracy of two different forecasting methods (Method X and Method Y). Here's the data:

Product	Actual Demand	Method X Forecast	Method Y Forecast
A	1,000	1,100	950
B	500	450	525
C	2,000	2,200	1,900

To calculate MAPE for each method:

Method X:

Absolute Percent Errors = |1,000 - 1,100| / 1,000 = 10%
                          |500 - 450| / 500 = 10% 
                          |2,000 - 2,200| / 2,000 = 10%

MAPE = (10% + 10% + 10%) / 3 = 10%

Method Y:

Absolute Percent Errors = |1,000 - 950| / 1,000 = 5%
                          |500 - 525| / 500 = 5%
                          |2,000 - 1,900| / 2,000 = 5%
                             
MAPE = (5% + 5% + 5%) / 3 = 5%

In this case, Method Y has a lower MAPE (5% vs 10%), indicating it's the more accurate forecasting model overall.

MAPE is a go-to metric for comparing forecast models because it's scale-independent and easy to interpret. A MAPE of 10% means your forecasts are off by an average of 10%. Generally, a MAPE below 10% is considered highly accurate, 10-20% is good, 20-50% is reasonable, and above 50% is inaccurate.

3. Weighted MAPE (WMAPE)

One limitation of MAPE is that it treats the accuracy of all forecasts equally, regardless of the item's sales volume or value. This can skew your perception of overall forecast accuracy.

For example, say you have two products:

• Product A: You sell 1,000 units at ₹100 each.

• Product B: You sell 100 units at ₹1,000 each.

If your forecast for Product A is off by 10% and your forecast for Product B is off by 20%, the MAPE would be 15% ((10% + 20%) / 2). But this doesn't account for the fact that Product B, while selling fewer units, generates the same revenue as Product A and thus has a bigger impact on your business.

Weighted MAPE addresses this by weighting each forecast's accuracy by the item's relative importance (usually determined by sales volume or revenue).

The formula for Weighted MAPE is:

WMAPE = Sum of (Absolute Percent Error x Weight)

Using our earlier example, let's say Product A accounts for 60% of your sales and Product B accounts for 40%. The WMAPE would be:

WMAPE = (10% x 0.6) + (20% x 0.4) = 14%

This provides a more nuanced view of forecast accuracy that accounts for each product's business impact. Pay extra attention to the accuracy of your forecasts for high-volume or high-value items.

4. Forecast Bias

Forecast Bias measures whether your forecasts tend to be consistently higher or lower than the actual demand. It helps identify if you have a systematic over- or under-forecasting issue.

The formula for Forecast Bias is:

Forecast Bias = Sum of (Forecasted Demand - Actual Demand) / Sum of Actual Demand

A positive Forecast Bias indicates a tendency to over-forecast, while a negative bias points to under-forecasting. Ideally, you want a Forecast Bias close to zero, indicating your forecasts are not skewed in either direction.

For example, consider these forecasts and actual demand for a product over five months:

Month	Forecasted Demand	Actual Demand
January	1,100	1,000
February	1,050	1,000
March	950	1,000
April	900	1,000
May	1,200	1,000

The Forecast Bias would be:

Forecast Bias = (1,100 + 1,050 + 950 + 900 + 1,200 - 5,000) / 5,000
              = 200 / 5,000 
              = 4%

The 4% positive bias suggests a slight tendency to over-forecast demand for this product. Armed with this insight, you can investigate the reasons behind the bias and adjust your forecasting approach accordingly.

5. Forecast Value Added (FVA)

Forecast Value Added measures the incremental improvement in accuracy that a forecasting process provides over a simple naïve forecast. A naïve forecast simply assumes that the next period's demand will be the same as the last period's actual demand.

FVA is calculated by comparing the Mean Absolute Deviation (MAD) of your forecast to the MAD of a naïve forecast:

FVA = 1 - (Forecast MAD / Naïve Forecast MAD)

Where MAD is calculated as:

MAD = Sum of Absolute Deviations / Number of Periods
Absolute Deviation = |Forecasted Demand - Actual Demand|

An FVA greater than zero indicates your forecasting process is adding value over a simple naïve forecast. The higher the FVA, the more value your process is adding.

Let's illustrate with an example. Say you have the following demand data and forecasts for a product:

Period	Actual Demand	Your Forecast	Naïve Forecast
1	1,000	1,200	-
2	1,100	1,150	1,000
3	1,200	1,250	1,100
4	1,150	1,100	1,200
5	1,250	1,300	1,150

To calculate FVA:

Your Forecast MAD:

Absolute Deviations = |1,200 - 1,000| = 200
                      |1,150 - 1,100| = 50
                      |1,250 - 1,200| = 50
                      |1,100 - 1,150| = 50
                      |1,300 - 1,250| = 50

Your Forecast MAD = (200 + 50 + 50 + 50 + 50) / 5 = 80

Naïve Forecast MAD:

Absolute Deviations = |1,000 - 1,100| = 100
                      |1,100 - 1,200| = 100
                      |1,200 - 1,150| = 50
                      |1,150 - 1,250| = 100

Naïve Forecast MAD = (100 + 100 + 50 + 100) / 4 = 87.5

FVA = 1 - (80 / 87.5) = 8.6%

The 8.6% FVA indicates that your forecasting process is 8.6% more accurate than a naïve forecast, providing tangible value.

FVA is a great way to justify investments in forecasting processes and tools. If your FVA is low or negative, it's a sign that your current approach may not be much better than a simple guess and needs improvement.

Putting It All Together

Demand forecasting is a critical capability for any D2C brand looking to optimize inventory and maximize profitability. By understanding and regularly tracking these key metrics – Forecast Accuracy, MAPE, Weighted MAPE, Forecast Bias, and FVA – you can objectively assess the performance of your forecasting process and continuously improve it.

Remember, no forecast will ever be perfect. The goal is not 100% accuracy, but rather to be as accurate as possible given the information and resources available. By leveraging data, employing the right metrics, and using them to inform decisions, you can bring more science to the art of demand forecasting.

At Fiscal Flow, we specialize in helping D2C brands in India streamline their financial operations and make data-driven decisions. Our team of experts can help you set up a robust demand forecasting process, select the right tools and models, and interpret your metrics to drive continuous improvement.

From improving forecast accuracy to optimizing inventory levels and minimizing stockouts and overstock, we partner with you every step of the way. We can also help ensure your forecasting process aligns with financial planning and tax compliance requirements.

If you're ready to take your demand forecasting to the next level, contact us today. Let's work together to turn demand uncertainty into a competitive advantage for your brand.