Do you demand forecast as an e-commerce brand owner or a retail business owner? Do you check the sales from the previous month and add some numbers? Do you use a spreadsheet and tons of AVERAGE formulas? Or are you just guessing? (No, we are not being funny.) 85% of Indian SMBs (Revenue<₹50L/month) use a spreadsheet for their demand forecasting and are losing money every month.
Cost of bad forecasting
Overordering: By ordering too much inventory, you end up wasting your cash. This inventory can be stored in your warehouses for months or it can get wasted as it is perishable.
Underordering: This results in stockouts. You end up losing sales, and your customers end up buying from your competitors. For e-commerce brands, a stockout during a sale event can cost you more than your entire marketing budget.
How to do it right
To achieve the right balance, you need to order the quantity you expect to sell plus a certain amount of cushion. This can be achieved if we can answer not just "how much" but also "how confident" and "what could go wrong".
Why is ARIMA better than your spreadsheet?
ARIMA is an AutoRegressive Integrated Moving Average. It is a statistical technique for forecasting time series data. It helps you to look at the hidden patterns in your data that you wouldn't be able to see just by observing it.
- Trend: Is the demand increasing or decreasing?
- Seasonality: Is the demand higher in December?
- Autoregression: How much of last month's sales predict this month's sales?
- Noise Vs Signal: Which of the changes in our data are noise and which of them are signals?
A simple average done in a spreadsheet can't differentiate between months and weights them equally. ARIMA can weigh recent months more heavily, find a trend and predict a confidence level. The result is a forecast with a confidence score. Not just a number but a range.
Good forecasting should look like this
For every product in your catalog, you should be able to forecast:
- Forecast value: How much is your product expected to sell next month?
- Confidence score: How confident are we with this forecast (0-100%)
- Trend tag: Rising, Falling, Stable or Volatile
- Risk indicator: Stockout risk, excess inventory risk or stable demand
- Story: A simple explanation of what is happening and what you should do for each product
E.g.
Herbal Shampoo
Trending Up 86% confidence
Forecast 220 units next month (last month sold 195).
Increase stock 15% to avoid stock out
Good. An excel sheet with "AVG: 170" is not.
How to do forecasting for your business
Step 1: Prepare data in CSV in this format
product_id: Unique numberproduct_name: Human readable product namecategory: Category of the productsales_m1tosales_m6: Monthly sales from older to newtrend_score: A number between 0 and 1 that indicates the strength of the trend in your sales history
A minimum of 6 months of sales data should be available at least. Having more will help provide accurate prediction.
Step 2: Pick a model
- Linear Regression: A model that works really well for simple use cases. Fast and good for products that don't see huge seasonal changes in sales
- ARIMA: A model that works well for products that have some form of trend or seasonality
- Both models combined: The best results. Use ARIMA, and if it's not confident, use linear regression
Step 3: Make sense of the forecast
- Trend up with high confidence: Increase the stock to avoid stock out
- Trending down with high confidence: Offer promotions to reduce inventory
- High volatility: Increase safety stock and re-evaluate frequently
- Low volatility: Maintain current orders
Try it out!
I built Demand Oracle, a tool that does this:
- Upload your sales data as a CSV file
- Get ARIMA model and Linear Regression model based sales forecast for all your products
- AI writes a narrative for each product and tells you what you should do
- Get a dashboard to see the history of all the runs of the tool in one place
It's pay per use. Credits don't expire. Don't have to use any excel sheets or anything.
You can try it out here: 4ugusta.dev/tools/forecast
The issue with enterprise solutions
Enterprise solutions use tools like SAP, Oracle, Prediko to forecast. All these tools cost anywhere from 50 to 500 USD per month. Most small businesses in India don't have that kind of money in their budget.
"Spreadsheet guessing" is where most of the money is lost. If you're in charge of managing 50+ SKUs and making ordering decisions based on your hunches, you're losing a bundle of cash each month. The tools to do it right are available and they're affordable now. Use them!