While the StockIQ forecasting algorithm does a good job of creating forecasts, it will not be right in all scenarios. This is when you as the forecast manager must decide how to make changes.
Common examples of when StockIQ might get it wrong because we don't have the whole picture:
- One-Time Changes
- Sales is running a promotion or there is some other 1-time event
- Large, permanent shifts in demand
- There is a new customer or distribution channel coming on line.
- E-Commerce trends, viral behavior or crowdsourced links lead to huge demand increases
- You have lost a customer
- Industry Trends / Reversal of recent trends
- Business is trending generally up or down across the board
- Other industry factors or leading indicators
- New Items
- This is a brand new item or the item is becoming hot/trendy/viral.
- Subtle / Muddy Seasonality
- Seasonality is unclear on a month-by-month basis, but seems to have overall Winter/Spring/Summer/Fall trends
Below are some strategies for what to adjust and when:
One-Time Changes (special events or promos)
For this, often the best solution is to create an event to explain away the unusual situation, either when that has already occurred, or it is in the future.
Large, Permanent Demand Shifts
Demand Increases
If you expect that a change in demand will be permanent, such as a brand new customer, distribution channel, or some viral effect in an eCommerce channel (e.g. Amazon Rankings) you have two options, in increasing order of automation:
- Put in a manual forecast for the next 3-6 months, and set auto-forecasting to resume at that time. Check the "Continue Auto Forecasting After Resume Date" checkbox.
- Use the Model Adjustments feature in Forecast Manager. This will create events that adjust the stat model up to your new expected level of demand on the start date you specify. See the Model Adjustments topic for more discussion
- Enable Bias Detection on the StockIQ algorithm, and see if it can discover the change for itself. Even if bias detection shifts the algorithm to high volatility, we will still lag large/rapid changes by around 2-3 months.
- NOTE: You can also manually enable "High" volatility in the StockIQ algorithm, but this can create long-term forecasts that are far too reactive if left active for too long
TIP: For the specific case of an item's demand going stratospheric due to some sort of internet viral/crowd-sourced reason, the right answer is almost always to put in a manual forecast for a few months based on what you know about how permanent (or not) this viral behavior may be.
Demand Decreases
When you have lost a customer, you can use the Lost Customer (if they are no longer buying ANYTHING from you), or the Lost Customer Item screens in System Configuration to tell StockIQ. StockIQ will then create events to "adjust out" the fact they will no longer be purchasing from you.
Industry Trends
If the adjustment is not nearly as drastic, but you just want to put some slight adjustment on the forecast, you can use the Edit Forecasts window. Using this, apply your percent change up or down for however long you like. When you save, StockIQ will ask you if you want to resume auto-forecasting the period after the final period you adjusted.
If StockIQ is seeing a consistent upwards or downward trend, we will follow that trend. If an item has been on a decreasing trend for the last six months, we will follow that trend downward, sometimes resulting in the future forecast for an item being zero. If you believe this is inaccurate, you can try:
- In Forecast Manager, Algorithm Settings, set the StockIQ Algorithm volatility to "Auto" and enable "Very Low Volatility" OR just manually select "Very Low Volatility" --> This will slow how quickly StockIQ follows this trend
- If the current month is showing a recovery in activity for the product, enable the Current Period Prediction, and the new point will be incorporated into the forecast.
New Items
Very often, on new items, StockIQ will not have enough history to work with. See the Forecasting New Items topic. The New Forecast Wizard can be helpful here, as well as taking seasonality from another established item, or ideally, setting up a replacement/supercession using the ItemSiteReplacements screen.
Subtle Seasonality
If the seasonality for your item does not have strong enough correlation for StockIQ to calculate it on its own, then you can force the "Highly Seasonal" mode of the StockIQ algorithm by setting "highly seasonal: Yes" in the seasonality settings.
If your item is part of a group of items that all behave similarly, using the Seasonality Index Type "Hierarchy Level", and then selecting the product's group or family can give you a good representative seasonality curve for the item to use.