Inventory how-to

Why simple forecasts often beat complex ones

Simple forecasts based on recent actual sales are easier to trust, easier to check, and react faster to real demand changes than complex models with many inputs.

Short answer: Simple forecasts based on recent actual sales are easier to trust, easier to check, and react faster to real demand changes than complex models with many inputs. Complex models often overfit old patterns and quietly break when something changes, and nobody notices until stock runs out or piles up.

Complexity hides errors

A forecast with many inputs (trend lines, multiple seasonal layers, external factors) has more places for a bad assumption to creep in. When the forecast is wrong, it is hard to tell which input caused it. A simple average of recent daily sales is easy to audit: you can look at the raw numbers and see if it makes sense.

Recent sales are the best signal you have

For most small and mid-size stores, what sold in the last few weeks is a better predictor of next week than a complex model trained on a year of noisy history. Demand shifts with trends, weather, social posts, and stock-outs. A short, recent sales window picks up these shifts quickly. A model weighted heavily on long history reacts slowly and can keep recommending quantities based on a demand pattern that no longer exists.

Overfitting is a real risk

Complex forecasts can fit past data very well and still predict the future badly. This happens when a model learns noise, like one unusual week, as if it were a repeating pattern. Simple methods are less likely to chase noise because they do not have enough parameters to memorize it.

Seasonality still matters, just add it carefully

The fix is not to ignore seasonality, it is to add it only once you have enough history to trust it, and to keep the base calculation simple. A plain recent-sales average with a seasonal adjustment layered on top, once you have about a year of data, is usually enough. Anything more complex rarely earns its keep for a single store with a few hundred SKUs.

Step by step

  1. Use a recent sales window (2 to 4 weeks) as your default forecast for each SKU.
  2. Only add a seasonal adjustment once you have at least a year of sales history for that product.
  3. Recalculate often rather than trying to build one perfect long-term model.
  4. Sense-check any forecast number against what you actually sold last week.
  5. Flag and manually review new products or items with erratic sales, since no simple average works well for these.
  6. Resist the urge to add more inputs just because they are available; each one adds a new way to be wrong.

Worked example

A store sells a ceramic mug and looks at the last 28 days: 56 units sold, so daily average is 2 units. A complex model using 12 months of data, including a one-off influencer spike in March, forecasts 4.5 units a day because that spike skews the average upward. The simple 28-day forecast of 2 units a day turns out to match actual demand far better over the next month, because the spike was a one-time event, not a repeating pattern.

How SteadyShelf helps

SteadyShelf deliberately uses a simple approach: a 28-day sales window by default, with seasonality only layered in once a SKU has about 13 months of history. This matches the point above, it will not build complex multi-factor models, so very erratic or brand-new products still need your own judgment.

$29/month flat, unlimited products and suppliers, 30-day free trial. Run the numbers now with the free reorder point calculator.

SteadyShelf is launching on the Shopify App Store. Join early access to get the install link first.

Related guides

All inventory how-to guides ยท Inventory glossary

FAQ

Does a longer sales history always make a forecast more accurate?

No. Longer history can include outdated demand patterns, like a discontinued promotion or a one-off spike, that drag the forecast away from current reality. More data only helps once it is relevant data, and it works best for adding seasonality rather than replacing a recent-sales baseline.

When should I override a simple forecast manually?

Override it when you know something the numbers do not, such as a planned promotion, a supplier price rise, a product being discontinued, or a known seasonal event with less than a year of history behind it. Simple forecasts are a strong default, not a rule that overrides your own knowledge of the business.

Let SteadyShelf do this math for every SKU

Order-by dates per supplier, seasonal forecasts and draft POs for your Shopify store. $29/month after a 30-day free trial.

Get early access