"How much historical data do you need?"
As little as 8 weeks lets the system pick up basic weekly patterns, but quality improves a lot with more. With under 6 months, it can only see weekly rhythms, not the annual shape. With under a year, it hasn't yet observed parts of the calendar it hasn't lived through (if you onboarded in March, it hasn't seen your December yet), so it uses conservative estimates for those periods rather than guessing. A full year gives the best results, since the complete annual cycle is visible, and two or more years unlock holiday-specific effects.
"Does it account for holidays?"
Yes, but only when there's enough data. Holiday effects such as Black Friday and Christmas are included for customers with at least 2 years of history. The system looks at a three-day window around each holiday (the day itself plus one day on either side) and compares demand there to normal non-holiday demand. For newer customers, it focuses on the broader monthly and weekly patterns, because isolating individual holidays from limited data would be unreliable.
"What if my product's seasonality changes from year to year?"
The system gives more weight to recent data than older data, and each layer adapts at its own pace. Weekly patterns update fastest: change your opening hours and the weekly shape adjusts within a few weeks. Monthly patterns adapt over a few months. Holiday effects change most slowly, since they only happen once a year and the system needs to accumulate evidence across multiple years. The result tracks genuine shifts in demand while staying stable enough not to overreact to noise.
"Why does the system group my products together? Won't that be less accurate?"
For most products, grouping makes the forecast more accurate. A product selling one or two units a day produces data too noisy to extract a reliable seasonal signal, because the randomness overwhelms the pattern. Pooling related products reveals the shared rhythm clearly, and the system then applies it back to each product at its own scale. Only very high-volume products could benefit from individual modelling, and even they gain from the noise reduction that grouping provides.
"What happens if my products were out of stock for a while?"
The system detects out-of-stock periods and estimates what demand would have been, so a stockout doesn't create a false "low demand" signal. Those estimates are based on the product group's typical sales for the same weekday and month, so they reflect realistic demand levels rather than a simple average.
"Does it work for new products with no history?"
A brand-new product with no sales history receives a flat seasonal forecast (multiplier 1.0, meaning average demand and no seasonal adjustment). As it accumulates sales, the system begins detecting and applying patterns. If it belongs to an existing group that already has established seasonality, that group's pattern is applied as soon as there's enough data to compute a baseline for the new product.
"How far ahead does it forecast?"
The standard horizon is 90 days, about a quarter, which is enough for most pricing and inventory planning. The forecast is recalculated every two weeks, so the 90-day window rolls forward continuously.
"Can my seasonal forecast be wrong?"
Like any forecast, it reflects patterns learned from the past and can't predict genuinely unprecedented events, such as a pandemic, a viral moment, a new competitor entering the market, or your first-ever participation in a major sale event. It's designed to be conservative: without strong evidence for a pattern, it defaults to a flat forecast rather than an unreliable one. Multipliers are also capped between 0.01 and 3.0, so outlier or thin data can't produce extreme predictions. And bear in mind that past events unrelated to a product's demand are just noise: including them tends to make the estimates worse, not better.
"Is the seasonality the same across all my stores?"
If you operate multiple stores, the system can produce store-level multipliers. The underlying seasonal pattern is currently learned at the product-group level and then applied across stores, so the seasonal shape is shared while the scale, each product's average sales, can differ from store to store.