When Price Optimization returns a suggestion, it also tells you how sure it is. That confidence label is one of the most useful signals in your results: it's the difference between "we're confident this is the right price" and "treat this one with care." This article walks through each label and what it should mean for your review.
Where confidence comes from
For each product that is modeled, the system works out a cautious version and an optimistic version of the demand pattern, alongside its central estimate. The gap between those two is the uncertainty around the suggestion. When the gap is narrow, the system has a clear picture and confidence is high. When the gap is wide, the data is noisier and the suggestion deserves more scrutiny. It's essentially saying "we're 90 % confident the actual value is likely to fall between these two values".
That gap sets the starting label. The system can then lower it if there is another reason to be careful, for example when the underlying fit came out weak (e.g., modelling an individual data-poor product with group/category-level -estimate), when the product sold out often enough to muddy the sales signal, or when customers barely reacted to the price changes it did make. So a low label doesn't always mean there is too little data. Sometimes the data is there but hard to read.
The labels
Label | What it means | What to do |
| The model is confident; the uncertainty band around the suggestion is tight and the model fit was accurate. | Safe to trust and approve, all else being equal. |
| A reasonable estimate, but meaningful uncertainty remains. | Fine to use; sanity-check higher-value products. |
| The data is noisy or the fit is weak, so the suggestion is more of a rough guide. | Review before approving; consider waiting for more history. |
| The product had little signal of its own. Its estimate is generalised from products in the same group, or from its broader category. | Plausible, but there's little direct evidence. Review encouraged. |
| The product did change price, but demand barely moved in response. It's treated as price-insensitive rather than given a fitted price response. | Useful to know: this product's sales, based on current evidence, are driven by something other than price. |
| The product had no sales at all in the period the system looked at, so there's no basis to grade confidence. | Check whether the product is genuinely active and in stock. |
| No model was involved. The product was given a simple static price change instead of an optimized price. | Part of data gathering process; see Why Some Products Aren't Optimized. |
About borrowed
When a product has little signal on its own but belongs to a group and/or category of similar products, the system can still estimate a price for it by drawing on what its better-selling siblings reveal about how demand responds to price. It's like a new team member picking up the ropes from experienced colleagues rather than starting from nothing.
The system can borrow from two places: the product's own group, and, where product categories are defined, the broader category the product sits in. That second route is what lets a product with flat prices of its own still get a sensible suggestion.
That estimate is often far better than a blind guess, but it rests on someone else's evidence, not the product's own. So a borrowed suggestion is plausible rather than proven. Give it a closer look before approving, especially for higher-value items.
About inelastic
inelastic means something different from the labels above, and it's easy to misread as a problem. This product did sell at different prices, and customers carried on buying at much the same rate anyway. In other words, the system got a clear look at how demand responds to price, and the answer was: barely at all.
Rather than invent a price response the data doesn't support, the system anchors the product at its normal sales level. This is genuine information about the product, not necessarily a gap in the data. It usually means something other than price is driving sales, such as brand loyalty, convenience, or the product being a routine repeat purchase. Products like these often have room for a price increase, so they're worth a look.
About no_sales
no_sales means the product recorded no sales at all in the period the system examined, so there is nothing to grade. Before reading anything into the suggestion, it's worth checking the basics: is the product actually listed, in stock, and visible to customers? A product carrying this label often has a practical explanation rather than a pricing one.
About rule_based
rule_based means no model ran at all: the product or its broader group didn't have enough of its own price history to learn from, so it received a simple, conservative instruction instead of an optimized price. This is about data coverage, not a configuration problem. It's covered in full in Why Some Products Aren't Optimized.
The common thread
borrowed, no_sales and rule_based all come back to the same thing: not enough sales history or price variation, for the product or its group. The remedy is better data over time, not a settings change. As a product builds up its own sales at a range of prices, its confidence label tends to climb toward medium and high on later runs.
inelastic is the exception. It isn't a data problem at all. The system had what it needed and found that price simply isn't what moves this product, based on the current saleshistory.
For how suggestions are produced overall, see How Price Optimization Works. For why a suggestion can shift between runs even at high confidence, see Why Suggested Prices Change Between Runs.