How should new products rank before they have any sales history?

The cold-start problem in product ranking

Every signal your ranking uses, click-through rate, conversion rate, revenue per view, is a historical metric. A new product has no history, so the algorithm treats it as the worst product in the catalog. This is rational for the algorithm and terrible for the business: launches die quietly at the bottom of page four, and the merchandising team concludes the product failed when it was never really tested.

The problem compounds for brands that launch frequently. If every new product needs manual intervention to get visibility, the merchandising team becomes a bottleneck and launches slip. The goal is a system that gives every launch a fair, automatic trial and then gets out of the way, promoting winners and demoting losers on evidence.

Launch boosts: how much and for how long

A launch boost should be big enough to generate a statistically meaningful trial and short enough that a failing product cannot hide. In practice, that means placing the product in a visible but not dominant position, page one but below proven bestsellers, for one to three weeks depending on your traffic. The boost buys impressions; the product has to convert them.

Calibrate the boost to your traffic. A high-traffic store can evaluate a launch in days; a low-traffic store needs weeks to accumulate enough views. Set the duration from your numbers, not from a rule of thumb: decide in advance how many product-page views constitute a fair trial, and end the boost when the product hits that number. Time-based cutoffs are simpler; view-based cutoffs are fairer.

Signals you can use before sales exist

Sales history is not the only signal. Category affinity, attribute similarity to proven sellers, price positioning, margin, and inventory depth all exist on day one. A new product that shares attributes with your bestsellers deserves a better starting position than one in an unproven category. Supplier and brand performance data can inform the prior too, if you track it.

Early behavioral signals arrive fast. Click-through rate on the product card, add-to-cart rate, and time on page accumulate within days and predict conversion better than demographics or attributes. Build the boost logic to listen to these leading indicators: a product with strong early engagement earns an extended trial, while one that shoppers ignore should lose its boost early. The system should be impatient with failure and patient with promise.

Knowing when to pull the plug

Every launch boost needs a kill criterion defined before it starts. The simplest version: if the product's conversion rate after the trial period trails the category average by a wide margin, the boost ends and the product ranks on its merits. Write the criterion down and automate it, because manual review creates exactly the bottleneck you were trying to avoid.

Pulling the plug is not the same as giving up. A product that fails its launch trial might succeed with better imagery, a different price, or placement in a different collection. The ranking system did its job by generating clean evidence; now merchandising can act on it. The failure mode to avoid is the zombie boost: a launch lift that nobody removes, quietly subsidizing a weak product for months while better products sit below it. Audit your active boosts quarterly and kill the ones that stopped being trials long ago.

Should new products outrank proven sellers during the boost?

No. The boost should buy visibility, not dominance. Placing an unproven product above your bestsellers taxes your most valuable traffic on a gamble. Page-one placement below proven winners gives the launch a fair trial without risking the revenue your bestsellers generate.

How long should a launch boost last?

Long enough to accumulate a fair trial of product-page views, which depends on your traffic. For most stores that is one to three weeks. Define the trial in views rather than days if you can, and always pair the duration with a performance criterion so strong launches earn more time and weak ones exit early.

What if the product flops after the boost ends?

Then the system worked: it generated evidence cheaply. A product that cannot convert its trial traffic is telling you something about the product, the price, or the presentation. Fix the underlying issue and relaunch with a fresh trial, or accept the verdict and allocate the visibility to products shoppers actually want.