Often when merchants install Candy Rack, they ask us two questions: what results I should expect from upselling and what is a good upsell strategy. For the latter one, I have already published our recommended upsell strategy guide.
And for the results, we knew internally what does "good" look like, but we lack some hard evidence we could share with the merchants. And thanks to AI development in the last few months, we were able to review the data and put together a solid insights that will answer questions like "Is 2% upsell take rate good?", "Are AI recommendations better than manual picks?", "How much does upselling increase my overall store revenue?".
Key takeaways
- The median upsell take rate (conversion) is 2.4% calculated as upsell sold vs. upsell viewed. The top 10% is close to 10%. That means if you spend enough time and create a "perfect" upsell offer, the take rate can be above 10%.
- Discounts don't matter that much – relevancy is more important. Upsells with no discount converts same as upsells with 1 – 20% off, so don't giveaway your margin for no reason.
- AI recommendations convert better (3.2%) than hand-picked offers (1.9%). We recommend testing Shopify vs. Gemini vs. Candy Rack AI to see which engines performs the best for your store.
- Showing more products in the offer increases the conversion. Offers with 1 product converts at 1.5% while offers with 3+ products converts at 3.1%. Our recommended number of products in one offer is 3 – 4.
How we analyzed the data
We used the same offer analytics that power the dashboard inside Candy Rack, for offers shown between January 1 and August 31, 2026. Candy Rack is installed on 5,000+ Shopify stores; 3,199 of them showed at least one offer during this period. Together they account for 39,007 offers, 218.6 million offer views (56.5 million on the product page pop-up) and 1.91 million upsell units sold. Offer-level tables include only offers with at least 1,000 views (8,854 offers), and all figures are medians across stores, so no single large store skews the result.
The analytics data records offer views, adds to cart and sales per store, offer and product. It contains no shopper-level or personal information, and results are reported in aggregate – no individual store is identified.
1. Take rate by placement
Candy Rack is one of the few upsell solutions on Shopify which covers the whole purchase funnel – from PDP to post-purchase. In total merchants can create offers on seven different placements.
But the conversion rate on each placement is different because buyer is in a different part of the shopping journey. In general the earlier in the journey the buyer is, the higher the upsell take rate is. That's why we recommend focusing on the pre-purchase upsells (i.e., on PDP and cart drawer).
* Checkout upsell offers placed inside the Shopify checkout. Require Shopify Plus plan (checkout extensibility).
** Embedded checkbox and button blocks combined. They count a view on every product page load, so their take rate isn't comparable to other placements.
On the other hand, post-purchase offers may not convert that well, but their gain is purely incremental, so it would be a shame not to use them.
2. Discounts
In the past, our recommendation was to offer a small discount, 10 to 15%, to increase the upsell take rate. But after doing this analysis, we have to update such recommendation as data doesn't support it.
The more important factor in the upsell conversion is the relevancy. When the offer is relevant and complementary, customers don't care much whether they get a discount or not. This is really important because keeping the 10 to 20% to yourself, has a huge impact on your margins.
To make sure we compare apples to apples, we have filtered out only the product page pop-up placement offers with 1,000+ views for the discount analysis.
Where discounts still make sense
The data above are stricly on the single item cross-sells and they don't include bundles. Using discounts on volume bundles (quantity breaks) and product bundles (i.e., different SKUs combined) is still recommended.
The difference is that bundle is typically much more expensive than the original product and customer need an extra motivation to take it. And that motivation are often savings.
From the merchant's point of view, the discount is justified by increasing the AOV and absolute margin. The actual discount amount depends on your product margins.
3. AI vs hand-picked product recommendations
Another very common question when it comes to upsell strategy is using the AI recommendations vs. picking offered products manually. The good news is that the AI recommendations are not just much faster to implement, but they also perform better!
The catch is that AI recommendations perform best at the PDP, i.e. when customer is adding a single product which is a clear anchor for the AI model. Further in the purchase flow, the performance of AI recommendations is decreasing and that's likely because the whole cart is an anchor and it's difficult for the AI engine to understand the exact customer's needs.
AI recommendations in Candy Rack have three options: Shopify Product Recommendations API (default), Gemini, and Candy Rack AI. We released the additional two models in July 2026, so the data are mostly based on the Shopify recommendations. In the future updates, we plan to split the AI into each specific model and see which actually performs the best.
To learn more about the engine logic and differences you can watch my video below or read this article.
Combined AI and manual recommendations
The good thing with Candy Rack is that can choose for which placement exactly to use AI/manual recommendations and you can even combine them together. So for example, the first product recommendation can be manual and the rest can use AI.
I typically advice merchants to enable AI recommendations for all products because that takes like 30 seconds and brings uplift immediatelly. And then start creating super tailored manual recommendations for best selling products.
4. Number of upsell products per offer
Another very important upsell strategy factor is the number of products in a single offer. That means how many upsell/cross-sell products the customer will see in each offer placement.
Quite logically the upsell take rate increases with the number of offered products because customers have more options to choose from. But I like to put a limit and avoid choice overload, especially because it can negatively impact the overall conversion rate (i.e., customer will drop off from the purchase funnel completely). I always recommend 3 to 4 upsell products per offer, same as Nick Kolenda.
In our data we see the take rate increasing even for 4–5 and 6–10 products, but the sample is relatively small and we also didn't measure the impact on the store conversion.
How to act on these numbers
Whatever upsell app you are using, compare these benchmarks with your data. If something is off, try to optimize it. You can also contact me directly via LinkedIn and I will be happy to personally review your upsell strategy and provide recommendations.
Here's also a quick checklist to go through:
- Compare your placement conversion with the take rate by placement table. If your offer performs below the median, try changing the offer product and/or messaging.
- Turn off discounts on your offers and monitor how it impacts the take rates. If you are using Candy Rack, you can launch a proper A/B test via Intelligems platform.
- Turn on AI recommendations for all products on product pages or cart drawer.
- Make sure all your offers have at least three upsell products, not less.
We plan to update this report regularly, especially with our volume bundle offers which we have launched in September 2026.
