Reporter's hands-on test of AI bill calculations during Double 11: Taobao, Doubao, Tongyi Qianwen, and DeepSeek frequently make errors in discount calculations

It's that time of year again for "Double 11." When deposit expansion, store coupons, category coupons, cross-store full-reduction discounts and other promotional rules all intertwine, this shopping carnival has long since turned into a mind-numbing "math olympiad."

An IT Times reporter first ran hands-on tests on the AI assistant embedded in the Taobao App.

In the "product selection" stage, the AI assistant performed notably well: it not only quickly understood the request but also rapidly filtered out several products from well-known major brands, with product quality matching the reporter's expectations. In terms of reading account information, the AI assistant also showed a relative advantage.

Shanghai consumer Ms. Lin (pseudonym) shared her test experience with the reporter: she bought some items on October 20 (the day for paying the balance of Double 11 pre-sales), and when she recalculated a few days later using the Taobao AI assistant, she found that the price given by the AI was more expensive than what she had paid before. After puzzling over it for several days, she discovered that the reason was not that the AI assistant had miscalculated, but that she had already used a "spend 600 yuan, get 80 yuan off" coupon. The AI assistant embedded in Taobao can indeed dynamically calculate prices based on the coupons currently in the user's account, something most other AI assistants cannot yet do.

The reporter selected multiple women's clothing and outdoor products and added them to the shopping cart. The Taobao page showed a combined payment price of 1,546 yuan, and the price initially read by the AI assistant was consistent with this. When the reporter asked it to "find a plan lower than 1,546 yuan, with split orders allowed," the AI assistant's logic began to fall apart. It suggested adding items to reach "spend 1,500 yuan, get 180 yuan off" or purchasing store shopping credit.

Even more absurdly, when the reporter requested a split-order operation, the AI assistant made multiple errors in succession. It miscalculated the total price of 1,546 yuan as 1,670 yuan because it misread the unit price of a certain item. In addition, it calculated a small order that actually exceeded 450 yuan as more than 170 yuan and suggested adding laundry detergent to reach the threshold. After the reporter corrected it multiple times, the AI assistant "suddenly realized" and repeatedly apologized and recalculated, but the "large order" it ultimately planned was still about 100 yuan short of meeting the 1,500 yuan threshold.

In cross-App testing, Doubao "lost its memory" and miscalculated again, and JD AI Shopping could not read the shopping cart

An IT Times reporter conducted cross-App testing and gave Doubao this test request: "I want to buy a cervical massager, personal care products and hygiene products on Douyin Mall. Due to coupon restrictions, I currently only have a 'spend 600 yuan, get 80 yuan off' coupon, and I hope to precisely hit the order amount so that the actual payment is 520 yuan."

At the very start of the test, Doubao exposed problems of outdated information and "memory loss."

On the first day of testing, it confidently claimed that Douyin had a "spend 300 yuan, get 50 yuan off" coupon. The next day, when the reporter asked it to recalculate according to "spend 600 yuan, get 80 yuan off," Doubao not only said it could not see the previous day's chat history and links, but also conflated the listed original prices of recommended products with their post-coupon prices, causing the total plan price to greatly exceed the requirement.

After multiple rounds of back-and-forth, Doubao recommended a personal care set priced at 34.93 yuan as a filler item. But when the reporter actually tried it, they found that this low price came from a "Doubao App channel exclusive 30% off subsidy coupon," and the rules for that coupon were extremely strict: it was limited to a single item and could not be used at the same time as Douyin Mall's "spend 600 yuan, get 80 yuan off" full-reduction coupon.

Once the reporter followed Doubao's suggestion and placed multiple items in a combined order, the channel subsidy immediately became invalid, the product price instantly rebounded to 49.9 yuan, and the final actual payment for the order soared to 734 yuan.

Within half an hour, the reporter communicated with Doubao multiple times before making it realize that it had "ignored the hard restriction that platform coupons are not interoperable," and it gave an extremely helpless fallback plan: have the reporter split the products into separate orders and place them separately in the Doubao and Douyin Apps. This compromise, which broke the logic of full-reduction discounts, completely deviated from consumers' original intention of "adding items to save money."

The reporter did not find an AI assistant within the JD App. After asking customer service, the reporter was told that it was necessary to separately download an independent App called "JD AI Shopping," and that there was no web version. After downloading and logging in, the reporter found that although the top of the App showed 99+ items in the shopping cart, when the reporter asked it to "calculate how to buy the items in the shopping cart most cost-effectively," the AI replied that it "temporarily could not directly see the specific products in your shopping cart" and required manual input of names or categories. The inability to synchronize basic data led to a poor user experience.

Shanghai consumer Susu (pseudonym) told an IT Times reporter that she planned to buy some women's clothing, beauty products and personal care products on Taobao, and then took a guide circulating online for adding items to reach thresholds in e-commerce livestreams and tried to have AI help calculate the optimal solution.

This guide was very detailed. It not only included multiple stacked discounts such as the 88VIP consumption coupon and beauty supplementary coupon, but also listed in detail whether various consumption coupons expand, how to choose expansion tiers, as well as the stacking order of balance red envelopes and complex dynamic threshold calculations. Under these rules, the total product price first needs to deduct pre-discounts, then the consumption coupon threshold is calculated based on the discounted amount, and only then are balance red envelopes applied. The logic is rather convoluted.

Susu sent this general set of full-reduction rules from the internet to three AI models—Doubao, Tongyi Qianwen and DeepSeek—along with the products she wanted to buy, and each large model gave completely different calculation plans and final prices. Compared with the actual payment price, the error was as high as 500 yuan. In fact, Susu's combined payment price on Taobao Mall was 2,500 yuan, while among these three AI models, some gave a minimum of 2,100 yuan, while others calculated 3,000 yuan.

Among them, Tongyi Qianwen, after repeatedly confirming with Susu that "products not marked with a consumption coupon cannot use that coupon," rechecked the rules and gave a payment plan split into two orders. However, after a round of calculations, the final estimated total price became 3,000 yuan. Faced with the result of "the more it calculated, the more expensive it got," Susu said helplessly: "The rules are too convoluted. No matter how much the AI calculates, it's not much use, and each one calculates a different price. It delayed me for more than an hour, and in the end I just paid 2,500 yuan directly as a combined payment."

The essence of the "Double 11" shopping festival should be to benefit consumers, yet now it has evolved into a complex math problem that confuses AI, which shows how much time and energy consumers have to spend on choosing products and making payments. When AI deviates in data reading, logical calculation and prediction of rule conflicts, it not only weakens its auxiliary value but also greatly diminishes the shopping experience. Platforms not only need to further optimize their algorithms, but also need to make discount rules simple and straightforward again.

This article comes from the WeChat public account "IT Times" (ID: vittimes), author: Mao Yu, published by 36Kr with authorization.