All professions

Business and operations

E-Commerce Operations

The problem

The assistant gives an online shop owner a full market overview: named competitors, suppliers, market growth in percentages. Several elements turn out to be invented, and the overview was about to go into an investor presentation.

30 courses · 5 specific to this profession

A process from your work

How long does a buyer's return or refund request take you?

  1. Step 1The tool reads the chat and the order, and proposes where the problem lies and how to solve it.
  2. Step 2You compare the solution with the order and with the shop's return policy.
  3. Step 3What the buyer is promised is your decision.
  4. Where the tool goes wrongThe proposed solution may not match the order or the return policy.
  5. The verification stepOpen the order and find the rule in the return policy that allows the solution.
8.2% faster to identify the issue, in a randomised field experiment. Source: Field experiment on a large online marketplace, working paper, July 2026.
More: what was measured and the full check

What changed, as measured

Randomised access to the assistant cut the time to identify the issue by 8.2%, about 3.2 seconds, and the length of the chat by 1.1%.

Who was measured: After-sales chat agents of a large online marketplace in China, handling returns, refunds and exchanges, randomised to have the assistant or not.

The caveat: A small average effect; the best agents saw their service quality decline. A working paper, with authors from the platform.

The check, step by step

  1. Open the order and compare the item, the date and the amount with what the draft says.
  2. Find the rule in the return policy that allows the solution; without a rule, the solution is not offered.
  3. Do not promise a deadline or refund amount that does not follow from the order and the policy.
  4. When the case fits no rule, pass it to the person in charge, not back to the tool.

The course covers drafting, editing and checking text with AI, with no invented facts and no personal data in public tools.

We can help you with this process: “E-commerce AI: Documents and communication”, 8 hours.

A documented failure pattern

The thermal bottle: “put it in the dishwasher without a worry”

  1. What the tool doesA product description an AI tool drafts for an online shop's operator from the supplier's specification sheet, to be published on the product page.
  2. Where the tool goes wrongThe draft promises the bottle is dishwasher safe, but the supplier's sheet says nothing about cleaning.
  3. The verification stepFind each product attribute's line in the supplier's sheet; remove any attribute without one.
Source: Amazon researchers, in Proceedings of the Seventh Workshop on e-Commerce and NLP, 2024, and 3 more.
More: why it is easy to miss and the full check

Where the tool goes wrong, and why it is easy to miss

Once published, the dishwasher claim becomes a product feature: the customer relies on it, and the shop answers for it, not the tool. The supplier's sheet says nothing about cleaning: the tool has filled the gap with what is often written for bottles of this category. The claim sits inside a benefit sentence ("easy to clean"), where the reader expects sales copy rather than technical data, so it reads as a given and not as a new claim. A check of the figures does not catch it: the 500 ml, the double-walled stainless steel, the 12 and 6 hours are all copied correctly.

The check that catches it

  1. Put the supplier's sheet next to the draft and underline in the draft every product attribute: size, material, duration, use, care, compatibility.
  2. For each underlined attribute, find the matching line in the supplier's sheet. Anything without a line is removed from the draft or sent to the supplier as a question.
  3. Read benefit sentences ("easy to clean", "suitable for") with particular care: that is where care and use instructions the supplier never gave slip in.
  4. Keep the supplier's written answer (an email or an updated sheet) with the listing, so it is at hand when a return or a complaint arrives.

Where the failure pattern is documented: A study by Amazon researchers at the ECNLP 7 workshop (May 2024); an arXiv preprint (October 2025); guidance from ASA/CAP, the UK advertising regulator (August 2023); Amazon's documentation of its AI listing tools (September 2024).Sources: Amazon researchers, in Proceedings of the Seventh Workshop on e-Commerce and NLP, 2024; arXiv, 2025; Advertising Standards Authority and Committee of Advertising Practice, 2023; Amazon, 2024.The case above is ours, written for work in Albania; the laws it names were checked against the official text.

What it looks like

The supplier's sheetThermal bottle 500 ml. Body: double-walled stainless steel. Lid: plastic. Keeps temperature: 12 hours cold, 6 hours hot.

The AI tool's draftA 500 ml thermal bottle in double-walled stainless steel, keeps drinks cold for 12 hours and hot for 6. Easy to clean: put it in the dishwasher without a worry.

Translated from the Albanian original.

Specialist courses for this profession

What you leave with

The market-overview verification table puts each competitor, supplier and growth percentage on its own row, with the register or official statistic that proves it, or the reason it was deleted. It is opened before the overview goes into an investor presentation.

Courses that apply to every profession

25 more courses are the same for every field: the foundations of AI, checking its answers, agents, and what to do when you get stuck.

E-Commerce Operations

The catalogue lists all 30 together, with filters for format and time.