Skip to content

AI in customer support

| Article: 5 min read | AI in e-commerce

AI in customer support means using language models and chatbots to answer customer requests: order statuses, delivery and return terms, routine questions. It works well on simple, verifiable requests backed by data from your systems, and poorly where a decision, empathy or an exception to the rules is needed. The main condition: the customer always has a way out to a human.

In plain words

Most requests to an online store repeat: "where is my order", "how long does Nova Poshta delivery to Kharkiv take", "how do I return an item". A bot connected to a knowledge base and order statuses answers these instantly. Harder cases (a damaged parcel, a warranty dispute, an irritated customer) are better handed to a person, ideally with the history the bot has already collected.

In its breakdown of agent systems, Anthropic explains why support suits agents well: the conversation needs outside data (customer profile, order history, knowledge base), the agent can take actions (issue a refund, update a ticket), and success is easy to measure. At the same time, autonomy means higher cost, and errors can pile up over several steps (Anthropic, "Building effective agents", 2024, source). What an agent is: AI agents.

What to give to AI

  • Order status and tracking. The data comes from the system, nothing needs to be invented.
  • Routine questions from the knowledge base: delivery, payment, exchange, working hours.
  • Help choosing from your catalog (provided the bot sees current specs and availability).
  • First-line information gathering: order number, photo of the damage, the gist of the issue, so a person doesn't start from zero.
  • Draft replies for an employee who checks and sends them. The least risky way to start.

What not to give to AI

  • Money decisions without limits: refunds, compensation, exceptions to the rules. Only with a cap and a human's confirmation.
  • Complaints, conflicts, sensitive situations (a payment failure, a wrong charge, an unhappy customer).
  • Legally significant answers: warranty, return terms in non-standard cases. The store is responsible for what the bot says. In the British Columbia tribunal's decision in Moffatt v. Air Canada (February 2024), the airline was held liable for its chatbot's wrong advice about a discount, and the tribunal rejected the argument that the bot is a separate entity: the chatbot is part of the company's website (McCarthy Tétrault, 2024). This is a Canadian body's decision and does not apply under our law, but it shows the logic that both customers and regulators may apply to your bot.
  • Answers the bot has no data for. It should say "I don't know" and hand over to a person instead of making something up.

What the data says

  • Customers want a way out to a human. In a 2026 Gartner survey, 87% of customers said it is essential that companies using generative AI offer the option to reach a live agent. 50% said the interaction gets easier when companies use generative AI. On their last contact, customers were three times more likely to use third-party AI services (ChatGPT, Gemini, Copilot) than company chatbots (Gartner via DestinationCRM, 2026). The sample and methodology are not given in this retelling.
  • Companies are in no hurry to replace people. A Gartner survey of 321 support leaders (September–October 2025): 20% cut agent headcount because of AI, 55% kept headcount while request volume grew, 42% hired specialists for AI projects (dialogue designers, analysts). Gartner estimates that by 2027 half of the organizations planning big AI-driven cuts will reverse them (The Register, 2025).
  • The Klarna example. In 2024 the company said its AI assistant did the work of 700 agents. In May 2025 the CEO said the investment should go into the quality of human support, and that customers should know: "a human will always be there if you want one" (CX Today, 2025). This is one example of a large company, not a pattern for everyone.

There is no recent data on how many Ukrainian shoppers accept store chatbots.

How to roll it out gradually

  1. Start with draft replies for an employee, then add bot answers to the 5–10 most frequent questions.
  2. Connect the bot to current sources: the knowledge base and order statuses. Don't rely on the model's general knowledge.
  3. Make an obvious "talk to a human" button and pass the conversation on with the context already collected.
  4. Set simple limits: refund amount, a list of topics the bot doesn't answer.
  5. Read real conversations every week: where the bot got it wrong, where the customer asked for a person.
  6. Track the metrics: share of requests resolved without a human, repeat requests on the same issue, customer rating. Response speed alone doesn't show quality.

Example (illustrative numbers)

The numbers are made up for the example. A store gets 600 requests a month. 240 of them are about order status and delivery. The bot closes 200 of those and hands 40 to a person. Managers free up about 30 hours a month (200 requests × 9 minutes). But if 20 of the 200 answers were inaccurate and customers came back, the real saving is smaller. So read the answers selectively.

Common mistakes

  • Making the bot the only channel with no way out to a human.
  • Not connecting it to current data and letting it answer "off the top of its head".
  • Measuring only speed and the share of "closed" requests.
  • Giving the bot the right to refund without a limit.
  • Not telling people that a bot is answering.

FAQ

Can an AI bot replace support? Partly. The bot handles simple questions well. Gartner's 2025 data shows that most companies are not cutting headcount because of AI, and in 2026 most customers demand access to a human.

What should an online store give to AI in support? Order statuses, routine questions from the knowledge base, data gathering for the manager, draft replies. Leave complaints, money and exceptions to people.

Is the store liable for a chatbot's mistakes? In the well-known 2024 decision in Canada (Moffatt v. Air Canada), yes: the bot is treated as part of the website. For your jurisdiction, consult a lawyer.

Do I need to tell the customer they are talking to a bot? It is good practice for trust. Legal requirements depend on the country, so check them against current law.

How do I measure the benefit of AI in support? Compare the share of requests closed without a human with repeat requests and customer ratings, not just first-response speed.

Sources

Want to check this on your store?

AI automation

An AI assistant checks the store every day, suggests fixes and makes them after your «Yes».

About the service →

Share

Related

  • Agentic commerce

    | Article: 5 min read

    Agentic commerce is shopping in which an AI agent acts on a person's behalf: it searches for and compares products, and within the permission it has been given it can also place the order and pay. For a store, this means part of the "visitors" will be…

  • Technical debt

    | Article: 6 min read

    Technical debt is the future cost of fixing and reworking that arises when software was built fast and without room for change. Like financial debt, it lets you move faster today, but with interest: every next change costs more and takes longer.

  • Vibe coding

    | Article: 4 min read

    Vibe coding is building programs where a person describes the task in ordinary language, AI writes the code, and the person accepts the result almost without checking it and goes by whether it "looks like it works", adding new wishes in the chat. For quickly…

  • How an online store team uses AI every day

    | Article: 2 min read

    AI in a store team's daily work is a set of small, repeating tasks in which a language model prepares a draft and a person checks it and decides. The value comes not from separate "smart" features but from simple habits assigned to specific people.