AI Chat Agent Use Cases 2026: Six Jobs from Sales and Support to Onboarding, with Limits and a Cost Calculation

In short: An AI chat agent on your website pays off where many identical questions arrive and the answer is in your documents: pre-sales, first-line support, appointment booking, lead capture, shop questions and onboarding. Through integrations it books appointments, looks up orders or creates contacts in your CRM, and your team takes over difficult conversations in the Live Inbox. Whether it pays depends on the share of chats the agent resolves without a person. With EchoCall, 1,000 sessions cost €400 a month, and you pay for every session, not just the resolved ones.

What customers expect from a chat agent

With questions about an online purchase, 62 % of online shoppers in Germany would most like a quickly reachable human, 36 % chatbots. 86 % were satisfied with human contact, 50 % with chatbots. [1]

In a Gartner survey of 3,566 customers, 50 % said interactions are easier when companies use generative AI. Yet 87 % consider it essential to be able to reach a human as well. [2] The same survey shows that customers use third-party AI tools about three times as often as company chatbots, whose use is statistically unchanged since 2022. [3]

57 % of companies in Germany with 20 or more employees use AI, up from 36 % a year earlier, most often in customer contact, for example in handling enquiries (72 %). [4]

Three rules follow for every use case below: answer quickly and correctly, keep the route to a human open, and measure whether the agent really resolves.

Six use cases at a glance

1. Pre-sales advice

Someone about to buy will not wait until the next morning. An agent explains at 10 pm which plan fits or whether two products work together.

• What the agent does: answers questions on products, plans, delivery times and compatibility, compares variants and points to the right page or offers a callback.
• What data it needs: product pages, price lists, data sheets and a list of typical objections with your answers. You read in the website by crawl with a refresh interval, data sheets as PDF.
• Where the limit is: individual discounts, negotiation, binding commitments. What the agent says is attributed to you. In Moffatt v. Air Canada, the airline's chatbot had wrongly explained its bereavement fare rules. The tribunal in British Columbia held the airline responsible for all information on its website and awarded the customer 650.88 Canadian dollars in damages. [5]
• How you measure success: the share of chats followed by an enquiry, booking or purchase, and the questions the agent could not answer. That list shows what the knowledge base lacks.

2. First-line support, with handover to the Live Inbox

The classic use case: recurring questions about accounts, usage, invoices or delivery. Which customer service problems can an AI agent solve? The ones whose solution is written down somewhere.

• What the agent does: answers from help articles and guides, asks back when a request is unclear and, if needed, creates a ticket in Zendesk, Freshdesk, ServiceNow or Jira. It can also notify your team in Slack or Microsoft Teams, as an action, not as a chat channel for customers.
• What data it needs: help articles, guides, known errors with fixes, your rules on deadlines and goodwill.
• Where the limit is: angry customers, special cases, anything that needs judgement. Then your team takes over in the Live Inbox. The danger is an invented rule. In April 2025 the support bot of software maker Cursor claimed a subscription was limited to one device for security reasons. No such rule existed, and the company apologised publicly. [6] So write in the prompt that the agent only says what is in the knowledge base and otherwise hands over.
• How you measure success: the share of chats that end without handover and do not return days later with the same question. Intercom cites an average of 76 % resolved for its agent Fin, HubSpot 70 % for its Customer Agent. [7] [8] These are vendor claims without a published method. Plan with your own figures.

3. Appointment booking through calendar integrations

• What the agent does: asks for the reason and preferred time, checks free slots and books directly in Google Calendar, Outlook, Calendly or Cal.com. A confirmation can go out via Gmail or SMTP.
• What data it needs: calendar access, appointment types with duration, buffer times and rules such as "first consultations on Tuesdays only".
• Where the limit is: appointments that need a professional judgement, such as urgency in a medical practice or the scope of a repair. Here the agent takes the details and a person decides.
• How you measure success: appointments booked from chat, the share booked outside opening hours and the no-show rate compared with phone bookings.

4. Lead capture into your CRM

• What the agent does: asks what your sales team would ask in the first call and creates the contact with a summary in HubSpot, Salesforce, Airtable or Google Sheets.
• What data it needs: a list of required details such as name, company, need, timeframe and callback request, plus your rules on which lead goes to whom.
• Where the limit is: only ask for what you really need. The GDPR requires data minimisation, and every extra required field costs enquiries. Quotes and price negotiation stay with sales.
• How you measure success: the share of complete leads, time to first callback and the share of chat leads that turn into a quote.

5. Shop questions: the chatbot for your online shop with Shopify or Shopware

Most shop enquiries repeat: where is my order, is this in stock, how do I return it?

• What the agent does: looks up order status and availability through the Shopify or Shopware 6 integration, explains shipping and returns and passes everything else to your team.
• What data it needs: the shop connection, shipping and returns terms, delivery times by country.
• Where the limit is: order data is personal data. Before launch, decide how the agent recognises the buyer, for example by order number plus email address, and test exactly that. Complaints, credit notes and goodwill stay with people, and since 62 % of online shoppers prefer a human, the way there must stay visible. [1]
• How you measure success: the share of order status questions that end without a person, and emails with the same question compared with the previous month.

6. Onboarding and internal knowledge questions

Onboarding has two sides. New customers ask the same things in their first weeks, such as how to set something up or invite colleagues. New employees ask about holiday requests, expenses or IT access.

• What the agent does: answers questions from guides and handbooks. For internal questions, set up a separate agent with its own knowledge base, embed it on an internal page and use the domain whitelist to allow the widget only there.
• What data it needs: setup guides, staff handbook and process descriptions, uploaded as PDF, TXT, Markdown or CSV or added as single pages that are re-read regularly.
• Where the limit is: each agent has its own knowledge base, but there are no permissions for individual users. Anyone chatting with the agent can learn what is in it, so salary bands, personnel files and confidential material do not belong there.
• How you measure success: fewer follow-up questions to support, HR or IT in the first weeks, and the list of questions the agent had no answer to.

The integrations overview lists the tools the agent can use.

What an AI chat agent costs: a worked example with assumptions

With EchoCall, chat agents are billed by sessions, meaning chat conversations. [9]

Minutes a visitor speaks by microphone in the widget come from the voice pool.

For the worked example we assume:

• An online shop receives 800 chats a month and books Professional for €400.
• A chat handled by a person takes 6 minutes on average.
• An hour of work costs the shop €35. The German average in 2025 was €45.00, according to the Federal Statistical Office. [10]
• The team handles everything else as before.

In this example the break-even point is around 114 resolved chats, just over 14 % of 800. Many calculations miss three points:

1. You pay for every session, not just the resolved ones. At 40 %, each resolved chat in the example costs €1.25 (€400 divided by 320).
2. Saved hours are only money if the time is used elsewhere. If the team stays the same size and simply has less to do, waiting times fall, not costs.
3. You only know the share after launch. Calculate cautiously and check after four weeks.

Chatbot costs compared: how other vendors bill

List prices from the vendors' pages, retrieved on 6 October 2026:

The units are not comparable. If you only pay for resolved conversations, you pay more per unit but nothing for handovers. Each vendor defines "resolved" differently, and some prices require a base subscription.

Klarna shows how big the lever can be and where it ends. According to Klarna, its AI assistant handled 2.3 million conversations in its first month, two thirds of all service chats, doing the work of 700 full-time agents. [15] Just over a year later, CEO Sebastian Siemiatkowski said, according to Bloomberg, that AI-driven cost cutting had gone too far, and Klarna planned to hire people again so customers could always speak to a real person. [16]

How to get started with one use case

1. Pick one use case. Usually first-line support or shop questions, where most identical questions arrive.
2. Fill the knowledge base. Website crawl, help articles as PDF or Markdown, prices and deadlines as a short table.
3. Write the prompt. Tone, form of address, what the agent does not do and when it hands over.
4. Embed the widget. One script snippet on any website or the WordPress plugin, plus the domain whitelist.
5. Test. Run 30 real questions from your inbox and try the handover to the Live Inbox.
6. Review after four weeks. Resolved share, handovers, unanswered questions.

To try it out there are 30 voice minutes and 30 chat conversations for 14 days, with no credit card required. Without signing up, use the live demo: enter your website address, and after about 20 seconds you can write to an agent built from your public pages or call it in your browser.

AI disclosure and data protection

Article 50 of the EU AI Act has applied since 2 August 2026: people must be told clearly, at the latest at the first interaction, that they are interacting with an AI system. Breaches of these transparency duties can cost up to €15 million or 3 % of worldwide annual turnover, whichever is higher. [17] The EchoCall chat widget shows an automatic AI notice in German, English and French. Before the microphone starts, a notice card appears.

For data protection the usual duties for any service provider apply: update your privacy notice, sign a data processing agreement, ask only for necessary data and keep sensitive material out of the knowledge base. The EchoHub runs in data centres in Frankfurt and Karlsruhe, and it also offers a Zero-PII mode. More on our GDPR page.

Frequently asked questions about AI chat agents

What can an AI chat agent on a website take over?

Recurring questions whose answer is in your documents, and simple actions through integrations: booking appointments, looking up order status, creating leads in the CRM, opening tickets. Judgement calls, complaints and negotiations stay with your team.

Which customer service problems can AI agents solve, and which not?

Questions with a documented answer, such as opening hours, delivery status, usage or plans. Not cases that need judgement, disputes or anything without a written rule. There the agent hands over instead of guessing.

What does an AI chat agent cost?

With EchoCall, €40 for 100 sessions a month, €120 for 300, €400 for 1,000 and €1,000 for 5,000+. You pay for every session. Whether it pays off depends on the share of chats the agent resolves alone. In the worked example above, break-even is just over 14 %.

How does the chat agent hand over to a human?

Your team can take over any chat conversation in the Live Inbox. The agent can also create a ticket or notify your team via Slack, Microsoft Teams or email.

Can the chat agent also reply on WhatsApp?

WhatsApp is currently not part of what you set up yourself in the EchoHub. Today the EchoCall chat agent runs on your website, with handover to your team. Our post on WhatsApp Business with AI explains what Meta allows for AI on WhatsApp.

Can AI agents in customer service be used in a GDPR-compliant way?

In principle yes, with a privacy notice, a data processing agreement, data minimisation and no sensitive data in the knowledge base. Since August 2026 the AI disclosure under Article 50 of the AI Act applies too, and the EchoCall widget shows it automatically. Clarify individual cases with your data protection adviser.

Sources

1. Bitkom, press release of 22 May 2025, "Kundenservice beim Online-Shopping: Mensch schlägt Chatbot", 1,006 people aged 16 and over, 978 of them online shoppers (in German): bitkom.org
2. Gartner, press release of 4 August 2026, survey of 3,566 customers: gartner.com
3. Gartner, press release of 8 July 2026, same survey: gartner.com
4. Bitkom, press release of 14 September 2026, "Erstmals nutzt die Mehrheit der Unternehmen KI", 603 companies with 20 or more employees (in German): bitkom.org
5. Civil Resolution Tribunal of British Columbia, Moffatt v. Air Canada, 2024 BCCRT 149, 14 February 2024: decisions.civilresolutionbc.ca
6. Ars Technica, 17 April 2025, report on Cursor's support bot: arstechnica.com
7. Intercom, vendor claim on Fin's resolution rate: fin.ai
8. HubSpot, Customer Agent, vendor claims on resolution rate and price: hubspot.com
9. EchoCall, pricing: echocall.de/en/pricing
10. Destatis (Federal Statistical Office), press release no. 148 of 29 April 2026, labour costs 2025 (in German): destatis.de
11. Intercom, Fin pricing: fin.ai/pricing
12. Zendesk, pricing (in German): zendesk.de
13. Freshworks, Freshdesk pricing: freshworks.com
14. Salesforce, Agentforce pricing: salesforce.com
15. Klarna, press release of 27 February 2024: klarna.com
16. Bloomberg, 8 May 2025, "Klarna Slows AI-Driven Job Cuts With Call for Real People", behind a paywall: bloomberg.com
17. Regulation (EU) 2024/1689 (AI Act), Articles 50 and 99: eur-lex.europa.eu

All links were checked on 6 October 2026. This post is not legal advice.

Revised in October 2026: we removed claims we could not source.

Read next

• AI Chat Agent and Chatbot Statistics 2026: Market Size, Adoption, What Customers Want, Cost per Resolution and the Law, Every Number Sourced
• WhatsApp Business with AI in 2026: What Meta Allows, What It Costs and How to Stay GDPR Compliant
• EU AI Act 2026: What the AI Disclosure Duty Means for Voice and Chat Systems
• Multilingual AI Voice Agent 2026: Serving International Customers in More Than 30 Languages by Phone and Chat