Table of Contents
- Key takeaways
- Which companies use chatbots for customer service?
- How were these 16 companies chosen?
- Customer service assistants with published details
- 1. Bank of America
- 2. Lyft
- 3. Klarna
- 4. Amtrak
- 5. Marriott
- 6. Amazon
- 7. Capital One
- 8. KLM
- Ordering and shopping bots from well-known brands
- 9. Domino's
- 10. Starbucks
- 11. Sephora
- 12. eBay
- 13. H&M
- 14. Whole Foods
- 15. Fandango
- 16. Mastercard
- What do these companies have in common?
- Which example should your business copy?
- How can a smaller business get the same result?
- Frequently asked questions
- Companies and industries
- What companies use AI chatbots for customer service?
- Which banks use AI chatbots?
- Which airlines and travel companies use chatbots?
- Which websites have chatbots?
- Results
- What results do companies see from customer service chatbots?
- Do chatbots replace human customer service agents?
- What went wrong with Klarna's chatbot?
- Technology
- What is an AI chatbot for customer service?
- What companies use ChatGPT or Claude for customer service?
- Getting started
- How can a small business use a chatbot like these companies?
- Should a small business start with every question at once?
- Sources
Created by
Do not index
Created time
Sep 15, 2023 12:35 PM
Companies using chatbots for customer service include Bank of America, Lyft, Klarna, Amtrak, Marriott, Amazon, Capital One, and KLM, and most of them describe a handoff from the chatbot to a human agent.
TL;DR
- 16 well-known companies are listed here: 8 run customer service assistants and 8 launched ordering or shopping bots.
- The largest published number belongs to Bank of America. Its assistant Erica passed 3 billion client interactions in August 2025.
- Lyft reported an 87% drop in average customer service resolution time after adding an AI assistant in 2025.
- Klarna's assistant handled two-thirds of its customer service chats in its first month. In 2025 Klarna started hiring human agents again.
- Lyft, Klarna, Marriott, Amazon, and KLM each describe a handoff to a person. A smaller business can copy that setup with a chatbot trained on its own content.
Key takeaways
Company | Chatbot | Main job | Published result |
Bank of America | Erica | Account questions and spending insights | 3 billion client interactions, nearly 50 million users |
Lyft | Customer care AI assistant | Rider and driver questions | 87% lower average resolution time |
Klarna | AI assistant | Refunds, returns, and payment questions | 2.3 million conversations in the first month |
Amtrak | Ask Julie | Booking help and travel questions | Over 5 million questions a year |
Marriott | ChatBotlr and Rewards chatbots | Stay requests and loyalty account help | 2 in 3 Aloft guests used ChatBotlr in early findings |
Which companies use chatbots for customer service?
The companies that use chatbots for customer service span banking, travel, retail, food, and ride-hailing. The table lists all 16 with the year of the announcement each entry is based on.
# | Company | Industry | Chatbot | Main job | Announced |
1 | Bank of America | Banking | Erica | Account questions, spending insights | 2018 launch, 2025 figures |
2 | Lyft | Ride-hailing | Customer care AI assistant | Rider and driver service requests | 2025 |
3 | Klarna | Payments | AI assistant | Refunds, returns, payment questions | 2024 |
4 | Amtrak | Rail travel | Ask Julie | Booking help, travel questions | Undated vendor case study |
5 | Marriott | Hotels | ChatBotlr, Rewards chatbots | Stay requests, loyalty account help | 2017 |
6 | Amazon | Ecommerce | Customer service chatbots | Refund status, order cancellations | 2020 |
7 | Capital One | Banking | Eno | Account questions, charge alerts | 2021 article |
8 | KLM | Airline | BlueBot (BB) | Ticket booking on Messenger | 2017 |
9 | Domino's | Food delivery | Dom | Ordering and order tracking | 2016 and 2017 |
10 | Starbucks | Coffee | My Starbucks Barista | Ordering by voice or message | 2017 |
11 | Sephora | Beauty retail | Reservation Assistant, Color Match | Appointment booking, shade matching | 2016 |
12 | eBay | Marketplace | ShopBot | Product discovery | 2016 |
13 | H&M | Fashion retail | Kik commerce bot | Outfit suggestions | 2016 |
14 | Whole Foods | Grocery | Messenger recipe bot | Recipe search | 2016 |
15 | Fandango | Movie ticketing | Messenger bot | Showtimes and ticket links | 2016 |
16 | Mastercard | Payments | Mastercard KAI bot | Account and spending questions | 2016 |
The list is not a ranking. It runs from the companies with the most published evidence to the ones with launch details only.
How were these 16 companies chosen?
These 16 companies were chosen on three criteria. Each entry follows the same template so the examples are easy to compare.
- The company named the chatbot in public. The source is the company's newsroom, a vendor case study, or trade press.
- The source says what the chatbot does. Vague claims about "AI-powered service" did not qualify.
- Any number comes from that source. An entry with no published number says so.
Every figure is self-reported by the company or its vendor. None is an independent audit, so read the numbers as direction. For assistants outside customer service, such as voice assistants, AI tutors, and phone agents, see these conversational AI examples.
Customer service assistants with published details
The eight companies in this group use a chatbot for service work: account questions, refunds, bookings, and requests during a stay.
1. Bank of America
Bank of America uses a virtual assistant called Erica inside its mobile app. Erica launched in 2018.
- What the chatbot does: Erica answers account questions and sends proactive insights, such as a cash back deal that matches a client's spending or a warning about where a balance is heading over the next 7 days.
- Published result: Erica has served nearly 50 million users and passed 3 billion client interactions. It averages more than 58 million interactions a month, and more than 98% of users find the information they need.
- Human handoff: Clients can schedule an appointment through Erica, which Bank of America describes as a handoff to its high-touch service channels.
- Pattern to copy: Erica draws on a library of more than 700 responses and has had more than 75,000 updates since launch. A chatbot needs steady maintenance after it goes live.
2. Lyft
Lyft uses a customer care AI assistant built on Anthropic's Claude, which it runs through Amazon Bedrock.
- What the chatbot does: The assistant resolves thousands of customer requests each day. Anthropic's case study lists charge disputes and ride issues for riders, and onboarding questions for drivers.
- Published result: Lyft reports that average customer service resolution time fell by 87%.
- Human handoff: The assistant directs the customer to a human specialist when more hands-on help is needed.
- Pattern to copy: Measure resolution time before launch. Lyft could report a clear result because it had a baseline.
3. Klarna
Klarna uses an AI assistant, powered by OpenAI, for customer service in its payments app.
- What the chatbot does: The assistant handles refunds, returns, and payment questions in 23 markets and more than 35 languages.
- Published result: In its first month the assistant held 2.3 million conversations, two-thirds of Klarna's customer service chats. Resolution time fell from 11 minutes to under 2 minutes, and repeat inquiries dropped 25%. Klarna estimated a $40 million profit improvement for 2024.
- Human handoff: In May 2025 Klarna began recruiting human agents again. CEO Sebastian Siemiatkowski said it is "so critical that you are clear to your customer that there will be always a human if you want".
- Pattern to copy: Do not judge a chatbot on cost alone. Klarna's CEO said cost had been "a too predominant evaluation factor" and that quality suffered.
More retail and payments examples, including Walmart and Lowe's, are in the guide to ecommerce chatbot examples.
4. Amtrak
Amtrak uses a virtual assistant called Ask Julie on its website.
- What the chatbot does: Julie helps travelers book rail travel by prefilling the booking form and guiding them through the remaining steps. She also answers questions such as which items are allowed on a train.
- Published result: Julie answers over 5 million questions a year. The vendor's case study reports an 8x return on the investment and 30% more revenue per booking made with Julie.
- Human handoff: The case study does not describe one.
- Pattern to copy: Let the chatbot fill in the form. Julie collects the trip details in conversation and passes them to the booking tool.
5. Marriott
Marriott uses ChatBotlr at its Aloft hotels and Marriott Rewards chatbots on Facebook Messenger and Slack.
- What the chatbot does: Guests text ChatBotlr to request hotel services or ask about the property. Rewards members use the Messenger and Slack chatbots to research and book stays at more than 4,700 hotels.
- Published result: Early findings showed 2 out of 3 Aloft guests interacting with ChatBotlr, with a five-second response time. Of the members who used the Messenger chatbot that year, 44% received stay-related customer service and 53% got help with their Rewards account.
- Human handoff: ChatBotlr connects guests to the front desk team. Rewards members can chat directly with Customer Engagement Center associates.
- Pattern to copy: Give the chatbot the routine requests so staff have time for the guests in front of them.
6. Amazon
Amazon uses automated agents in its customer service chat, and in 2020 it described a test of neural-network versions.
- What the chatbot does: The tested agents handled two request types: return refund status and order cancellations.
- Published result: Amazon measured automation rate. That metric counts a success when the agent completes the request and the customer does not contact customer service again within 24 hours. The new agents "significantly outperform" the older rule-based ones. Amazon published no percentage.
- Human handoff: A request the automated agent cannot complete goes to a human customer service representative.
- Pattern to copy: Count a conversation as resolved only if the customer does not come back within a day.
7. Capital One
Capital One uses an assistant called Eno in its mobile app, online, and by text message.
- What the chatbot does: Eno answers questions about recent transactions, balances, and due dates at any hour. It also sends alerts for a suspicious charge, a duplicate charge, an unusually large tip, and a free trial that is about to end.
- Published result: Capital One's article gives no usage numbers.
- Human handoff: The article does not describe one.
- Pattern to copy: Alert the customer before they have to ask. A duplicate-charge alert removes a service contact.
8. KLM
KLM uses a service bot called BlueBot, or BB, which it introduced on Facebook Messenger in 2017.
- What the chatbot does: BB helps customers book a ticket in conversation, without a KLM agent.
- Published result: KLM published no figures for BB alone. At launch, its social media service team handled more than 16,000 cases a week in nine languages.
- Human handoff: BB is backed by 250 human service agents and refers customers to them when it cannot help further.
- Pattern to copy: Staff the handoff. KLM put its agent team in the launch announcement.
- Source: KLM newsroom, September 26, 2017
Ordering and shopping bots from well-known brands
The eight companies in this group launched bots that take orders or help customers find products. Most arrived in 2016 and 2017 on messaging apps, and each entry gives the facts from the launch announcement.
9. Domino's
Domino's uses an ordering bot called Dom on Facebook Messenger, part of its AnyWare ordering platforms.
- What the chatbot does: The first version, in September 2016, let customers with a saved Pizza Profile repeat a recent order or track an order. In February 2017 Domino's opened the full menu to anyone, with no profile needed.
- Published result: The announcements give no chatbot figures. Domino's said digital channels produced over 50% of its US sales at the end of 2015.
- Human handoff: Not described.
- Pattern to copy: Start with the repeat order. Domino's shipped reordering first and the full menu five months later.
- Source: Domino's investor relations, September 15, 2016 and Domino's press release, February 1, 2017
10. Starbucks
Starbucks announced My Starbucks Barista in January 2017 as an ordering assistant inside its mobile app.
- What the chatbot does: Customers order and pay by voice or message, and can modify a drink the way they would with a barista. Starbucks also released an Alexa skill for reordering a usual item.
- Published result: None. The beta opened to 1,000 iOS customers, with a wider rollout planned through summer 2017.
- Human handoff: Not described.
- Pattern to copy: Test with a small group. Starbucks limited the first release to 1,000 customers.
- Source: TechCrunch, January 30, 2017
11. Sephora
Sephora launched two Facebook Messenger bots in November 2016: Reservation Assistant and Color Match.
- What the chatbot does: Reservation Assistant books makeover appointments at Sephora stores. Color Match scans a photo and returns the closest lipstick shade from Sephora's range.
- Published result: The launch coverage gives no figures.
- Human handoff: The booking leads to an in-store makeover appointment.
- Pattern to copy: Use the chatbot for the booking step. A short conversation replaces a longer form.
- Source: Retail Dive, November 3, 2016
12. eBay
eBay launched ShopBot in beta on Facebook Messenger in October 2016.
- What the chatbot does: ShopBot is a shopping assistant for eBay's one billion listings. A shopper describes an item by text, voice, or photo, and the bot asks follow-up questions before it recommends listings.
- Published result: None published in the announcement.
- Human handoff: Not described.
- Pattern to copy: Ask a clarifying question before answering. ShopBot narrows the request first.
- Source: eBay Inc., October 17, 2016
13. H&M
H&M was among the first retailers to run a commerce bot on the messaging app Kik, in 2016.
- What the chatbot does: US customers answered multiple-choice questions about their style, and the bot sent outfit and product suggestions. A tap on an item opened H&M's mobile site.
- Published result: None published.
- Human handoff: Not described.
- Pattern to copy: Offer buttons as well as free text. Multiple-choice answers keep the conversation inside what the bot can handle.
14. Whole Foods
Whole Foods launched a Facebook Messenger chatbot for recipes in July 2016.
- What the chatbot does: Customers search recipes by keyword, by cuisine, or by emoji. Sending a jalapeno emoji returns recipes that use jalapenos. Conversable built the bot.
- Published result: None published.
- Human handoff: Not described.
- Pattern to copy: Train the chatbot on content the company already owns. Whole Foods used its recipe database.
- Source: VentureBeat, July 12, 2016
15. Fandango
Fandango announced a Facebook Messenger bot for movie discovery and ticketing in April 2016.
- What the chatbot does: The bot gives movie information, trailers, showtimes, and nearby theaters, with a link to buy tickets on Fandango.
- Published result: None published. The announcement set the launch for May 2016.
- Human handoff: Not described.
- Pattern to copy: Answer the question, then link to the page where the customer completes the task.
16. Mastercard
Mastercard announced a chatbot for banks in October 2016, built with Kasisto on its KAI platform.
- What the chatbot does: The plan was for US consumers to ask the Mastercard KAI bot about their accounts, purchase history, spending levels, and cardholder benefits on Facebook Messenger. A second bot for merchants would let shoppers check out with Masterpass.
- Published result: None. The report covers an announcement, with consumer access planned for early 2017.
- Human handoff: Not described.
- Pattern to copy: Build on a specialist platform when the subject is regulated. Mastercard partnered with a banking AI vendor.
- Source: Finextra, October 25, 2016
What do these companies have in common?
These companies have four things in common.
- The chatbot answers from the company's own data. Erica reads account data. Julie answers from Amtrak's travel information. Whole Foods used its recipe database. This grounding method is called retrieval-augmented generation.
- The scope started narrow. Amazon tested two request types. Domino's started with repeat orders. Starbucks opened to 1,000 customers.
- A person stays reachable. Lyft, KLM, Marriott, Amazon, and Klarna each describe a handoff. The comparison of AI chatbots with human handoff shows how tools pass the conversation on.
- Someone measures it. Lyft tracks resolution time. Amazon tracks whether the customer returns within 24 hours. Klarna tracks repeat inquiries.
The same four points appear as rules in the guide to chatbot best practices.
Which example should your business copy?
Your business should copy the example that matches where its answers live and what a wrong answer costs. The table maps common needs to a company on this list.
If you need... | Copy... | Why |
Answers to repeat questions from pages you already have | Amtrak | Julie answers travel questions and fills in the booking form. |
Faster resolution on a large volume of requests | Lyft | The assistant resolves routine requests and people take the rest. |
Answers about a customer's own account | Bank of America or Capital One | Both connect the assistant to account data. |
Fewer front desk or phone requests | Marriott | ChatBotlr takes routine requests by text. |
Repeat orders in one message | Domino's | A saved order is the simplest transaction to automate. |
Product discovery in a large catalog | eBay or Sephora | Both ask questions first, then recommend. |
A clear rule for when a person takes over | Klarna | Klarna now promises a human whenever the customer wants one. |
How can a smaller business get the same result?
A smaller business can get the same result with a chatbot trained on its own content, with no in-house AI team. The large companies above built custom systems. The part that transfers is the answer layer: a chatbot that reads the pages and documents the business already has. To see how that kind of chatbot works step by step and which type fits, read what a customer support chatbot is.
The steps are short:
- List the 20 questions customers ask most.
- Check that each one is answered on a page, in a help article, or in a document.
- Train the chatbot on that content and test it with the 20 questions.
- Set the rule for when a person takes over, then test the handoff.
SiteGPT is one tool for this answer layer. It trains on 12+ source types, including websites, files, cloud storage, and help centers, and it escalates to a person on every plan. Starter costs $59 a month billed monthly, or $39 a month billed yearly. It has limits: there is no free plan, the 7-day trial needs a credit card, and Starter has one seat, manual content refresh, and no chat integrations. Results from two customers are in the chatbot case studies.
The full build is covered in how to create an AI chatbot. To compare vendors instead of examples, see the list of AI customer service agents.
Frequently asked questions
Companies and industries
What companies use AI chatbots for customer service?
Companies that use AI chatbots for customer service include Bank of America (Erica), Lyft, Klarna, Amtrak (Ask Julie), Marriott (ChatBotlr), Amazon, Capital One (Eno), and KLM (BlueBot). Domino's, Starbucks, Sephora, eBay, H&M, Whole Foods, Fandango, and Mastercard have launched ordering or shopping bots.
Which banks use AI chatbots?
Bank of America and Capital One both use AI chatbots. Bank of America's Erica has passed 3 billion client interactions. Capital One's Eno answers account questions and sends alerts about suspicious or duplicate charges.
Which airlines and travel companies use chatbots?
KLM, Amtrak, and Marriott use chatbots. KLM's BlueBot books tickets on Messenger. Amtrak's Ask Julie helps with rail bookings. Marriott's ChatBotlr takes service requests by text at Aloft hotels.
Which websites have chatbots?
Amtrak's website has Ask Julie, and Capital One offers Eno online and in its app. Most of the other bots on this list launched inside an app or on a messaging platform such as Facebook Messenger.
Results
What results do companies see from customer service chatbots?
Published results vary by company. Lyft reported an 87% drop in average resolution time. Klarna's assistant handled two-thirds of its service chats in the first month. Amtrak's vendor reports over 5 million questions answered a year. All of these figures are self-reported.
Do chatbots replace human customer service agents?
No company on this list reports removing people entirely. Klarna came closest, then started hiring human agents again in 2025. Lyft, KLM, Marriott, and Amazon each describe a handoff from the chatbot to a person.
What went wrong with Klarna's chatbot?
Klarna's CEO said cost had been "a too predominant evaluation factor" and that quality dropped as a result. In May 2025 Klarna began recruiting human agents again and said customers would always be able to reach a person.
Technology
What is an AI chatbot for customer service?
An AI chatbot for customer service is software that answers customer questions in a chat window using a language model. The reliable ones answer from the company's own content and pass the conversation to a person when they cannot help. The difference from an agent that takes actions is covered in AI agent vs chatbot.
What companies use ChatGPT or Claude for customer service?
Klarna's assistant is powered by OpenAI, the company behind ChatGPT. Lyft built its customer care assistant on Anthropic's Claude through Amazon Bedrock. Both companies announced the partnerships in their own press releases.
Getting started
How can a small business use a chatbot like these companies?
A small business can train a chatbot on its own website and documents, test it with real customer questions, and set a rule for handing off to a person. This takes no custom development with a content-trained chatbot tool.
Should a small business start with every question at once?
No. Amazon tested two request types and Domino's started with repeat orders. Start with the questions that already have a written answer, then widen the scope.
Sources
All sources were opened and checked on October 4, 2026.
Written by
