What Companies Use AI for Customer Service

The Short Answer: Almost Every Serious Company Is

If you're asking what companies use AI for customer service, the honest answer is: every company that can afford to figure it out is actively doing it. Delta Airlines, Bank of America, H&M, Klarna, Amtrak, Airbnb, Lyft, Verizon, Heathrow Airport - the list is long and getting longer every quarter. This isn't a future trend. It's already the operating norm for competitive businesses.

But knowing who is doing it matters less than understanding how they're doing it and what that means for you - whether you're building a support operation yourself, or you're an agency selling services to companies making this transition.

Let's break down the real examples, the tools they're using, and the angle that most articles miss: this AI shift is creating a massive opening for outbound sellers who know where to look.

Real Companies, Real Results

Airbnb - 40% of Inquiries Resolved Without a Human

Airbnb is one of the most instructive case studies in this space right now. Rather than chasing flashy AI travel planning tools like some of its competitors, Airbnb deliberately started at the bottom of the funnel - customer service. CEO Brian Chesky publicly said the company chose this because they saw it as the hardest problem in AI to solve well.

The results have been significant. Airbnb's custom-built AI assistant now handles roughly 40% of customer inquiries without escalating to a live agent - up from about 33% at the start of the same year. The company also noted a 15% reduction in customers needing to contact live human agents at all. Their approach is channel-agnostic, handling issues through in-app messaging and automated phone calls using what they call an Intelligent Automation Platform - a task-oriented dialog system that detects user intent, generates dynamic responses, and routes customers through product workflows to resolve issues or hand off to human agents when needed.

The Airbnb example matters because it shows measured, deliberate AI rollout working. They didn't try to automate everything at once. They picked the hardest, most impactful use case and built around it systematically.

Delta Airlines - AI Inside the Call Center

Delta deployed an AI chatbot called "Ask Delta" that helps customers check in, track bags, and find flights. More importantly, they embedded AI inside their reservations team to help agents quickly parse historical policies and pull up answers during live calls. When a passenger calls with a question about traveling with a pet, for example, the AI instantly surfaces the relevant section of the procedural manual so the agent doesn't have to hunt for it. The result: Delta's call center volumes dropped by 20%. That's not a rounding error - that's a structural reduction in headcount cost.

Bank of America - Erica, the Banking Assistant

Bank of America's AI assistant Erica handles everything from tracking spending to canceling subscriptions to helping users make trades on their Merrill investment accounts. Erica has processed more than 3 billion customer interactions since launch - handling approximately 58 million interactions per month. The system operates against a library of roughly 700 topics using supervised machine learning rather than generative AI. That constraint is deliberate. It gives Bank of America precise control over what the AI attempts to resolve and, critically, when it stops trying. When Erica escalates, it transfers full interaction context to the receiving agent. That architecture is a big part of why the product has scaled without collapsing under edge cases.

Klarna - 700 Agents Worth of Conversations

Klarna's AI assistant made headlines when, in its first month live, it managed 2.3 million conversations - the equivalent workload of 700 human agents. Customer resolution times dropped from an average of 11 minutes to just under two. The tradeoff? Some customers found the bot frustrating for complex disputes. That's the consistent pattern across all these deployments: AI handles volume extremely well, but edge cases still leak through.

What happened next is equally instructive. Klarna's fix centered on escalation design - adding AI-generated handoff summaries so agents received full context, and introducing confidence scoring so the system would know when to escalate rather than push forward with a low-confidence answer. The result? The AI actually ended up handling more interactions than before, not fewer, because the handoff was finally trustworthy. That's the real lesson from Klarna: the automation isn't what fails, the escalation path is.

Lyft - 87% Reduction in Resolution Time

Lyft partnered with Anthropic and AWS to deploy an agentic AI system for customer and driver support. The results were dramatic: an 87% reduction in average resolution time, with more than half of customer and driver support requests handled in under three minutes. The system is available 24/7 in English and Spanish.

What makes Lyft's implementation interesting is the stated goal - not to replace agents, but to unbundle the work. High-volume routine issues go to AI. Human agents are repositioned as "true specialists" for complex inquiries where empathy and accountability actually matter. The VP of Safety and Customer Care at Lyft put it plainly: drivers were waiting for help on issues affecting their livelihood, while riders weren't getting fast personalized support. When generative AI matured, they recognized agentic AI was uniquely suited to solve that specific problem.

Amtrak - 5 Million Requests a Year

Amtrak's AI virtual assistant Julie handles customer inquiries ranging from booking tickets to checking train schedules. Julie handled over 5 million customer requests in a single year, achieved a 25% increase in self-service bookings, and reduced average handling time across the board - all while lowering operational costs during peak travel seasons.

Verizon - Predicting Why You're Calling Before You Say a Word

Verizon's AI system correctly anticipates the reason behind 80% of its 170 million annual customer calls, routing them to the best-suited agent or automation path before a rep even picks up. That's predictive support at an absurd scale - and it directly reduces average handle time and agent stress. Beyond prediction, Verizon avoids 100,000 potential churn cases annually and reduces average in-store visit time by seven minutes per customer through this system.

What sets Verizon apart is how they treat every transfer as a data point. The system tags and tracks every handoff, feeding outcomes back into the learning algorithm. Issues that consistently require escalation get flagged for retraining. As one Verizon Business executive described it, the goal is a "sixth sense" model for human agents - arriving at a handoff already armed with context, sentiment data, and predicted next steps.

Heathrow Airport - Summarizing and Responding at Scale

Heathrow handles over 1,300 flights a day with nearly 90% international travelers. They use generative AI to reply to service queries and automatically summarize cases, which feeds back into continuous improvement. The system is saving agents significant time and delivering what Salesforce's CEO Marc Benioff described as "incredible new levels of productivity."

H&M - The 24/7 Digital Stylist

H&M deployed an AI chatbot that helps customers check product availability, track orders, and get style recommendations - acting as a 24/7 shopping assistant that reduces the load on live agents while improving response times. It started on messaging platforms like Kik, asking shoppers about style preferences to help them find products and put together outfits. Simple concept, real reduction in ticket volume.

Uber - AI for Measuring and Improving CX

Uber's use of AI in customer service leans heavily on measurement and feedback loops. After rolling out a new rider app, Uber used AI to identify unhappy users and surface that feedback to the product team before complaints flooded the support inbox. The logic is sound: if you can detect a bad experience pattern early via AI sentiment analysis, you can fix the root cause before it becomes a support ticket. That's a fundamentally different use case than a chatbot - it's AI functioning as an early warning system upstream of the support team.

DoorDash - Solving the Unanswered Call Problem

DoorDash identified a specific, painful problem: restaurants were leaving 50% of customer calls unanswered. During peak times, staff can't split their attention between in-store customers and ringing phones. DoorDash launched AI-powered voice ordering technology to answer those calls automatically, take orders, and handle common inquiries - turning a massive revenue leak into a solved problem. The framing matters here. DoorDash didn't lead with "AI for efficiency"; they led with a specific measurable problem that the AI addressed directly.

What These Companies Are Actually Deploying

Most people think AI customer service equals chatbot. That's one piece. The full picture is wider:

• AI chatbots and virtual agents - Handle FAQs, order tracking, appointment scheduling, and common requests 24/7 without a human. These range from simple rule-based bots to sophisticated generative AI systems that understand context and take action.
• Agent assist tools - AI copilots that sit inside the helpdesk interface, pulling up relevant past conversations, suggesting responses, and summarizing long threads in real time. These make your existing reps faster, not redundant. Companies that deployed agent assist tools saw a 27% reduction in average handle time, and organizations pairing agents with virtual assistants handle 7.7% more simultaneous chats on average.
• Predictive support - Analyzing usage patterns and behavioral signals to identify which customers are about to have a problem, then proactively stepping in before they submit a ticket. Companies using predictive analytics for support see a 20-30% increase in operational efficiency and a 10-15% boost in customer satisfaction.
• Automated ticket routing and triage - AI detects whether an incoming ticket is a bug report, billing issue, or feature request, then routes it to the right team automatically. Companies using this report up to 65% reduction in first response times. Merchants using automation respond 37% faster than those who don't.
• Sentiment analysis - Scanning thousands of support tickets, reviews, and survey responses to flag trends like "frustrated onboarding" or "confusing checkout" before they turn into churn events. American Express uses AI-driven feedback analysis to review survey data and uncover customer satisfaction drivers across product features.
• Voice AI and intelligent call routing - AI analyzing incoming calls in real time to predict intent, route intelligently, and handle tier-1 queries via voice before a human ever touches them. Verizon's system at 170 million calls per year is the extreme end of this.
• Multilingual support - AI translating incoming messages and outgoing responses in real time, letting companies serve global audiences without hiring multilingual reps. Lyft's system handles English and Spanish natively. Heathrow uses it to handle near-90% international traveler volume.

The Human Handoff Problem Nobody Talks About Enough

Here's the part that separates companies that win with AI customer service from those that generate PR nightmares: the handoff.

The biggest service failure in most AI deployments isn't the bot itself - it's the broken transition when context doesn't move cleanly from AI to a human agent. Customers end up repeating themselves, which is consistently cited as one of the most frustrating support experiences possible. One study found 73% of consumers say having to repeat information is one of the most frustrating parts of a support interaction, especially after being transferred.

Three failure modes show up repeatedly in poorly designed deployments:

• The amnesia problem - A customer spends five minutes explaining their situation to a chatbot. The bot determines the issue requires human attention and initiates a transfer. The agent picks up with no record of the prior conversation. The customer repeats everything from the beginning. This happens when AI tools, CRM systems, and support platforms operate independently with no shared context layer.
• Poor escalation logic - Customers stuck in conversation loops, desperately typing "AGENT" or "HUMAN" with no escape. The system can't recognize when it's failing and a handoff is required. This is one of the most common causes of brand damage in AI customer service deployments.
• Metric misalignment - Performance is measured by deflection rate (keeping users away from agents) while ignoring critical indicators like escalation failure rate, intent recognition accuracy, CSAT, and time-to-human. You end up with a bot that looks great on dashboards and terrible to customers.

The Commonwealth Bank of Australia found this out publicly. They rolled out an AI voice bot claiming it reduced call volumes by 2,000 per week. Their staff union challenged the data - members reported the opposite: call volumes rising, staff being pulled onto phones to handle overflow. The bot was creating escalations rather than reducing them. CBA reversed course and publicly apologized. The AI worked. The system surrounding it didn't.

The companies that get this right - Bank of America with Erica, Verizon's tagged handoff system, Klarna after their pivot - all treat the transition as a first-class engineering problem, not an afterthought. A good handoff feels invisible: the agent picks up exactly where the AI left off, fully informed and ready to act. A bad handoff forces the customer to start over, breaking trust instantly.

If you're selling AI implementation services to companies going through this transition, the handoff architecture is where your value is. Most companies have someone to help them pick a chatbot platform. Very few have someone who understands how to design the escalation logic, wire up the CRM context pass-through, and build the feedback loops that make the system get smarter over time.

The Tools Powering All This

If you're looking at the actual platforms behind these deployments, the names that keep showing up are:

• Intercom (Fin) - Intercom's AI agent Fin handles customer queries in over 45 languages, pulling answers directly from your knowledge base. Strong fit for SaaS and tech companies that already live inside Intercom.
• Zendesk AI - Zendesk's platform connects AI agents with ticketing, helpdesk, knowledge base, analytics, and quality assurance in one system. Their AI agents learn from each service interaction through what they call a Resolution Learning Loop. Pre-trained on billions of CX interactions, so setup time is lower than building from scratch.
• Ada - Enterprise-grade AI agents with extensive customization. Integrates with Freshworks, Amazon Connect, ServiceNow, Zendesk, and Salesforce. Pricing is custom and quote-based.
• Tidio - More accessible option for smaller teams. Has a free plan with basic live chat, paid plans that scale up. Includes an AI phrase matcher and a reply assistant for human agents. Their AI product Lyro can automate up to 70% of customer requests according to their published data.
• Chatbase - Train a custom AI agent on your own data, connect it to your CRM and order systems, and deploy it across channels. Good for teams that want control over their knowledge base without heavy dev work.
• Salesforce Agentforce - Salesforce's AI-native agent platform with configurable escalation triggers, seamless human handoff, and full conversation history preservation. Built for enterprise with Omni-Channel skills-based routing baked in.
• Amazon Connect + Claude - Lyft's implementation used Anthropic's Claude AI via Amazon Bedrock as the intelligence layer powering their support system. This approach gives companies access to frontier model quality without building training infrastructure themselves.

If you're building out AI-powered outreach sequences or want to brainstorm SaaS angles in this space, grab the SaaS AI Ideas Pack - it covers a range of productized opportunities in the AI and business tooling space that I've been tracking.

Industry Breakdown: Who's Moving Fastest

Not all industries are adopting at the same rate. If you're doing outbound into this space, it helps to know where the density of active buyers is highest.

Financial services is one of the most aggressive adopters. Bank of America's Erica is the most-cited example, but it runs deep across the sector. The unit economics are compelling: agents cost $3-6 per interaction versus $0.25-0.50 for AI-handled contacts. At millions of monthly interactions, even a 30% AI containment rate saves tens of millions annually.

Travel and hospitality is another high-density sector. Airbnb, Delta, Amtrak, Heathrow, and Expedia are all active. The driver here is 24/7 global demand with time-sensitive queries. A customer locked out of their hotel room at 2am in Tokyo doesn't want a ticket queue - they want an answer now. AI handles that clock without adding overnight headcount.

Telecom is extremely active, with Verizon being the standout example. Call volume at telecom scale - 170 million annual calls - makes even marginal efficiency gains worth tens of millions. Predictive intent detection is particularly valuable here because most calls fall into a handful of high-frequency categories that AI can learn to route instantly.

E-commerce and retail is growing fast. H&M, Sephora (which uses a virtual assistant for product discovery and service booking), and DoorDash all represent different angles - order tracking, product recommendations, and voice ordering. The common thread is high ticket volume with mostly repetitive queries, which is exactly where AI delivers the best ROI.

Gig economy platforms (Uber, Lyft, DoorDash) have a unique challenge: they have to simultaneously support multiple user types - customers, drivers, and merchants - each with different needs, urgency levels, and emotional states. AI that can detect intent and route correctly across all three is genuinely complex to build, which is why these companies have invested heavily in it.

The Data on ROI

A few numbers worth knowing if you're pitching into this space or building for it:

• Companies see average returns of $3.50 for every $1 invested in AI customer service.
• AI agents cost $0.25-0.50 per interaction compared to $3-6 for human agents - an 85-90% cost reduction per contact.
• Gartner benchmarks the median cost per self-service contact at $1.84 versus $13.50 for agent-assisted - a 7x difference.
• Chatbots can reduce customer service costs by up to 30%, which is why pilots keep turning into full deployments.
• 74% of high-performing service agents say AI tools make it easier to deliver high-quality service - meaning AI isn't primarily a replacement play, it's an augmentation play.
• 92% of businesses report improved CSAT after implementing AI. The common objection - "customers hate bots" - dissolves when the implementation is actually good.
• 75% of customer inquiries can now be resolved by AI tools without human intervention, according to data from contact center analytics.
• Agent assist tools have been adopted by 40% of support units, with companies that deployed them seeing a 27% reduction in average handle time.
• The global AI for customer service market is projected to hit $117.87 billion by 2034. The spend is real and accelerating - the market is already at $15.12 billion this year, growing at 25.8% CAGR.

One important nuance: 88% of contact centers report using some form of AI, but only 25% have fully integrated automation into daily workflows. That gap - between adoption and integration - is where the real service opportunity lives for agencies and implementation consultants.

What AI Customer Service Still Gets Wrong

I want to give you the honest picture here, because if you're selling services to companies deploying this stuff, understanding the failure modes is as valuable as knowing the wins.

Hallucination and bad information. Generative AI chatbots can invent facts. Not occasionally - enough that companies need systematic quality controls on AI responses. New York City's MyCity chatbot famously advised restaurant owners they could serve cheese that a rodent had nibbled on, directly contradicting local health regulations. The legal precedent now exists in multiple jurisdictions: companies are responsible for all information on their websites, including chatbot responses. Your AI's mistakes are your legal liability.

Emotional intelligence gaps. AI handles structured, repeatable, factual queries well. It handles emotionally charged situations - a flight canceled on the day of a family emergency, a fraudulent charge that wiped out someone's account - poorly. Customers know this. 79% of Americans still prefer interacting with humans over AI for customer service, and while 51% prefer bots when they want immediate service, that preference reverses fast when the stakes go up. The companies that win are designing systems where AI handles volume and speed, humans handle complexity and emotion, and the handoff between the two is seamless.

Knowledge base decay. AI is only as good as the information it's trained on. 61% of leaders say they have a backlog of articles to edit, and more than one-third have no formal process for revising outdated knowledge base articles. An AI confidently answering customer questions based on a policy that changed six months ago is worse than no AI at all - it erodes trust fast.

Over-deflection mentality. The most common implementation mistake is designing AI around deflection metrics - how many customers can we keep away from human agents? Companies that optimize purely for deflection end up with bots that trap frustrated customers in loops. The better metric is resolution rate. Did the customer actually get their problem solved? Deflection and resolution are not the same thing, and confusing them is why some high-profile AI deployments have blown up publicly.

The Outbound Angle Nobody Talks About

Most articles about AI customer service are written for the buyer - the VP of Customer Success deciding which platform to implement. I want to talk to the seller: the agency owner, the SaaS founder, the outbound rep who's looking at this industry shift and asking "where's my angle here?"

The angle is this: every company deploying AI for customer service needs three things they often can't build internally.

1. Clean prospect and customer data. AI support tools are only as good as the data they're trained on. Companies upgrading to AI customer platforms are simultaneously auditing their CRM data, cleaning their contact lists, and building tighter customer profiles. If you sell data services, list enrichment, or CRM cleanup - this wave is your friend.

2. Implementation and integration help. Deploying Intercom's Fin or Zendesk AI isn't plug-and-play for most mid-market companies. There's a knowledge base to build, escalation rules to define, integrations to wire up, handoff logic to design, and ongoing training to manage. Agencies that can position themselves as "AI customer service implementation partners" - and specifically as experts in escalation design and handoff architecture - are getting into deals that used to go to Big 4 consultants. That gap between 88% adoption and 25% integration is the entire addressable market for implementation services.

3. Outbound to reach the decision-makers driving this shift. The person approving an AI customer service purchase is usually the VP of Customer Success, Head of Support, or COO. They're reachable. Cold email still works if you lead with specificity - reference their tech stack, their industry, or a specific pain (call center volume, CSAT scores, agent turnover). If you want a tested framework for writing that kind of outreach, the Cold Email GPT Prompts pack has templates built around exactly this kind of technical buyer.

How to Find Companies Investing in AI Customer Service

If you want to build a prospect list of companies actively in this transition, there are a few reliable signals:

• Recent tech stack additions - Companies that just added Intercom, Zendesk, Salesforce Service Cloud, or Ada to their stack are likely mid-deployment. A BuiltWith scraper can pull this data at scale - filter for companies running these platforms and you've got a warm list of buyers actively invested in customer service tech. This is one of the highest-signal prospecting methods I've seen for this category.
• Job postings - A company hiring a "Conversational AI Specialist," "CX Automation Manager," or "Knowledge Base Manager" is in active build mode. That's a buying signal. Use it to time your outreach when budget is already allocated and the decision-maker is in motion.
• Company size and industry - E-commerce, fintech, SaaS, travel, and telecom are the highest-adoption sectors right now. If you want a B2B database you can slice by industry, company size, and decision-maker title, ScraperCity's B2B lead database lets you filter exactly that way without paying per contact - useful when you're pulling hundreds of prospects across multiple verticals.
• Press releases and earnings calls - Publicly traded companies like Airbnb and Lyft announce AI customer service milestones on earnings calls. Track those announcements and you'll know exactly when a company is in post-launch optimization mode - a prime time to sell implementation help, training data services, or adjacent tooling.
• LinkedIn signals - When a VP of Support starts posting about "AI transformation" or a company's LinkedIn page starts featuring customer service AI content, they're in the awareness-to-consideration phase. That's earlier in the buying cycle than a tech stack trigger, but it gives you more runway to build a relationship.

Once you've got the list, you need the contact data. For finding direct emails for the CX and ops buyers at these companies, an email finding tool like Findymail is worth having in the stack - high deliverability, works well with Apollo exports and LinkedIn-sourced leads.

If you need to find direct phone numbers for the decision-makers you're targeting, a mobile finder tool can surface direct dials that aren't in any database - useful when email isn't getting responses and you need another channel.

And once your sequences are running, tools like Smartlead or Instantly handle the sending infrastructure so your cold emails don't land in spam.

For the lead research side specifically, check out the GPT Lead Gen Prompts pack - there are some solid prompts in there for building targeted lists of companies in specific tech categories, including AI infrastructure buyers.

What a Good AI Customer Service Pitch Actually Looks Like

If you're an agency or consultant going after companies making this shift, the worst thing you can do is lead with "we do AI." Every vendor says that right now. The companies spending real money on this are sophisticated enough to see through vague positioning immediately.

What works is leading with the specific pain. Here's how I'd think about it by buyer type:

VP of Customer Success at a mid-market SaaS company (200-2,000 employees): Their pain is CSAT declining as the product scales and the support team can't keep up with ticket volume. Lead with: "You're probably handling [X] tickets per month with a team of [Y]. Here's how similar companies reduced first response time by 40% without hiring." Reference their current tech stack if you can.

COO at an e-commerce brand: Their pain is return and order-tracking queries eating agent time during peak seasons. Lead with: "Your support team is probably getting buried in WISMO queries. Here's how [comparable brand] automated 60% of those while improving CSAT." Specific, seasonal, financially framed.

Head of Support at a telecom or financial services company: Their pain is call volume and agent turnover - support is expensive, agents burn out fast, and the business wants to reduce headcount cost without killing service quality. Lead with: "What's your cost per contact right now? Here's how comparable companies have moved that number."

The specificity of the opener determines whether you get a reply. Generic AI pitches get deleted. Pain-first pitches that reference the prospect's actual situation get responses. Every company deploying AI for customer service has at least one of these problems: too many tickets, too slow, too expensive, or too inconsistent. Your job is to figure out which one before you hit send.

What the Smaller Company Can Steal From the Enterprise Playbook

You don't need Airbnb's engineering team or Bank of America's budget to start using AI in customer service effectively. Here's what translates directly to smaller operations:

Start with one high-volume, low-complexity query type. Don't try to automate everything. Pick the one question your team answers 50 times a day - order status, return policy, pricing questions, account access - and build the AI around that single use case first. Lyft, Airbnb, and DoorDash all started narrow and expanded. That's the right order of operations.

Obsess over the handoff before you obsess over automation rates. A bot that handles 40% of queries with a clean handoff is better than a bot that handles 70% of queries but traps frustrated customers in loops. Get the escalation path right first. Make sure when the AI can't help, the human agent picks up with full context - not a blank screen.

Keep your knowledge base current. Your AI will only be as accurate as the information it's trained on. If your help docs are out of date, your AI will confidently give wrong answers. Assign someone to own knowledge base hygiene before you deploy. This sounds boring. It's the difference between a successful deployment and a liability.

Measure resolution rate, not just deflection rate. If you're only tracking how many customers never reached an agent, you're measuring the wrong thing. Track whether the AI actually solved the problem. CSAT on AI-handled conversations versus human-handled conversations is the honest scorecard.

Be transparent. Customers are more willing to interact with AI than you think - 51% prefer bots when they want immediate service - but they want to know they're talking to AI. Companies that pretend the bot is a human and then get caught erode trust fast. Name your bot, state its capabilities upfront, and make the path to a human easy to find.

The Bottom Line

The companies deploying AI for customer service aren't doing it because it's trendy. They're doing it because the math works. Klarna's example alone - one AI assistant doing the work of 700 agents - is hard to argue with from a unit economics standpoint. Lyft cutting resolution time by 87%. Airbnb resolving 40% of inquiries without a human touch. Verizon avoiding 100,000 annual churn cases through predictive routing. These aren't edge cases. They're the results competitive companies are building toward right now.

The nuance is that the companies getting real results aren't just buying a chatbot and calling it done. They're building systems - with clean data, integrated context, well-designed escalation paths, and feedback loops that make the AI smarter over time. That complexity is exactly where the opportunity lives for sellers, agencies, and consultants who understand this space.

For everyone reading this who's in the business of selling, growing, or advising: this shift is a pipeline opportunity. The buyers are identifiable, the pain points are specific (call volume, agent churn, CSAT scores, support costs, botched AI deployments), and the budgets are moving. You just need to get in front of the right people with the right message before your competitors figure out the same thing.

If you want to go deeper on how to structure outbound campaigns targeting these buyers - including messaging frameworks and follow-up sequences - I cover this inside Galadon Gold.