What is Tier-1 support in customer service?

Tier-1 support is the first layer of a support queue: routine, repetitive questions that don't need specialist knowledge, like order status, password resets, billing basics, or account changes. It's the layer most support teams spend the most hours on, and the layer every example in this article automated first, before touching anything more complex.

Can an AI agent really resolve most support tickets on its own?

For the Tier-1 share of a queue, yes, based on the published numbers here: 50 to 95 percent depending on the company and how narrow the scope was kept. None of these teams claim their AI agent handles everything. Each one still escalates anything outside its scope to a human, which is exactly why the numbers hold up instead of collapsing under edge cases.

Does automating Tier-1 support hurt customer satisfaction?

Not in these cases. Lightspeed Commerce reported a 60% increase in CSAT after automating with Fin, and Chime doubled its member satisfaction score while cutting support costs 60%. A fast, correct answer to a routine question tends to satisfy customers more than a slow one from a human, which is the opposite of what most teams expect going in.

How long does it take to see results from an AI customer service agent?

Fundrise reported over 50% of support cases automated in under three months. Anthropic, a separate Fin customer, reported a 50.8% resolution rate in just over a month. The common factor across faster results is a narrow starting scope. Teams that try to automate everything at once tend to take longer to reach a comparable number, not faster.

Which company had the best AI customer service results?

By raw percentage, ClassPass's 95% cost reduction and Duolingo's 80% chat deflection stand out, but "best" depends on what you're optimizing for. Chime optimized for both cost and satisfaction at once. Lightspeed optimized for speed without losing quality. Match the example to what your own team is actually trying to fix, not the biggest headline number.

Are these companies using LiveAgent's AI agent?

No. These five examples use Intercom's Fin and Decagon's AI agent, and we're citing their own published results as evidence the underlying pattern works, not as an endorsement of either tool. LiveAgent has its own AI Agent feature, built to work department by department, which we cover in a separate setup guide.

AI Customer Service Agent Use Cases: 5 Teams That Replaced Tier-1 Support

Most “AI customer service use case” content is speculation. These are 5 real teams with real, published numbers: what got automated, what didn’t, and what the result actually was.

Published on Sep 28, 2026 by Adam Khaled.

AI Customer Support Case Studies

Start your free trial Schedule a demo

Five companies have published real numbers from putting an AI agent on Tier-1 support: Fundrise automated 50%+ of cases in 3 months, Lightspeed resolved 65% and raised CSAT 60%, Chime cut costs 60% while doubling satisfaction, Duolingo deflected 80% of chat, and ClassPass cut costs 95%. None are LiveAgent customers; every number is cited from the company's own case study.

In this article:

What Tier-1 support actually means
Fundrise: 50%+ of support cases in under 3 months
Lightspeed Commerce: 65% resolved, CSAT went up
Chime: 60% lower cost, satisfaction doubled
Duolingo: 80% of chat deflected
ClassPass: cost cut 95%, deflection beat their own estimate
What these 5 teams actually have in common

Most examples of AI in customer service describe what an AI agent could theoretically do. This one only uses cases where a real company published a real number, with a named source, so you can check it yourself instead of taking a blog’s word for it.

None of the five companies below use LiveAgent. We’re citing their own published results as evidence for the pattern, not as an endorsement of the tools they used.

What Tier-1 support actually means

Tier-1 is the first layer of any support queue: routine questions that don’t need specialist knowledge to answer correctly. Order status. Password resets. Billing basics. Account changes. It’s also, in most teams, the layer that eats the most agent hours for the least reward, since a skilled agent spends the same ten minutes on a routine question as a junior one would.

That’s exactly why it’s the layer every example below started with. None of these five teams tried to automate their hardest tickets first.

Illustration of an AI customer service agent auto-resolving routine Tier-1 tickets like order status and password resets, then escalating a complex dispute to a human agent

Fundrise: 50%+ of support cases in under 3 months

Fundrise , a direct-to-investor real estate platform, deployed Intercom’s Fin AI agent to handle its growing support volume without adding headcount. Luke Ruth, Fundrise’s Chief Product Officer, said: “Less than three months after launching Fin, it now handles more than 50% of our total support cases.”

The context matters here as much as the number. Fundrise was trying to keep response times fast for a growing investor base while still leaving room for agents to give thoughtful answers to complex financial questions, the kind of question that shouldn’t go to an AI agent at all. Automating the routine half of the queue is what created that room.

Lightspeed Commerce: 65% resolved, CSAT went up

Lightspeed Commerce , a global e-commerce and point-of-sale platform, runs Fin across a support team split between longtime Intercom users and agents new to the tool. Fin now participates in 99% of conversations and autonomously resolves up to 65% of them.

The number worth paying attention to isn’t the resolution rate. It’s that Fin CSAT rose 60% alongside it, while chat response times dropped 38% and email response times dropped 67%. The common assumption is that automating support trades speed for quality. Lightspeed’s numbers say the opposite happened, once the AI agent had full control over when to hand a conversation to a human instead of forcing every case through it first.

Chime: 60% lower cost, satisfaction doubled

Chime , a fintech company, deployed Decagon’s AI agent and reported a 60% decrease in customer support costs, a 70% chat and voice resolution rate, and a doubled member satisfaction score in the process.

Cost and satisfaction moving in the same direction, instead of trading off against each other, is the least intuitive part of Chime’s result. It only makes sense once you separate what got automated (the routine, high-volume share of the queue) from what stayed with agents (the questions where a human genuinely adds value). Neither side got worse. Each side got more of what it was actually good at.

Duolingo: 80% of chat deflected

Duolingo deployed Decagon’s AI agent and reported an 80% chat deflection rate, meaning 4 out of every 5 chat inquiries never reach a human agent at all. The case study doesn’t break down exactly which categories make up that 80%, but the stated effect is that human agents now spend their time on the more complex inquiries the deflected volume used to crowd out.

At Duolingo’s scale, an 80% deflection rate isn’t a small efficiency gain. It’s the difference between a support team that can keep up with a growing user base and one that structurally can’t, without either hiring at the same pace or accepting slower answers.

ClassPass: cost cut 95%, deflection beat their own estimate

ClassPass , a fitness and wellness subscription platform, already ran a mature program for understanding what could be deflected before deploying Decagon’s AI agent. Even with that head start, the team said: “We saw 10x higher deflection at launch than we anticipated,” alongside a reported 95% reduction in support costs.

That detail is the most useful one in this whole list. ClassPass wasn’t guessing. They had existing customer research and still underestimated what the AI agent could take off their queue by an order of magnitude. If a team with a mature program can be off by 10x, most teams deploying for the first time should expect their own early estimate to be wrong too, in the same direction.

What these 5 teams actually have in common

Strip away the vendor names and the five results above share the same shape:

Bar chart comparing each company's headline AI agent result: ClassPass 95% cost reduction, Duolingo 80% chat deflected, Chime 70% chat and voice resolution, Lightspeed Commerce 65% conversations resolved, Fundrise 50% support cases automated

Team	Automated	Main result
Fundrise	50%+ of support cases	Full-time reallocation of agent hours to complex questions in under 3 months
Lightspeed Commerce	65% of conversations	Resolution AND satisfaction rose together, not traded off
Chime	Routine chat/voice volume	60% lower cost, satisfaction doubled
Duolingo	80% of chat	Support kept pace with a growing user base without matching headcount growth
ClassPass	Far more than expected	95% cost cut, 10x their own deflection estimate

None of them automated everything. All five kept a human layer for anything outside the AI agent’s scope. And in every case, the team started with the routine, repetitive share of the queue, the Tier-1 layer, not the hardest tickets they had.

That’s the actual takeaway, more than any single percentage: the pattern isn’t “replace support with AI.” It’s “stop spending skilled agent time on unskilled questions,” and let the numbers follow from there. If you’re ready to test this on your own queue, our guide on deploying an AI customer service agent covers exactly how to scope that first phase, and setting up an AI agent in LiveAgent covers the LiveAgent-specific configuration steps.

Share this article

Frequently asked questions

: Tier-1 support is the first layer of a support queue: routine, repetitive questions that don't need specialist knowledge, like order status, password resets, billing basics, or account changes. It's the layer most support teams spend the most hours on, and the layer every example in this article automated first, before touching anything more complex.
: For the Tier-1 share of a queue, yes, based on the published numbers here: 50 to 95 percent depending on the company and how narrow the scope was kept. None of these teams claim their AI agent handles everything. Each one still escalates anything outside its scope to a human, which is exactly why the numbers hold up instead of collapsing under edge cases.
: Not in these cases. Lightspeed Commerce reported a 60% increase in CSAT after automating with Fin, and Chime doubled its member satisfaction score while cutting support costs 60%. A fast, correct answer to a routine question tends to satisfy customers more than a slow one from a human, which is the opposite of what most teams expect going in.
: Fundrise reported over 50% of support cases automated in under three months. Anthropic, a separate Fin customer, reported a 50.8% resolution rate in just over a month. The common factor across faster results is a narrow starting scope. Teams that try to automate everything at once tend to take longer to reach a comparable number, not faster.
: By raw percentage, ClassPass's 95% cost reduction and Duolingo's 80% chat deflection stand out, but "best" depends on what you're optimizing for. Chime optimized for both cost and satisfaction at once. Lightspeed optimized for speed without losing quality. Match the example to what your own team is actually trying to fix, not the biggest headline number.
: No. These five examples use Intercom's Fin and Decagon's AI agent, and we're citing their own published results as evidence the underlying pattern works, not as an endorsement of either tool. LiveAgent has its own AI Agent feature, built to work department by department, which we cover in a separate setup guide.