AI customer service agent

AI customer service agents, at the level you are ready for

Build an agent for each job. Start with answers from your knowledge base, add skills that work the ticket, and let it take real actions in your systems when you decide it should.

Three levels

"AI agent" means three different things

One answers questions. One works the ticket. One issues the refund. Each level includes the one below it and asks more of the company running it. Theymes does all three, and you choose the level agent by agent.

Read why we think about it this way

Level 1

Answers

Reads your help center and documents

Integrations
Security
Maturity

Read only

Level 2

Skills

Works the ticket: fields, notes, routing

Integrations
Security
Maturity

Writes in the helpdesk

Level 3

Actions

Refunds, account changes, record updates

Integrations
Security
Maturity

Reads and writes in your backend

Each level includes the one below it, and asks more of the company running it.

To make the difference concrete, here is one customer message handled at each level: "I was charged twice for my subscription this month. Can I get a refund?"

Level 1 Answers

It answers from what you have already written

The agent reads your help center and the documents you give it, finds what applies and writes the answer in the customer's language. Most repeat questions end here, before a ticket exists.

What it takes A knowledge base worth reading. Theymes shows you the questions it could not answer, so you know which article to write next.

Resolution trace
Level 1 · Answers

I was charged twice for my subscription this month. Can I get a refund?

  1. 0.00s Received Web widget · English
  2. 0.34s Knowledge “How refunds work”
  3. 0.81s Answered From the help center article

Sorry about the double charge. You can request a refund under Settings → Billing → Request refund, and our billing team will review it within 3 business days.

Answered · the work is still on the customer
Level 1: a correct answer, and the customer still has the problem.

Level 2 Skills

It works the ticket the way your best triage person would

Give the agent a scenario and tell it what to do when it sees one. It recognizes what the customer wants, asks for the missing details, fills in the fields, writes a private note and routes the ticket to the right team.

What it takes Your routing rules, written down. If you can name the ten things customers ask for most, you can write the scenarios.

Resolution trace
Level 2 · Skills

I was charged twice for my subscription this month. Can I get a refund?

  1. 0.00s Received Web widget · English
  2. 0.22s Intent Billing · duplicate charge · refund request
  3. 0.41s Skill “Billing dispute” scenario matched
  4. 0.63s Enriched Plan, invoice dates and amount added to ticket
  5. 0.79s Note Private note: summary and suggested next step
  6. 0.94s Routed Billing team · Priority high

Sorry about that. I've passed this to our billing team with your invoice details, so you won't need to repeat anything. They'll confirm the refund today.

Handed off · full context attached
Level 2: a human still approves the refund, but the ticket arrives ready to solve.

Level 3 Actions

It fixes the problem in your own systems

Connect the agent to billing, accounts or orders and it can look things up and make changes: issue a refund, restore a purchase, update a record. The customer's problem is gone and nobody on your team had to touch it.

What it takes An API the agent can reach, and a decision from you on what it may do: which actions, up to what amount, and when a human has to approve.

Resolution trace
Level 3 · Actions

I was charged twice for my subscription this month. Can I get a refund?

  1. 0.00s Received Web widget · English
  2. 0.18s Identity Signed-in customer verified
  3. 0.52s Lookup Billing API · 2 charges of €9.99 on Nov 3
  4. 0.71s Policy Duplicate charge · under auto-approve limit
  5. 1.26s Action Refund of €9.99 issued
  6. 1.38s Logged Customer record updated · audit entry written

You're right, we charged you twice on November 3. I've refunded €9.99 to your card. It should show up within 3 to 5 business days.

Resolved · refund issued, no ticket created
Level 3: the agent checks the charge, applies the policy and fixes it.

In practice

One agent per job, working three ways

An agent can close a conversation by itself, prepare it for a human, or work for your own agents inside a ticket. Pick a mode below to follow one conversation from start to finish.

AI agent

Purchase agent

Its job, in your words

“Help customers whose purchase is missing. Check the order, find out why, and walk them through the fix.”

Gets the ticket when

Issue type: PurchaseAny language

I bought the season pass but it's not showing up. Restoring purchases didn't help.

  1. Assigned to Purchase agent Issue type Purchase · iOS
  2. Read the order Order 7731 · paid with another Apple ID
  3. Replied Steps to switch accounts in the app
  4. Customer confirmed “It's there now, thanks!”

Resolved by the agent No human involved

71%
Resolved alone
29%
Handed over
4.6
Satisfaction

How to improve it 18 handovers last week were about family sharing. Add it to this agent's instructions.

  • One agent per job, described in your own words
  • Theymes decides which conversations each agent takes
  • See where every agent hands over, and what would fix it

Control

How far it goes is your decision

The models are capable enough. What matters is how much you trust software to act for you, and that answer is different for every company and every kind of request.

You write the job

Each agent has one job, described in plain language. You can read exactly what it has been told to do, and so can the person who takes over from you.

You set the limits

Decide which actions an agent may take, up to what amount and for which customers. Anything outside those limits goes to a human.

Your team sees what matters

Agents run separately from the ticket queue. Conversations they resolve never land on your team, and the ones they hand over arrive with the details collected.

You can see how to improve it

For every agent you see what it resolved, where it handed over and why. The gaps become your to-do list: an article to write, a rule to add, an instruction to change.

Before you start

What each level asks of you

Going from level 2 to level 3 turns a support project into a company project. It is better to know that before you begin.

Level 1Answers Level 2Skills Level 3Actions
Works with Help center and documentsPlus ticket fields, notes and queuesPlus billing, accounts and backend records
Access Reads contentWrites inside TheymesReads and writes in your backend
Setup Connect a knowledge baseDescribe scenarios and routingAn API integration per system
Worst case A wrong answerA misrouted ticketA wrong refund or a changed record
Guardrails Content quality and toneScenario rules and reviewIdentity checks, limits, approvals, audit log
Who's involved SupportSupport operationsSupport, engineering, security and finance

CX AI maturity assessment

Which level is your support organization ready for?

Ten questions, about two minutes. You get a score, the level you can run today and the three things to fix next. No email needed.

Take the assessment

FAQ

Questions about AI agents

What is an AI customer service agent?
An AI customer service agent is software that handles a customer conversation by deciding what to do and then doing it. Depending on how far you let it go, it answers from your knowledge base, works the ticket by filling in fields and routing it, or takes real actions such as issuing a refund.
What is the difference between an AI agent and a chatbot?
A chatbot follows a script or retrieves an answer. An agent chooses between actions. The useful question to ask any vendor is what the agent can change. If the answer is nothing, it is a chatbot with better language skills, which can still be the right product for you.
Which level should we start with?
Start with answers. Every level builds on the one below, and an agent with poor knowledge will take the wrong action with confidence. Add skills once you know your top contact reasons, and add actions one at a time, beginning with something high in volume and low in risk.
Can we run different agents at different levels?
Yes. Each agent has its own job and its own permissions. You can let one agent issue small refunds on its own while another only collects details for your billing team.
What happens when the agent cannot solve something?
It hands the conversation to a human, with a summary and the details it has collected, and routes it to the right team. The customer does not have to repeat anything.
Can our human agents use AI agents too?
Yes. Your team can start an agent from inside a ticket, for example to check an account and suggest compensation that follows your policy. The human reviews the result and sends it.
How are AI agents priced?
Every Theymes plan includes a monthly allowance of AI answers, which covers questions resolved before a ticket is created. Agents that work inside the ticket are an add-on on every plan, sold as monthly conversation packages.

Find out which level your setup is ready for

Bring your top contact reasons. We'll show you which ones an agent can take today, and what the next level would need.