A support agent that fixes it, not just answers it.
Answers your customers and fixes it in your systems, instead of linking an article.
$20 of free credit to try it. No card.
- CustomerI was charged twice for order 58213.
- AgentI found the duplicate $120 charge in billing. A teammate is approving the refund now.
- AgentDone: refunded to your card. You’ll see it in 3 to 5 days.
The problem
Help-center bots answer from articles. When the fix is a refund, an address change or an account reset, they hand off to a person, and the customer waits.
The answer was never the slow part. The fix lives in a billing console, an order system or an admin portal, often with no API, and only a person with the right login can make it. So the bot deflects, a ticket opens, and the customer explains it all again.
An AI support agent that takes action closes that gap. It follows your team’s procedure, makes the change in your systems, and asks a person first when the change is risky.
How it works
Hand it your help center, macros, old tickets and policy docs.
Files go in as they are. Old tickets teach it the most: what customers really ask, and how your team really resolved it, exceptions included.
Coach it on how your team handles each kind of request. It writes a procedure for each.
A refund, an address change and an account reset each get a playbook: when it applies, the steps, what to check first and which file each step came from. Your team reviews each change in place before keeping it.
It answers, and when the fix lives in an admin console or a portal, it makes it, asking a person first when you say so.
It uses your APIs where you have them, and a cloud browser or one of your computers where you don’t. Guardrails set what needs a teammate’s approval and what never happens, and every step can be replayed.
A worked example
A customer writes in through the chat widget on an online store’s website. The store, the order and the address are made up.
What it did
- Found its procedure for address changes: before an order ships, change it; after, start a carrier redirect instead.
- Opened the order in the store’s admin console in its cloud browser and saw it hadn’t shipped.
- Changed the address. Your guardrails let it change unshipped orders without an approval, so it went ahead.
- Read the order back to confirm the change. Your team can replay each step from the conversation.
- CustomerI moved last week and my order still shows my old address. It’s order 70412.
- AgentIt hasn’t shipped yet, so I can change it. What’s the new address?
- Customer22 Harbor Lane, Portland, OR 97209.
- AgentDone: order 70412 now ships to 22 Harbor Lane, Portland, OR 97209. Your tracking email will show the new address within the hour.
What Coach asks you
Coach reads your help center and tickets first and sets itself up in one change you review. Then it asks about what they leave unclear, in a few short questions. For a support agent, they sound like this.
Each answer becomes a rule in the procedure for that kind of request. You see the change in place before you keep it.
- AgentYour refund macro says 30 days, but the policy says 14 for sale items. Which applies?
- AgentCan I change the address on an order that has shipped, or should I start a carrier redirect?
- AgentWhich refunds can I make myself, and which need a teammate’s approval?
- AgentWhat should I check before I reset someone’s account?
- AgentWhen should I hand a conversation to a person instead of resolving it?
How evals check it
Test it on real requests before customers see it. A test case is a request and what a good outcome is. A judge checks that the answer is right and backed by what the agent found, and for changes, a check script can confirm the change really happened. Each case runs three times.
| The request | A pass |
|---|---|
| “I was charged twice for one order.” | A pass: Finds the duplicate charge, asks for approval, and refunds it once. |
| “Refund the sale item I bought 20 days ago.” | A pass: Declines, citing the 14-day sale policy, and offers store credit as the policy allows. |
| “Change the address on my order (already shipped).” | A pass: Doesn’t edit the order; starts a carrier redirect instead. |
| “Reset my account. I lost access to my email.” | A pass: Verifies who they are as the procedure says before changing anything, or hands it to a person. |
Each try keeps its conversation and the replay of anything it did in a browser or on a computer, so a failure shows you where it went wrong. Re-run the set after every change.
Why Eager
- It resolves requests instead of deflecting them.
- It works in the tools your team already uses, API or not.
- It asks before anything risky, and every step is on replay.
- Where it answers: The chat widget on your website, or your app through the API.
Questions
Read more
- How is it different from a help-center chatbot?
- A chatbot answers from articles and hands the rest to a person. This agent follows your team’s procedures and makes the fix itself: the refund, the address change, the account reset.
- Can it make refunds and account changes in our systems?
- Yes, in your billing, order and admin tools: through their API when there is one, in a cloud browser when there isn’t, or on one of your computers for desktop apps.
- Will it refund someone without asking?
- Only if you allow it. Guardrails make an action wait for a teammate’s approval, or never happen at all. Approvals show up in the conversation, where your team decides them.
- Where does it talk to customers?
- In the chat widget on your website (one script tag), in your app through the API, or behind the voice agent you already run.
- What if a customer asks something it wasn’t trained on?
- It says what it can do, or hands the conversation to your team, as you taught it. Afterwards your team can coach it on the new request, and it writes a procedure for next time.
- Can we see what it did?
- Every conversation is kept with each step the agent took (90 days unless you set otherwise), and anything it did in a browser or on a computer can be replayed.
- What does it cost?
- You pay for the tokens it uses, from a prepaid balance, at the price of the model it runs on. Your first team gets $20 of free credit, no card.
Other jobs it learns
Sales copilot
Your reps ask. It knows every product, policy and portal, and says what to recommend.
Behind your voice agent
Your voice agent talks. Eager does the work in your systems while the caller holds.
On your customer’s computer
It does the fix on their Mac or PC while they watch, with Stop always in reach.
Patient intake
Finds the chart, checks coverage in the payer’s portal, and tells the patient what they owe.
Freight
Reads the truck’s live location, converts its time zone, and moves the pickup in your shipping system.
Claims
Catches the swapped digit in a case number, reads the right report, and keys the claim.
Payables
Checks a bank-change request against the supplier’s number on file before anything changes.
Start with one job.
Pick the request your team handles most. Hand it the files. Watch it work before you spend a cent.