Confirms it is really them.
Verifies the customer against your own records before it touches anything. Email match, order ID, last-four, magic link. Your rules. No identity, no action.
Identity confirmed. Email and last-four matched.
Auto-Resolve
Most customer support AI agents read your help center back to the customer. Kal authenticates them, calls Stripe or your CRM, runs the action, and closes the ticket. No human in the loop.
The difference
An AI chatbot that quotes the manual
never cut your headcount.
The market is flooded with AI chatbots that summarize help articles. They deflect a question and call it a day. Your queue, and your payroll, never noticed. Resolution is a different job.
Summarizing chatbot
"Here is how to request a refund."
Settings, then Billing, then submit a request. The ticket is still open and a rep still has to do the work.
Resolution layer
"Done. I've refunded $49.00 to your card ending 4242."
The money moved in Stripe, the customer was told, and the conversation closed. Nobody on your team opened it.
Refunded $49.00. Closed in 41s, logged.
The loop
Ticket in.
Ticket closed.
Four steps run between the customer's message and a resolved conversation. Here is one $49 refund moving through every one of them.
Checks if it is allowed.
Reads the request against your policy. Inside the refund window? Plan eligible to downgrade? It only proceeds when the answer is yes. Everything else routes to a human.
Eligible. 6 days into the 30-day window.
Makes the API call.
Fires the real action in Stripe, your billing system, or your CRM. A refund is a refund. A downgrade is a downgrade. Not a draft for someone to copy-paste.
POST /v1/refunds. Refunded $49.00 to •••• 4242.
Replies and closes the ticket.
Tells the customer it is done, logs every step, and marks the conversation resolved. Your team never opens it. Most finish in under a minute.
Resolved in 41s. Full log saved.
No human ever sees
the easy ones.
The refund. The downgrade. The address fix. They come in, get done, and close. Your team only works on the conversations that actually need a person.
Actions, not answers
The tickets that should
never reach a person.
Hooked straight into the systems where the work happens. Each one authenticated, executed, and closed.
Stripe
Refunds
"I want my money back." Inside the window and within policy, Kal issues the refund and confirms the amount and card.
Refunded $49.00 to •••• 4242
Billing
Plan changes
Upgrades, downgrades, seat changes. Kal updates the subscription, prorates it, and shows the new amount before the next invoice.
Removed 2 seats, prorated −$38
CRM
Account updates
Email, billing address, company name. Kal verifies the request and writes it straight to the record. No internal ticket.
Email updated, record synced
Orders
Shipping fixes
Wrong address caught before it ships? Kal updates the order and tells the customer the corrected destination.
Address corrected before dispatch
Billing
Cancellations
Cancels the subscription, sets the end date, and confirms what they keep until then. Logs the reason for your retention team.
Canceled, active until Jul 2
Stripe
Invoices and receipts
Resends the invoice, generates a receipt, updates the VAT number. The errands that eat an L1 queue, gone.
Invoice #2231 resent
What counts
Resolved means resolved.
Not "they stopped asking."
Some AI tools mark a ticket resolved the moment a customer goes quiet, then put it on your bill. Silence is not a solution. Here is the standard Kal is held to, and the only thing we ever charge a resolution for.
Counts as resolved
- Kal ran a real action, a refund, a plan change, an account update, and the system that did it confirmed it.
- Kal gave a verified answer with a citation to the source, and the customer got their actual question answered.
- The conversation closed. Not deflected to a help article. Not left half-done for a rep to finish.
Never counts
- The customer went quiet. A timeout is not a resolution.
- Kal handed the thread to a human. Handoffs are free, every time.
- A canned "did this help?" with nothing behind it.
$0.49 per resolution, and we only count one when the left column is true. Half of Fin's $0.99, billed on a standard you can hold us to. See pricing
The math
Automate 30% of your actionable tickets,
and it pays for two of your L1 reps.
A mid-market support org spends most of its L1 hours on refunds, plan changes, and account edits, the exact work Kal can finish end to end. Take those off the queue and you stop backfilling two full-time seats.
30%
Actionable tickets resolved with zero humans
2
L1 seats you stop backfilling full-time
0
Humans in the loop on resolved tickets
<60
Ticket to closed seconds, typical
Figures are illustrative of a typical mid-market support org and depend on your ticket mix and policies. You pay only for resolutions Kal actually closes. See plans and per-resolution pricing.
Trust
Autonomous.
Not unsupervised.
Letting an agent touch billing is only a no-brainer if it stays inside the lines. Kal does. By design, not by hope.
Authenticates first, always
No action runs on an unverified person. You set the proof you require, and Kal will not move without it.
Scoped to what you allow
Refund up to a ceiling. Downgrade but never delete. Kal acts inside the exact limits you grant. Nothing past them.
Every step is logged
A full audit trail of who asked, what ran, and which API responded. Export it for finance or compliance any time.
Hands off when unsure
Edge case, angry customer, anything outside policy. Kal stops and passes the full thread to a teammate. It never guesses.
The guarantee
No unauthorized actions. Ever.
Kal never runs an action you did not grant it. Every step, who asked, what was checked, which API answered, and what came back, is written to an audit trail you can export for finance or compliance any time. If it is not in the log, it did not happen.
The receipts
See exactly
how Kal decided.
Every action comes with its reasoning attached. Open any resolved ticket and read the trail. Here is the $49 refund from earlier, start to finish.
Audit trail · ticket #10472 Resolved in 41s
Ali K. · 14:02:10 · "I want my money back. Order #10472."
- 14:02:11
- 14:02:11 Read intent parsed
Refund request · $49.00 · frustrated tone. Routed to the refund policy.
- 14:02:11 Check policy eligible
30-day refund window. Order placed 6 days ago, so it qualifies.
- 14:02:12 Check scope in scope
Refunds up to $100 are auto-approved. $49.00 is inside the limit you set.
- 14:02:12 Score confidence 0.96
0.96 against your 0.85 threshold. High enough to act without a human.
- 14:02:12 Execute Stripe
POST /v1/refunds → Stripe → $49.00 to •••• 4242.
- 14:02:12 Reply and close resolved
Confirmation sent to the customer. Conversation marked resolved.
- 14:02:12 Write log logged
Full decision trail saved, immutable and exportable for finance.
Refunded $49.00 · no human touched this ticket · trail exportable
FAQ
Questions.
Honest answers.
What teams ask before they let an AI agent touch billing. Written by the people who built it.
Yes. Auto-Resolve is an execution layer, not a chatbot. Kal authenticates the customer, checks the request against your policy, runs the action in Stripe or your CRM, and closes the ticket. No human touches a resolved conversation. Anything outside policy is handed to a teammate with the full thread.
Refunds, plan changes, cancellations, account and address updates, invoices and receipts. The repetitive Level-1 tickets that fill a support queue. Each one is verified, executed in the real system, and logged. Job-level walkthrough: billing and refunds. This is resolution, not ticket deflection that leaves the work for a rep to finish.
You pay per resolution Kal actually closes, on top of $19 per seat per month. The first 50 resolutions each month are included, then $0.49 each. 14-day trial, no card required. See plans and per-resolution pricing for the full breakdown.
Both. A small business gets an AI helpdesk that closes routine tickets from day one, without hiring an L1 rep. A mid-market team stops backfilling seats. You set the scope and the limits; Kal stays inside them and hands off anything it is unsure about. See smart handoff for how escalations work.
Updated June 2026