Product
It doesn't answer questions.
It resolves tickets.
Most support bots write a nice paragraph and leave the work to your team. Wilph looks up the order, creates the label, issues the refund, and tells the customer it's done.
How rollout works
Connect your stack
Forward your support inbox and connect your shop, carriers, and payment provider. Most setups take an afternoon.
Teach it your policies
Your return window, your tone, your escalation rules. The agent works from your process, not a generic script.
Start in copilot
Every reply is drafted for a human to approve. You see exactly what the AI would have sent, on every ticket type.
Graduate to autopilot
As acceptance rates climb, switch ticket types over one at a time. You choose what runs unattended and what never does.
The controls are the product
Handing your customer relationships to an AI is only reasonable if you can draw the line yourself. Here's where the lines are.
Copilot before autopilot
Every ticket type starts as a draft for a human. Wilph tracks how often your team sends the draft unchanged, and only suggests graduating a topic to autopilot once that acceptance rate holds up.
Approval on destructive actions
Looking up an order is free. Issuing a refund, changing an address, or sending an external email can each require a teammate to approve — per channel, per agent.
Escalation rules you write
Refunds over a threshold, mentions of a lawyer, a customer contacting three times about one order. When a rule fires, the ticket goes to a human with a summary attached.
Confidence thresholds
Set how sure the agent has to be before it sends without review. Below the line, it drafts and waits instead of guessing.
What's under the hood
Tags that route everything
Wilph classifies every incoming message against your tag set — order status, return request, damaged item, B2B inquiry. Tags drive routing, reporting, and which topics are cleared for autopilot. Four tags are permanently locked to human review: fraud, legal threats, explicit requests for a human, and spam.
Skills capture your process
A skill is your internal know-how written down: what customers ask, what your policy says, the steps to follow, and when to hand off. The agent reads them the way a new hire would read your support handbook.
A sandbox before anything ships
Replay real historical tickets against a new agent version and see what it would have said. Run a whole batch at once. Nothing leaves the building during a simulation.
Analytics that answer one question
Is the AI actually deflecting work? Automation rate, response time, acceptance per tag, and how the split between AI and human resolutions moves week over week.
What it handles day to day
Order Tracking
Wilph reads the order from your shop, the parcel status from the carrier, and answers with the actual delivery window — not a link to a tracking page.
Returns
The agent checks the return window, creates a prepaid label, and follows up when the parcel is scanned. Exchanges ship before the refund clears.
Refunds
Refunds run against your payment provider inside the limits you set. Anything above your threshold waits for a teammate to approve.
Product Questions
Answers come from your own catalog, product pages, and the policies you've taught the agent — not from guesswork.
Cancellations
If the order hasn't shipped, the agent cancels it and confirms. If it has, it switches to a return flow instead of failing.
Handover to Human
Fraud signals, legal threats, and repeat contacts escalate automatically, with a summary of what's already been tried.
Ready to see it on your own tickets?
Start free, connect your inbox, and watch it draft replies before you let it send anything.