AI customer support. Fast answers, human escalation.
The routine majority arrives answered, sourced and ready to send. Your agents spend the day on the cases that need them.
What it does
Answers from your docs
Drafts the reply to a repeat question out of your help pages and past resolutions, with the passage it used attached.
Honest escalation
Anything outside the material it was given goes to a person with the thread and the reason attached, and the draft says why it stopped.
Checkable drafts
Every draft carries its source, so an agent can verify the answer in seconds instead of rereading the documentation.
Night shift
A ticket that arrives at 3 a.m. is read and drafted on the same schedule as one that arrives at noon.
How it works
Receive
A ticket arrives through the form and lands as one record.
Draft
The answer is written from your knowledge base, source attached.
Send
An agent checks the draft against the source and sends it.
Improve
Edge cases go back into the knowledge items the role reads.
24/7
Tickets accepted through your site form, weekends included
0
Answers that reach a customer without a person on the send
60–70%
Routine tickets answered from a draft
Target range we model in pilots
CSAT held
While the routine tickets move to drafts
Target we hold in pilots
What the first month looks like
Week one is shadow mode: the assistant drafts, an agent checks every answer against its source, and you get a real accuracy number. By week two the obvious tickets go out on the draft with a glance — the 60–70% we model in pilots.
By week four, the human team is focused on complex cases, escalations, and relationship recovery — the work that actually drives loyalty.
Evidence · AI support
What runs today
The parts of this role that exist in the platform right now, and the parts a deployment has to build with you. Nothing in between.
Ships today
A support form on your site posts straight into one record
The endpoint runs in production: it caps the request body before parsing it, absorbs bot submissions without writing anything, and stores the ticket in Postgres. A request that fails validation gets a reason back instead of a silent success, and the form on your own site also emails you the moment a ticket lands.
Ships today
Answers are written from your material and sent by a person
Knowledge items are stored per brand and read into the prompt on every run, so two answers a month apart cite the same facts. When the material does not cover the question, the role says so in its own output instead of inventing a plausible policy.
Scoped with you
A helpdesk connection is built during the deployment
One platform connector ships today, and it is Meta ads. Zendesk, Intercom, Freshdesk and the rest are integrations we build against your account during a deployment. We name what we have built and leave the logo wall to someone else.
The gates behind the drafts — approvals, spend caps, the kill switch — are listed on the product page.
Frequently asked questions
Today a person reads and sends every answer, so what reaches the customer is your team writing with a head start. When a role starts replying directly, disclosure goes in the reply template and we set that up with you.
Your help center, PDFs and past tickets become knowledge items the role reads on every run, which takes an afternoon to load. Week one runs in shadow mode so you can measure how often the draft is right before anything depends on it.
Actions that move money or records sit behind the approval gate, and an action the rulebook has never seen counts as needing approval too. By default the role drafts, triages and escalates.
None out of the box today. Tickets come in through a form that posts to our lead endpoint, and a helpdesk connection is scoped and built as part of a deployment.
The agent sees the thread, what the assistant already drafted, and the reason it stopped. The customer never has to repeat themselves.
First-contact resolution, average handle time, satisfaction, and the share of tickets where the draft went out unedited. Every run is journaled with its outcome, so the last of those is a count you can pull.
Yes. The rate is per employee per month, with no overtime and no shift premium. Overnight tickets are read and drafted on the same schedule as daytime ones; sending still waits for whoever is on shift.