For customer support and CX leads
Site chat, support email, social comments and messages, and the calls your team logs – in one queue, each conversation on the customer's record. Around the clock, the desk answers the kinds of question your team has switched on, and hands everything else to a person.
Where the day goes
A customer writes in about an order. Before anyone can reply, someone has to find out who this is: the store's back office for the order, the CRM for the history, the inbox for the last complaint. By the time the context is on one screen, the customer on the chat has left the page.
Questions arrive at three in the morning the same as at noon. Most of them are answered in documents your team already wrote: delivery times, payment methods, the terms of this month's promotion. They still wait until someone logs on, and the first reply of the day goes to a customer who asked hours ago.
A bot that answers everything will one day answer a chargeback, a legal threat or a request to self-exclude as if it were a question about opening hours. So teams switch it off, or shrink it to a menu of links, and every message comes back to a person, sensitive or not.
The customer who complained on Tuesday gets the win-back offer on Wednesday. The desk cannot see the campaign and marketing cannot see the complaint, so an unhappy conversation that should have paused a send, or started a save play, ends as a ticket marked resolved.
What you get
Site chat on your own website, your support email, comments and messages from Instagram, Facebook, YouTube and Reddit, App Store and Google Play reviews, and the calls and visits your team logs by hand. Views for Mine, Unassigned and Needs reply, and every row shows how long the customer has waited.
Lifetime spend, orders, segment, last purchase and every previous conversation sit beside the thread. A chat from a customer your site signs in arrives with their record attached; a name the site did not sign shows as not verified, so nobody discusses an account with whoever typed an address.
Observe: a suggested reply when someone asks for one. Draft: a reply waiting on every new message, and nothing sent without a person. Auto: it answers on its own from the knowledge you marked for customers, never guesses a price or an opening time, and brings in a person after three replies.
The Customer Support Agent starts by learning: it writes the reply it would have sent, keeps it to itself and compares it with your team's. When it believes it is ready, it says so in numbers you can check. Whether it answers on its own, and on which topics, is an administrator's decision.
Add your terms, payment rules and help-center pages, each marked for customers or internal only, with an end date when a promotion closes. Every draft names the passage it came from, and every question it could not answer is listed, so your team answers it once.
Contacting support, an issue resolved, a customer turning unhappy and a rating are events a journey can start on. When a high-value customer's tone turns negative, the card raises a banner with your active promos one click away; a self-excluded or blacklisted customer never receives one.
How the night runs
From the chat, the inbox or a social account, it lands in the queue on the customer's record. A request to self-exclude, a sign of harm, a chargeback or a legal threat never reaches a model: it goes to a person, with a holding reply prepared.
On a topic an administrator switched on, the agent answers in the customer's language, from your knowledge. Anything else waits in the thread as a draft for the first person on the desk to approve, edit or discard.
Your team opens the desk on the conversations that need a person – each with a note saying why, high priority and a draft ready. Replies sent overnight sit in their threads with the sources they used, and Reporting counts them apart from your team's.
Governance
Topics you switch on
the agent answers alone only the kinds of question an administrator turned on
Sensitive matters to a person
self-exclusion, harm, chargebacks and legal threats never reach a model
Identifiers replaced first
the customer's name, email, phone and customer number become placeholders before a model reads
Confirmed in writing
drafts on arrival and answers on its own wait for an administrator to confirm the AI support addendum
The business case
“Choosing The AI CMO represents an important step in building a marketing operation that can scale across our online and land-based businesses. Its unified approach will allow us to coordinate acquisition and retention across brands and markets, increase the leverage of our teams, and measure impact against incremental margin.”
Vojtěch Krňoul
Director of Online Casino · NET and GAMES
Questions
An AI helpdesk is a helpdesk where AI works inside the queue your team already runs: it reads each message with the customer's record beside it, suggests or drafts the reply and, where your team allows it, answers on its own. In The AI CMO it answers only from the knowledge you marked for customers, and hands anything uncertain or sensitive to a person.
Yes, on the kinds of question an administrator has switched on, once the Customer Support Agent has been moved to its third step. Everything else waits as a draft for a person. Even on a switched-on topic it hands over when it is not sure, when it cannot back the answer with what you gave it, and after three automatic replies in a row.
Site chat on your own website; support email by forwarding, a hosted address, Microsoft 365 or your own mail server; comments and messages from Instagram, Facebook, YouTube and Reddit; App Store and Google Play reviews; and the calls and visits your team logs by hand. Phone calls are not answered by AI – logging one on the desk takes about twenty seconds.
It never reaches a model. It goes to everyone who takes the brand's tickets, marked as needing a person, with a short holding reply prepared for them to review. On a chat, the AI stays paused on that conversation until someone on your team allows it again.
Before a conversation reaches a model, the customer's name, email address, phone number and customer number are replaced with placeholders, and so is anything in the text that looks like an email, phone, card or bank account number, or an identity or document number – best effort for free text. The values are put back on our servers when the draft returns. No customer data is used to train models.
Yes. Site chat runs on the Website SDK script, in your colors and your visitor's language, and you can try it on your real site in Test before anyone else sees it. If a visitor leaves before reading your reply, it can reach them by email once your brand has a support mailbox.
The Helpdesk and Site chat are part of the platform, not a separate product to buy. Manager is $799 a month for one brand, and Enterprise starts at $80,000 a year; what each plan includes is on the pricing page. The Customer Support Agent's work is charged in credits per piece of work, from the brand's own budget.
Read next
See the desk on your own channels – arrange a walkthrough, or talk to us about enterprise volumes.