For support and CX teams of three to thirty
An AI front line your team supervises. Not a bot your customers fight.
OU Chat resolves the repetitive questions from your own documentation with the source cited, escalates everything else into a shared inbox with the transcript attached, and gives your leads the roles and the audit trail to run it across every channel.
Cited answers · Escalation with the transcript · Roles, permissions and an audit trail
Six responsibilities, and the queue is only the visible one
Volume is the part everyone sees. The rest of the job is what makes volume survivable, and it is the part that quietly stops happening when the queue wins.
Cover the hours
Shifts, breaks, the gap between the last person logging off and the first one logging on, and the weekend nobody wants.
Hold the SLA per priority
Messaging channels are judged in minutes, not the hours an email queue was allowed. Missing it quietly is worse than not promising it.
Quality, not just throughput
Reading replies, coaching the newer agents, catching the answer that was fast and wrong before a customer does.
Keep the documentation true
Every changed price, policy and process. Macros drift, help centres go stale, and the queue absorbs the difference.
Own the escalation path
Who takes a refund dispute, who takes an outage, who wakes up for the account that cannot wait until morning.
Explain it to the business
Why headcount, why the backlog, what customers keep asking for. All of it needs numbers you can defend line by line.
Volume grew three times over. Headcount did not.
Your agents paste the same macro into WhatsApp, Instagram and email in three separate tabs. The SLA holds until six in the evening and collapses after. Nobody can say which ten questions make up half the queue, because the queue lives in four tools and none of them agree.
Meanwhile the last "just add a chatbot" pilot produced angry customers and a widget the team learned to route around, so the bar for the next attempt is higher than it was for the first.
A resolution you can check, not a number we chose
Resolved means the customer did not ask for a person and did not come back within 72 hours. That definition is the same for every workspace, so the figure in your console is the figure we would quote, and you can audit it against your own transcripts.
The four on the right are the job. Everything faded above them closed without a person and without coming back, which is the only version of deflection worth reporting. Per-topic breakdown is not something we ship yet, so we do not put it on this page.
Escalation that respects the person receiving it
Most AI handovers arrive as a customer repeating themselves to a human who has no idea what just happened. These arrive as work in progress.
Everything attached
The transcript, the language the customer writes in, who they are, and what the agent already told them, so nobody opens with a request to explain the issue again.
Routed, then owned
A conversation is claimed by a named person, shows an assignee and a timestamp, and closes when it is genuinely done rather than when a bot decides.
Roles with a matrix behind them
Owners, managers and agents, with per-teammate permissions rather than an honour system, and an audit trail that answers who replied and when.
Quiet is a valid state
If nothing needs a person, the queue is empty. An empty inbox is the system working, and it is worth saying so before someone reports it as a bug.
The answers your security review will ask for
Your data is not training data
Encrypted in transit and at rest, never used to train a model, and exportable in full whenever you ask.
Your provider keys, if policy requires it
A workspace can run on your own LLM keys, which matters when your compliance team has already approved one vendor and will not repeat the exercise.
Access that matches the org chart
Role-based access with per-teammate permissions, so the intern sees the intern's view and the audit trail records the difference.
Regional hosting is a conversation
There is no region picker in the product. If your jurisdiction requires one, raise it before signature and we will tell you honestly what we can do.
What we are not claiming
You have been sold this category before. Here is the part that usually appears in week three, in advance.
No review queue
There is no mode where the agent drafts and a human approves before sending. It answers from your content or it escalates, and any tool promising both should be asked to show it.
No per-topic deflection report
You get resolution, CSAT and automated QA review. A breakdown of which topics deflect best is not built, so do not plan a knowledge roadmap around it yet.
Webhooks, not connectors
Escalations reach your existing desk over webhooks. Naming a helpdesk we have not built an integration for would be the fastest way to lose your trust.
Bad documentation, bad answers
The agent is as good as what it was given. If your help centre is two years stale, the first month is a content project wearing an AI project's name.
Straight answers
How do you stop it making things up?
It answers from your content and cites the page it used. When the answer is not in what you gave it, it escalates instead of guessing, and the escalation carries the thread so your agent does not restart the conversation.
Can we start in shadow mode?
Not as a product feature today, and we would rather say so than sell you a switch that does not exist. What teams do instead is start on one channel or one topic set, read the transcripts in the inbox daily for a week, and widen from there.
How does it work with the helpdesk we already run?
Your team can work directly in the OU Chat inbox, or escalations can be pushed to your existing tool over webhooks. There is no native Zendesk or Freshdesk connector, so we do not list one.
Where is our data stored, and is it used for training?
Encrypted in transit and at rest, and never used to train a model. If you need a specific region, raise it before you sign: there is no self-serve region picker, so it is a conversation rather than a setting.
What does migration actually look like?
Point it at the documentation and macros you already have, connect one channel, and run it alongside your current desk. Nothing has to be switched off to find out whether the answers are good.
What happens to our SLA reporting?
The inbox records who claimed a conversation and when it closed, so first reply and resolution are measurable per channel. Reporting today covers resolution, CSAT and automated QA review of transcripts.
Bring your hardest queue
Twenty minutes, your channels, your escalation rules and your team roles mapped with you. If the answer is that your documentation is not ready, we will say that instead of selling you a pilot.