What Is an AI Copilot for Merchants in Support Ops

Monday morning you want to know how support held up over the weekend. The old way: open the dashboard, click into reports, pick a date range, wait for the chart to load, do the comparison math against last week yourself. Now you can just ask “how did weekend volume and resolution rate compare to the week before,” and the answer comes back in seconds, calculated live from your own data instead of pulled from a canned weekly report.

The thing answering isn’t the AI agent—that’s the layer facing your customers, and it doesn’t talk to you about your business metrics. The one answering is Yuna, a merchant-facing copilot that only talks to you, never to your customers. This piece breaks down what that actually is, what it does for you, and where its edges are.

Two AIs, two very different jobs

YundaDesk runs two layers of AI, and mixing them up is where confusion starts.

The AI agent faces your customers. It answers around the clock by pulling from your knowledge base, and it hands off to a human whenever it can’t find an answer, a customer asks for one, or the request touches something high-risk like a refund. It doesn’t know you—it only knows the customers asking it questions.

Yuna faces you—the store owner, the support lead. It never touches a customer conversation, and customers have no idea it exists. Think of it as an assistant sitting next to you who knows your store’s data cold: you ask, it answers or acts, but it never makes the call for you.

One faces outward to customers, the other faces inward to you. That’s the line that matters most when you’re trying to understand what a merchant-side AI copilot actually is.

What Yuna does #1: answer questions about your numbers

The most direct use case is treating it like a report you can talk to. No hunting for the right menu, no fiddling with date pickers, no doing subtraction in your head—just ask in plain language:

• “What’s this month’s resolution rate compared to last month?”
• “How has response time on WhatsApp been lately?”
• “Which three questions come up most often but the AI handles worst?”

The value isn’t that you can see the numbers—your dashboard already has that. The value is turning “click through pages to find a number” into “ask a question and get an answer,” and the answer keeps up with your follow-ups in the same thread. Ask about resolution rate, then ask whether one channel is dragging it down, and it can keep digging in the same conversation instead of sending you back to a filter panel.

What Yuna does #2: surface specific conversations and anomalies

Beyond summary numbers, Yuna can also help you locate specific conversations out of thousands. Say you suspect customers were unhappy with how returns policy questions got answered last week—instead of paging through logs, just ask “were there complaints about the returns policy response last week” and it pulls the relevant conversations for you.

This is especially useful in two situations:

1. Reviewing a specific event—after a promotion wraps, you want to know what questions spiked and where the AI struggled.
2. Catching signals that get missed—a question your team has been manually answering dozens of times but nobody ever formalized into the knowledge base. That kind of gap is hard to spot by scrolling through logs; a single question to Yuna can surface it.

What Yuna does #3: reconfigure settings by conversation

Configuration used to mean figuring out which submenu a setting lived under, clicking in, and matching fields against documentation. Now you can just describe the outcome you want, and Yuna translates that into the actual configuration.

What conversational configuration actually saves isn’t clicks—it’s the mental translation from “what I want” to “where that lives in the system.” You don’t need to memorize a menu tree, just describe the outcome.

What Yuna does #4: feed your team’s know-how back into the AI agent

This is where Yuna connects most directly to our “gets smarter with every use” loop. When the AI agent can’t answer something, or an agent corrects one of its replies, the system generates a pending learning suggestion—and Yuna can help you process that backlog more efficiently.

You can ask Yuna to “pull together all the learning suggestions about shipping delays from last week,” and it groups the scattered items sitting in your review queue so you can judge them in a batch instead of clicking through one at a time. The actual decision—which suggestion gets adopted and goes live—stays with you. See our piece on teaching the AI that gets smarter for how that loop works end to end.

Yuna’s role here is organizer, not decision-maker. It puts the right things in front of you and cuts down the time spent hunting and sorting, but the final approval is always a human action.

What Yuna doesn’t do

The boundaries around a merchant-facing copilot are just as clear as they are for the AI agent—arguably stricter:

• It never touches the customer conversation itself. Yuna doesn’t reply to customers on your behalf; it only serves you.
• It never executes sensitive actions automatically. Refunds, compensation, and price changes always require human approval—Yuna won’t act on them just because you mentioned them in passing.
• It never adopts a learning suggestion on your behalf. It can organize and summarize what’s pending review, but clicking “adopt” is a human action.
• It’s not a universal natural-language query tool. It’s scoped to operations, support, and knowledge base tasks—not a general assistant wired into every system you run.

Why a conversational layer is worth building at all

Someone might ask: reports and settings pages already have this functionality, so why bother wrapping a conversational assistant around them?

The answer is entry cost. Owners and support leads juggle too many things to memorize every menu path, especially for settings they touch rarely. Traditional interfaces require you to already know where a feature lives before you can use it. A conversational interface flips that—you only need to know what you want, and Yuna maps it to the right place.

It’s the same design philosophy behind the one-click AI-to-human handoff in the shared inbox, which we cover in omnichannel inbox, explained: good tooling should lower the cost of finding the right action, not dump that complexity on the person using it. What Yuna does is take the most cognitively expensive part of running support—remembering where things are—and turn it into simply saying what you need.

The AI agent talks to your customers. Yuna talks to you. That’s the simplest way to define a merchant-side AI copilot. It doesn’t decide, doesn’t touch money, doesn’t approve anything on your behalf. What it does is turn data into answers, gather scattered signals into a list, translate your intent into configuration, and put the right learning suggestions in front of you. The time you get back is the actual product.