One AI, Two Jobs: Elly in the Customer Chat and in Your Workspace

When a CRM says it has AI, the sensible next question is: AI doing what? In Convelly the answer is one AI, Elly, with two very different jobs. Our guide to what an AI assistant actually does for a business owner draws that line early, because the two jobs share a name and the same customer records, and almost nothing else. In the chat, Elly is part of your customer-facing team. In the app, Elly is part of your back office. Each job has its own limits, its own tone and its own settings.

In the chat: fast, narrow, always visible

This is Elly answering inside a customer conversation. It writes as your business and in your tone, not under its own name. It works from three short texts that describe your business, your rules and your tone, from the documents you upload, and from a set of things it is allowed to do: look up the customer, check availability, book, move or cancel an appointment, send an approved message. It is judged on speed, tone and accuracy, and it is always one switch away from being turned off so a person can take over the conversation.

Its risk is external. A wrong answer reaches a customer, and you may only find out when they complain. That is why this job is built on limits: you choose what Elly may touch in the chat — customers, deals, bookings, documents, the web — and every service has its own switch for whether Elly may book it. Ground its answers in real documents and give it a clean route to hand over. If you have not set those up, start with building a knowledge base your AI can actually use before you switch it on for a live number.

In the app: broad, careful, private

In the app, Elly works for you and your team. Anything you would do in Convelly by clicking, you can ask Elly for: find a customer, add a service, build an automation, move a batch of deals, or ask how last month went. It answers you rather than your customers, so it can afford to be thorough and to show the numbers it looked at. Even when you ask it to message a customer from the app, you see the message before it goes out.

Its risk is internal, and it is the bigger of the two. A chat answer that goes wrong costs you one conversation. A change to a pipeline or a live automation that nobody checked can cost you a quarter. That asymmetry is why approval gates exist for AI changes: looking things up is instant, and every change is shown to you first as a proposal you approve or reject.

“

The Elly your customers meet should be limited. The Elly you talk to can do a great deal — but never changes anything without asking.

— Convelly team

What goes wrong when you mix up the two jobs

Three mistakes come up again and again, and all three come from treating both jobs as one set of settings.

• Expecting one tone for both. Customer replies follow your business texts: short, warm, free of jargon. In the app, Elly answers you directly, in the language you write in, and can be as blunt and numeric as the question needs.
• Expecting Elly's memory to reach your customers. What you tell Elly in the app, such as "we are closed on 24 December", stays in its memory for working with you. What it tells customers comes from your business texts, documents, services and calendar, so a change like that belongs there as well.
• Putting internal-only material in your documents. Cost prices and supplier terms do not belong in an uploaded document that Elly can quote back to a buyer. Your team's private notes are different: Elly reads them before it answers, but never reveals or hints that they exist.

How to divide the setup

Set up each job in its own sitting, with its own checklist. For the chat, decide which services Elly may book, what it may touch, what it must never discuss — pricing exceptions, medical advice, legal commitments — and whether new conversations on each channel start with it switched on. Test it like a customer before it touches a live conversation; testing an AI before you let it talk to customers covers a workable method.

For the app, the questions are different. Everyone on the team can work with Elly, but a few changes stay with admins: inviting people, creating, changing or deleting a pipeline, and connecting a channel. Tell Elly the facts about your business and the preferences it should keep in mind; you can read and edit that memory at any time.

Where the two jobs meet

Both jobs work from the same customers, conversations, deals and bookings, and that is where they help each other. In the app, ask Elly to read through last week's conversations and point out where customers did not get a clear answer. The way to phrase questions like that is covered in asking your CRM questions instead of building reports. It is the fastest way to find the gaps in what Elly knows, and the fixes land where customers feel them: in your business texts, your documents and your services.