AI Connector Tips & Examples

Overview

Once your AI assistant is connected to Quiq through the AI Connector, you can ask it questions about your conversations in plain language. This page covers how to write a request that gets a useful answer, followed by requests you can copy and adapt.

Writing a good request

Your assistant knows your Quiq data but not how your business describes it, so the more specific your request, the less it has to guess. These tips apply to any request, including the ones below.

• Give a time period. "Last month", "yesterday" or a specific date. Without one, the assistant has to guess how far back to look.
• Use the names already in Quiq. Refer to your queues, contact points, AI agents and custom fields by their names in Quiq, for example "the Service queue" or "the topic field from our Conversation Analyst". The assistant then works from your real setup instead of inventing its own categories.
• Define your terms. Say what you mean by words like "resolved" or "complaint", for example "contained means no human agent joined". For wait times, say whether to measure from when the customer wrote or from when the conversation was assigned.
• Say how to group the results. By contact reason, queue, product, or in-hours and out-of-hours. Grouped results are easier to act on than one long list.
• Ask for counts and examples together. The count tells you how big something is. The example conversations show whether the assistant grouped them correctly. Read a few before you change a flow or process based on a number.
• Ask how each number was counted. Check what a percentage is a share of, whether it covers every conversation or a sample the assistant read, and what makes up the rest if a breakdown doesn't add to 100%.
• Compare with an earlier period. Ask what grew or shrank compared with the month before, or how a day compares with a normal day.
• Expect one issue to span several conversations. A customer who comes back later usually starts a new conversation. When it matters, ask the assistant to check the same customer's earlier conversations.
• Follow up on what you find. Start broad, then ask about one group in more detail. If a result looks thin, ask what data the assistant couldn't see.

Requests to try

Replace anything in square brackets with your own details.

Why customers contact you

"What were the top reasons customers contacted us last month? Rank them by how often they came up and give two example conversations for each."

You get your contact reasons ranked, with examples to read.

What your AI agent can't finish

"Of last month's conversations the AI agent handed to a human agent, group them by what the customer wanted. For each group, say whether the AI agent was missing information, access to a system, or permission to act."

You get the handoffs grouped by cause, with what the AI agent would need to handle each group itself.

Customers who leave before anyone answers

"How many customers last month left before a human agent replied, and how long had they waited? Split it by queue and by in-hours and out-of-hours."

You get how many customers left while waiting, and which queues and times of day it happens in most.

What customers complain about

"What did customers complain about last month? Group the complaints by product, service, or process, and show which grew compared with the month before."

You get complaint themes ranked, with the change from the previous month and examples.

Sales opportunities your team missed

"Find last month's conversations where a customer showed buying interest. How many got a reply while the customer was still there, and how many led to a sale or a booking?"

You get how many buying conversations you had, and how many went unanswered or didn't convert.

Why customers cancel

"Find last month's conversations where a customer asked to cancel. Group the reasons they gave and show what happened next."

You get cancellation reasons ranked by how often they came up, with what happened in each and example conversations.

What's behind low survey scores

"For conversations with a low satisfaction score last month, what were customers contacting us about, and what went wrong in the conversation?"

You get the topics and problems behind low scores, with examples.

How often a problem happens

"How often did customers ask for [X] last month? Give me a count and examples."

You get a count and example conversations before you build an AI agent flow for it. For example, how many service requests stall while an agent waits for a serial number or a photo.

What your AI agent tells customers

"Find every conversation last month where the AI agent told customers to [do X], and check whether it made sense for that customer."

You get how often the AI agent gave the instruction, and how often it didn't apply to the customer, such as telling a customer who is already logged in to log in.

A daily report you build by hand

"Every morning we export the Queue Insights report and combine it in a spreadsheet with [these columns]. Pull the same numbers for yesterday."

You get the same table, built from your Quiq data whenever you ask.

What happened on a specific day

"What were customers contacting us about on [date]? Compare it with a normal day and show examples of anything that spiked."

You get what changed that day, with example conversations.