Insights from Implementing 100+ AI Agents: Part 1

Over the past year, we have implemented over 100 AI agents in online stores worldwide. Our bots have answered more than 145,000 inquiries and resolved 86,000 tickets, resulting in an average resolution rate of 59%. We have gathered many insights about bot operations that we want to start sharing with you regularly. Today, we present the first eight, which we hope you will also find interesting.

1. Dealing with Slang Language

Sometimes customers use slang in chats that is difficult to understand, even for a human agent.

The conversation below comes from one of our clients in the fashion industry.

Fortunately, the AI agent understood the customer’s jargon and emotions, leading to an escalation to a human agent.

2. More Content Doesn’t Mean Better

The AI agent can be trained using various sources, such as FAQ pages, past customer tickets, Shopify product details, policies, etc.

Recently, one of our clients asked if we could add their entire blog with all its articles to the bot’s knowledge base.

Until now, the bot has been performing excellently, and we realized that adding such a large amount of unprocessed, duplicate content could do more harm than good.

We proposed to the client that we collaboratively select specific blog articles that, after proper processing, would enhance the bot’s knowledge.

3. Customers Rarely Thank the Bot

Most customers whose issues were correctly handled by the AI agent do not thank it.

This happens in only 20% of conversations where the request was resolved.

4. Business Hours and Bot Activity

Our clients use various scenarios for bot activity throughout the day.

Scenario #1: the bot operates 24/7
Acts as the first responder to help offload the CX team. Ideally, it solves the customer’s problem autonomously. If the issue goes beyond the AI capabilities, the bot transfers the conversation to an agent. This scenario therefore benefits both: clients (who gain a quick response) and human agents, as a significant number of cases are resolved without their participation.

Scenario #2: the bot works only during business hours. This scenario is preferred by clients who want the bot to work only when it is possible to immediately switch the conversation to a live chat agent. The AI handles the repetitive inquiries and escalates the complex issues to human agents.

Scenario #3: the bot works only outside business hours. This scenario is used by brands that want to keep offering the ‘human touch’ CX to their clients but be available 24/7. The conversation with the bot is triggered only when agents are not at work. The escalated inquiries will be solved by the CX team via email during business hours.

5. Immediate Handover for Sensitive Topics

Customers want the ability to immediately take over 'sensitive topics’ (e.g., product returns, complaints, allergic reactions to products) even if the AI bot can answer these questions. What topics are considered sensitive varies greatly and depends on each business individually.

Such topics can be specially tagged in the Gorgias dashboard (see the screenshot below).

6. ‘Was This Helpful’ Helps Maintain the Flow of the Conversation

We introduced a mechanism that asks the customer „Was this helpful?” after providing an answer. We were concerned that this might cause unnecessary escalations, but thousands of subsequent conversations have not confirmed this.

On the contrary, it appears that „Was this helpful?” encourages customers to clarify their questions rather than immediately responding with a firm „no” and escalating the issue.

7. Bots Require Patience

The quality of a bot depends on its optimization and knowledge enhancement. The initial period (often overlapping with the trial offer period) is the most labor-intensive, as the bot requires many updates. This is completely normal. However, it would be unwise to give up at this stage.

8. Bots Often Receive 5-Star Ratings

When customers see that a bot has handled their issue efficiently, they are more motivated to rate the conversation and give the highest rating.

Stay tuned—we will soon share more insights with you. If you’d like to see an AI agent trained on your content, you can try PerfectBot for free with no strings attached.