Going beyond AI Agents in customer service

15 Apr 2024

5 min read

In customer service, AI Agents are becoming more and more popular. But what distinguishes an AI agent from the raft of other terms for similar tools and systems, such as chatbots, virtual assistants, customer service automation, generative AI and so on? We even talk about an AI concierge, how is that different?

Different providers will call their technology different things, but we can start to dig around and parse it out a little, and provide a bit more clarity. Along the way, you will hopefully understand the different levels of service and capabilities that you can get in this world. 

What are AI Agents?

In general, these sorts of tools have sprung up to use generative AI tools (similar to ChatGPT, and other tools built on Large Language Models), to answer customer questions. 

How it works is that the domain knowledge of a brand is used to train the AI. This kind of data includes things like returns policies, FAQs, product specifications, and other proprietary information. This could include things on your website, but also things that a customer service agent could have access to, such as more detailed product specs. 

The idea then is that, armed with all this knowledge, the AI agent can give factual answers to questions customers ask. The reason why they need to be trained is so they don't start pulling in information from other sources and giving wrong answers, known as "hallucinations". 

Previous chatbots might use keywords to search for a relevant article that the customer wants and then send them a link, but what these tools can do is digest the information and then rephrase and simplify it. So rather than having to scan a 500 word article or page, customers get the answer in 50 words. 

So for example, if you were asking an AI agent how long the returns policy was, the assistant would scan the relevant document, pull out the answer and rephrase it like: "hey thanks for asking. We offer free returns on all items up to 30 days." This same answer could be rephrased again and again to appear fresh to a different customer. 

It can even remember context and remember previous conversations, though it does need to remember who it is speaking to. This could be done with logged in users, or perhaps with IP identification. In theory you could then have a long-running conversation over many months and years. 

What are the limitations of AI Agents in customer service?

The tools are great for generic questions, by which mean, questions that anyone could ask and that the answer would be true. For instance, that question about returns policy is true for everyone. But if a customer asked "Can I still return my order?" then the assistant may not be able to answer as effectively. The answer "You can, if you ordered in the last 30 days" is fine, but it's not a definitive response. 

The limitations therefore are based on what data the assistant has access to. If the assistant can find the order information, and then calculate the time elapsed since order, it can then give an accurate answer. But that requires more complex set up and integrations. 

The other thing to consider is whether the AI agent actually has access to more information than the customer. For instance, if everything the AI assistant is trained on is public information on the website, how much is it adding to the experience? Of course some customers don't want to sift through page after page of help articles, but could you avoid a lot of the questions the AI assistant is answering if you A) made the information available, and B) made the information more easily searchable. 

Going beyond AI Agents: AI Concierge

It's perhaps an arbitrary difference in terms given that both are AI chatbots, and some tools that are labelled as agents will share functionality with concierges. But let's explain how we think the two terms differ. 

Loosely speaking, an AI Agent is great for telling you things. An AI Concierge is great at doing things for you. But what does that mean? Through integrations with a retailer's technology stack, a concierge can find the relevant information and also perform tasks just like a human agent could. 

For instance, when it comes to returns, the concierge would be able to find the specific order based on your email or order number, and extract all of the items from that order. Then, it could ask the customer to select the specific items they want to return, and then use that information, plus the customer's details, to generate a return label automatically.  

If a customer asks where an order is, an AI agent would send them to a tracking page or share a tracking link. Whereas a concierge would go to the tracking page and extract the status. This means that if the tracking page says "delivered" but the customer says it has not been delivered, the concierge can see the mismatch and start to escalate the situation. Through integrations with order management, the concierge can even order a replacement item, or through connections to payment systems, it can issue a refund, depending on the customer's preferences. 

In short, an AI agent is about arming an AI with as much knowledge as a human agent would have, allowing it to answer simple questions that don't require expertise. In the same way, an AI Concierge has been given many of the skills of a human agent to find the answers, perform actions, and resolve problems. You can watch our demo to see one in action.

​Are both good at Ticket Deflection?

Yes, both AI Agents and AI Concierges will deflect tickets, and both will do it through ticket resolution, which is the best form of deflection. The difference is the range of tickets that both can resolve. Agents can generally only do the generic questions before a purchase, whereas Concierges can do a full range from pre- to post-purchase. 

Do they work on chat and email?

This will depend on the provider, but in theory they can each work on both chat and email, and any other text-based support. However, given that AI agents tend to be best at the pre-purchase side of the customer journey, often they are only deployed on chat. 

Vendor Characteristics to Consider

When evaluating AI conversational platforms, there are a few key factors to keep in mind: 

  1. Use case handling: Can the platform handle your specific use cases and integrations? 

  2. Vendor support: What kind of onboarding, training, and ongoing support does the vendor provide? 

  3. Expertise: Does the vendor have experience in your industry and a track record of success? 

  4. Channels: Does the platform solve across a range of channels, including ones you are already using?

  5. Continuous improvement: How does the platform learn and adapt over time? 


By leveraging the power of AI, businesses can provide better, faster, and more personalized customer service, while also reducing costs and increasing efficiency. Whether it's Agents or Concierges, both can have an impact on the bottom line. 

But here's the thing: it's not just about the technology. It's about the human touch. AI platforms can handle the routine tasks and free up your human agents to focus on the more complex and emotional issues. It's a win-win for everyone.

If you're not already using AI in your customer service strategy, now's the time to start.