Blog Article
AI x CRM
AI Agent
Customer Relations

AI Agents for enhanced customer relations

Charles Dognin
March 25th, 2025

Introduction

You may recall a Google keynote from May 2018 showcasing Google Duplex, an AI assistant capable of performing real-world tasks over the phone. During this demonstration, Sundar Pichai, CEO of Google, asked the assistant to book a hair salon appointment.

The smooth and natural exchange impressed the audience and made headlines, promising a new era where AI would become an everyday assistant.

However, several years later, concrete use cases for Google Duplex remain rare and limited. While the demonstration left a lasting impression, it did not transform the landscape as one might have expected.

Today, the concept of AI assistants has evolved into what we now call AI agents. These agents offer far more plausible use cases and promise to revolutionize entire sectors, particularly customer relations. It is becoming difficult to imagine any aspect of the customer experience that cannot be touched or improved by AI agents.

Anyone looking to make their customer experience smoother, more efficient, and better tailored to expectations should seriously consider AI agents.

In this article, we will explore what AI agents are, how they work, and, most importantly, how to integrate them today to sustainably enhance your customer experience.

What are AI Agents?

AI agents are autonomous systems that leverage artificial intelligence to interpret their environment, make decisions, and perform actions to achieve specific goals.

They combine perception, reasoning, and execution capabilities to solve problems autonomously and efficiently.

To fully understand, it's helpful to distinguish between AI Agents and LLM Prompts:

AI Agents:

  • Are capable of solving complex, multi-step problems (like creating a comprehensive presentation on a specific topic), which involves planning.
  • Have access to tools to help solve these complex problems (such as APIs, Google searches, code interpreters, vector databases, etc.).
  • Possess memory that allows them to recall different steps.

LLM Prompts:

  • Generate text or respond to queries without autonomy, without interacting with the environment, and without access to tools like the internet, code interpreters, APIs, etc.

Numerous use cases

AI agents stand out for their ability to combine planning, execution, and continuous learning. Here are some concrete examples of what they can accomplish:

  • Monitor stock levels in real time and automatically place orders with suppliers.
  • Sort resumes, schedule interviews, and analyze candidates' skills.
  • Manage personal tasks, such as organizing calendars, booking tickets, or handling finances.
  • Analyze customer feedback: Identify trends in customer reviews, detect recurring issues, and recommend improvements to enhance satisfaction.

Many tasks that require thinking, decision-making, and automatic action can be optimized—or even entirely managed—by an AI agent. To fully measure their potential, it is essential to understand how they work and how to integrate them into your processes. That's what we'll explore next.

How do AI Agents work?

AI agents generally follow the same steps, from perceiving their environment to acting upon it.

To make this easier to understand, let's follow the example of a restaurant that has implemented an AI agent to manage reservations via email.

1. Perception: data collection

The first step is to connect your data, which forms the agent's environment (emails, phone calls, sensors, web interfaces, etc.). This step is crucial as it determines which information will be stored and prioritized.

The AI agent monitors the restaurant's email inbox in real time and analyzes each received email to detect reservation requests. It identifies important information in the messages: Customer name, desired date and time, special requests…

"Hello, I would like to book a table for 4 people on Saturday, January 20th, at 7:00 PM. It's to celebrate a birthday—could you arrange for a cake? Thank you, Julien."

The agent automatically extracts key information:

  • Name: Julian
  • Date: January 20th
  • Time: 7pm
  • Number of people: 4
  • Special occasion: Birthday

2. Memory flow: information storage

The memory flow acts as the agent's internal database. It stores and organizes all collected data, including past decisions and actions, with timestamps and descriptions. This allows the agent to quickly retrieve the most relevant information and prioritize recent or critical data.

The agent records the reservation data in its memory system, including special requests. This allows it to:

  • Check availability in the restaurant’s schedule.
  • Maintain a history of reservations for future reference.

3. Memory retrieval: checking availability

When the agent needs to make a decision, it retrieves relevant memories from its memory flow based on their recency, relevance, and importance. This targeted retrieval helps the agent focus on the most useful information to guide its actions.

The agent consults the restaurant’s scheduling system (table management system) to check if a table is available on January 20th at 7:00 PM for 4 people.

  • If a table is available, the agent prepares a confirmation response.
  • If no table is available, the agent searches for alternative time slots close to the initial request.

4. Reasoning: analysis and suggestions

After analyzing the retrieved information, the agent generates complex insights and implications. These reflections are then reintegrated into the memory flow, allowing the agent to improve its learning and adaptability for future decisions.

The agent takes special requests into account to formulate a tailored response:

  • It notes that Julien wants a birthday cake.
  • It includes a proposal in its response to confirm this service and request details (flavor, number of servings).

5. Planning: preparing a response

Here, the agent formulates actions based on the analyzed data and generated insights. The decisions made are also stored in memory to ensure consistency and inform future actions. Careful planning helps the agent act precisely and strategically.

The agent automatically drafts a response email based on the previous information and analysis.

For example, if the table is available:

"Hello Julien, your reservation for 4 people on Saturday, January 20th, at 7:00 PM is confirmed. We have noted that you would like a birthday cake. Could you please specify the flavor and the number of servings? Thank you for your trust!"

6. Acting or reacting: sending responses and adjusting if necessary

In this final step, the agent implements planned actions or reacts to new data if unexpected changes occur in its environment. This dual capability allows the agent to execute predefined strategies while remaining flexible in the face of new challenges.

The agent sends the email to the client and looks out for the response. If Julien accepts the reservation or requests a modification, the agent automatically adjusts the booking and updates the schedule.

  • If Julien replies with the cake flavor details, the agent automatically informs the pastry chef.

As AI agents become more adept at reasoning, planning, and self-monitoring, they will be able to handle tasks that assist users, such as specialized coding, or manage more tedious tasks quickly and at scale.

Enhancing customer experience with AI Agents

Faster, more efficient, and cheaper? Is improving the customer experience as simple as plugging in AI agents? The answer is nuanced.

Data collection: the foundation of everything

Now that you better understand how AI agents work, you can also see how each step can potentially cause issues, such as:

  • Data Collection: If your data is unreliable (lack of standardized formats, spam, duplicates, etc.), the resulting analyses may be incorrect, leading to poor actions (e.g., delivery to the wrong address, poor recommendations).
  • Communication with Tools: Tools can fail or evolve without the agent being updated. Unlike a human who can adapt to an interface or process change, an AI agent requires technical reconfiguration.

Despite these challenges, the benefits of AI agents for customer experience are immense. Major brands are already leading the way by deploying these technologies to create more personalized and efficient experiences.

Deutsche Telekom offers an AI agent named askT, allowing employees to access information about internal policies and benefits, and is considering assigning it administrative tasks.

Cosentino is deploying a "digital workforce" to handle customer service operations, effectively supporting multiple teams and enabling staff to focus on other tasks.

H&M and Amazon use AI agents to assist customers in selecting products that best suit their needs.

AI Agents accessible to businesses of all sizes

Mark Zuckerberg recently stated that in the future, there will be more AI agents than individuals, as businesses, creators, and individuals begin designing their own agents to perform various tasks.

This prediction is based on an emerging reality: more and more solutions, like Microsoft Copilot or OpenAI's agents, are designed to be accessible to both the general public and businesses of all sizes. These technologies even allow small businesses to benefit from the advantages of AI agents, which were previously reserved for large organizations.

Today, a small e-commerce site can implement a chatbot connected to multiple AI agents to manage various tasks such as:

  • Automatically responding to frequently asked questions.
  • Processing orders in real time.
  • Providing personalized product recommendations.

Thanks to these tools, such a site can offer a 24/7 customer experience without needing a large team to cover these services while reducing operational costs.

From customer data to concrete actions

To maximize the potential of AI agents, it is crucial to provide them with a reliable data environment. AI agents can only function effectively if the collected data is:

  • Filtered: Removing spam, duplicates, or unusable data.
  • Standardized: Ensuring information is organized and consistent.
  • Analytically Exploitable: Enabling analysis to generate relevant actions.

This preparation step is critical to ensuring the successful integration of AI agents.

At Glanceable, we have developed a proprietary AI to help you efficiently analyze your customer data, better understand the voice of the customer, and plan the right actions to enhance the customer experience through AI.

It is important to note that AI agents are not designed for large-scale data analysis tasks. For such needs, it is preferable to use custom models optimized to process vast amounts of data quickly and cost-effectively. Indeed, AI agents generally rely on LLMs, which are expensive to host and use for inferences. Even for a small LLM, costs can easily reach several thousand euros per month, making them less suitable for data-intensive use cases.

AI Agents: still in their early days, but already revolutionary

Although these technologies are still in their early stages, their potential is immense. Few companies today fully grasp the range of possibilities offered by AI agents, and even fewer have conducted concrete tests. Unlike Google Assistant presented six years ago, AI agents now have real-world use cases that are already transforming customer relations in major brands.

We hope this article has made their functioning clearer for you and sparked new ideas. So, why not consider testing an AI agent in your business today?

Unparalleled customer feedback analysis within reach

Fill out the form to get in touch with one of our experts and see your future AI partner in action.

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.