Episode 5 of 9 AI for Managers

Start from the beginning

AI in Customer Service — Chatbots and Beyond

Episode 5 15 minutes

Customer Service: Where AI Makes Immediate Impact

Customer service was one of the first areas AI entered — and one where it has had the most success. The reason is simple: a large portion of customer questions are repetitive, being available 24/7 is expensive, and response speed directly affects customer satisfaction.

But — and this but is important — AI does not work magic in customer service. If implemented poorly, it drives customers away. In this episode I will tell you where chatbots work, where they do not, and how to build a hybrid system that both reduces costs and keeps customers happy.

Old Chatbots vs AI Chatbots

Old Chatbots (Rule-based)

You may have had a bad experience with chatbots. Those chatbots that responded with I did not understand to everything you typed. Those were rule-based chatbots — they had a fixed list of questions and answers and only responded if you typed the exact same words.

AI Chatbots (LLM-based)

The new generation of chatbots is built on Large Language Models — the same technology behind ChatGPT. These truly understand natural language. A customer can ask their question in any phrasing and the chatbot understands it.

The difference is like the difference between a receptionist who only knows four sentences and an experienced receptionist who gives a logical answer to any question.

Key difference: An old chatbot only answers questions that were previously defined for it. An AI chatbot can answer new and unpredicted questions — provided it has the necessary information.

Where Do AI Chatbots Excel?

1. Frequently Asked Questions (FAQ)

What are your business hours? What is the return policy? How do I register a return? These questions might make up 60 to 70 percent of customer service contacts. An AI chatbot answers them instantly, accurately, and 24/7.

2. Order Tracking

Where is my order? is one of the most common questions. An AI chatbot connected to the system takes the tracking code and provides the exact status. No human operator needed.

3. Initial Guidance

Which product is right for me? The chatbot can ask a few questions and suggest the appropriate product based on the customer’s needs — like a salesperson in a store.

4. Collecting Initial Information

Before a call is connected to a human operator, the chatbot collects initial information: name, order number, type of issue. This means when the operator takes the call, all the information is ready and they can help faster.

5. Multilingual Support

If you have international customers, an AI chatbot can respond in multiple languages without needing to hire foreign-language operators.

Where Do AI Chatbots Fail?

Read this section carefully because it is more important than the previous one:

1. Serious Complaints and Angry Customers

When a customer is angry and has a serious complaint, the last thing they want is to talk to a robot. Even if the chatbot gives the right answer, the customer feels they are not being valued. A human is absolutely necessary here.

2. Complex, Multi-step Issues

Problems that require checking multiple systems, special decision-making, or exceptions — humans still solve these better.

3. When Mistakes Are Costly

In areas like banking, insurance, or healthcare, a wrong answer can cause financial or legal damage. Here AI should only be a helper, not a decision-maker.

4. When the Customer Explicitly Wants a Person

Some customers are simply not comfortable with chatbots. If a customer says I want to speak to a person, connect them immediately. Nothing is worse than a customer wanting to escape the chatbot and not being able to.

Golden rule: Always — I emphasize always — have a clear and accessible connect to a human operator button. The chatbot should be a helper, not a gatekeeper.

The Hybrid Approach: Best of Both Worlds

The best customer service system is neither fully AI nor fully human — it is hybrid. Let me show you a practical model:

Tier 1: AI Chatbot (Front Line)

  • Answer frequently asked questions
  • Order tracking
  • Initial guidance
  • Collect customer information
  • Categorize request type

Goal: 60 to 70 percent of requests are resolved here.

Tier 2: Human Operator + AI Assistance

  • More complex issues that the chatbot could not resolve
  • AI helps the operator: previous conversation summary, suggested response, relevant customer information
  • The operator makes the final decision

Goal: Operator speed and quality doubles.

Tier 3: Specialized

  • Serious complaints, legal issues, VIP customers
  • Connected directly to a senior specialist
  • AI only documents and records

The best customer service is when the customer feels they matter — whether AI answers or a human does. Technology is the tool, customer feeling is the goal.

Sentiment Analysis: Understand How the Customer Feels

Sentiment Analysis is one of the most fascinating applications of AI in customer service. AI can determine from the text of a customer message:

  • Whether the customer is happy (e.g., that was great, thanks)
  • Whether the customer is unhappy (e.g., terrible experience, never buying again)
  • Whether the customer is angry (e.g., what kind of service is this?!)
  • Whether the customer is confused (e.g., I do not understand what to do)

How to Use It?

When AI detects a customer is angry, it can automatically raise the ticket priority and connect the customer to an operator sooner. Or when it detects a customer is happy, it can ask them to leave a positive review or refer friends.

At a broader level, sentiment analysis across thousands of customer messages gives you an overall picture of customer satisfaction — better than any survey.

Practical example: A service company using sentiment analysis discovered 40 percent of dissatisfied messages were related to a specific problem in the delivery process. They fixed that issue and customer satisfaction increased 25 percent — without hiring a single additional person.

Smart Ticket Routing

Another area where AI excels in customer service is ticket routing. In traditional systems, someone has to read each ticket and send it to the relevant department. AI does this automatically and more accurately:

  • Automatic categorization: AI determines from the ticket text whether the issue is technical, financial, or shipping-related — and routes it to the right team
  • Prioritization: Based on issue urgency and customer sentiment, AI sets the ticket priority
  • Assignment to the best operator: AI knows which operator is better at solving which type of problem and assigns the ticket accordingly
  • Resolution time prediction: AI estimates how long solving the issue will take based on similar past tickets

Result: faster response time, higher customer satisfaction, and operators focus on tasks that truly need their expertise.

Measuring Success: How to Know AI Is Working

Every AI project in customer service should be measured with specific metrics. Here are the most important ones:

Quantitative Metrics

  • First Contact Resolution rate: What percentage of problems are solved on the first contact? Target: above 70 percent
  • Average response time: How long from customer contact to first response? With AI it should be under 30 seconds
  • Deflection Rate: What percentage of requests does the chatbot resolve without human intervention? Initial target: 40 to 60 percent
  • Cost per contact: How much does each customer interaction cost? AI should reduce this by at least 30 percent
  • Escalation Rate: What percentage of conversations transfer from chatbot to human? If above 50 percent, the chatbot needs improvement

Qualitative Metrics

  • CSAT (Customer Satisfaction Score): Customer satisfaction with chatbot interaction — ask after each conversation
  • NPS (Net Promoter Score): Would the customer recommend you to others?
  • Qualitative feedback: Read customer comments. If they say the chatbot was useless, you have a problem
Important note: Measure metrics before starting the AI project so you have a baseline. Without a baseline, you cannot say how much AI has improved things.

Step-by-Step Implementation

If you are convinced AI in customer service is useful for you, follow this path:

Phase 1: Preparation (Weeks 1-2)

  1. Write a list of the 20 most common customer questions
  2. Prepare an accurate and complete answer for each question
  3. Measure current customer service metrics (baseline)
  4. Talk to the customer service team — listen to their concerns

Phase 2: Pilot (Weeks 3-6)

  1. Choose an AI chatbot tool (like Intercom, Zendesk AI, Tidio, or a custom chatbot)
  2. Activate it only for the 20-30 most common questions
  3. Test on a portion of traffic (e.g., only the website, not phone)
  4. Review conversations daily and see where the chatbot makes mistakes

Phase 3: Improvement (Weeks 7-12)

  1. Based on real data, improve the chatbot
  2. Add more questions
  3. Activate sentiment analysis and smart routing
  4. Re-measure metrics and compare with baseline

Phase 4: Expansion (Month 4 onwards)

  1. Add to other channels (WhatsApp, Telegram, phone)
  2. Add more advanced capabilities (product recommendations, feedback collection)
  3. Use collected data to improve products and services

Common Implementation Mistakes

Avoid these mistakes:

  1. Launching without sufficient content: If the chatbot does not have answers to questions, it frustrates customers. Prepare content first, then launch
  2. Hiding the human option: If the customer cannot reach an operator, they get angry and leave. The connect to operator button must be clear and always accessible
  3. Ignoring feedback: Review unsuccessful conversations every week. If customers get stuck at a specific point, fix it
  4. Expecting instant results: The first month, the chatbot is weak and that is normal. By month three it gets better. Be patient
  5. Forgetting the team: Tell the customer service team that AI is going to help them, not replace them. If the team is scared, they will resist the project
Real results: Companies that properly implement the hybrid approach (AI + human) typically experience 30 to 50 percent reduction in customer service costs and 15 to 25 percent increase in customer satisfaction.

The Future of Customer Service with AI

Several important trends in the near future:

  • Voice chatbots: Instead of typing, the customer speaks and voice AI responds — more natural and faster
  • Predicting problems before contact: AI detects a problem before the customer calls and proactively reaches out
  • Deep personalization: AI knows the complete customer history and personalizes the conversation
  • Video analysis: AI can detect customer emotions from facial expressions in video calls

Series Summary

In 5 episodes of this series, we covered a complete path:

  1. We understood what AI is and why it matters now
  2. We learned where to identify AI opportunities in business
  3. We explored ready-made tools and learned how to use them
  4. AI in marketing and sales — reality separated from hype
  5. AI in customer service — chatbots, sentiment analysis, and hybrid systems

The main point of all these episodes is one thing: AI is a tool, not magic. Like any other tool, when used correctly it is extraordinary, and when used poorly, it causes harm. Your most important job as a manager is to understand, test, and proceed gradually.

A manager who understands AI does not need to become a programmer. They just need to know where to use this tool, where not to, and how to measure the results. You are now that manager.