WEBVTT 00:00:00.600 --> 00:00:06.765 Welcome to lesson 4.1, AI in customer service: build and deploy your own chatbot. 00:00:07.115 --> 00:00:10.529 Many businesses now use them to help customers day and night. 00:00:10.879 --> 00:00:16.767 In this lesson you will learn what a customer-service chatbot does and how a modern AI chatbot works. 00:00:17.117 --> 00:00:20.445 You will see how to build and test one yourself, with no coding. 00:00:20.795 --> 00:00:24.635 And you will learn your legal duty to tell customers they are talking to AI. 00:00:26.135 --> 00:00:29.100 By the end of this lesson you will be able to do four things. 00:00:29.450 --> 00:00:36.511 First, explain the difference between an older rule-based chatbot and a modern AI chatbot built on a large language model. 00:00:36.861 --> 00:00:41.235 Second, write a system prompt that sets the bot's role, tone, scope and fallback. 00:00:41.585 --> 00:00:47.302 Third, build a knowledge base and test your chatbot for wrong answers and a smooth hand-off to a person. 00:00:47.652 --> 00:00:53.113 Fourth, apply the EU AI Act rule that says users must know they are talking to AI. 00:00:54.613 --> 00:00:56.235 Here are some key terms. 00:00:56.585 --> 00:01:02.771 A chatbot is an AI assistant that answers questions in everyday language, often for customer service. 00:01:03.121 --> 00:01:07.879 An LLM, or large language model, is the AI behind modern chatbots. 00:01:08.229 --> 00:01:10.319 It predicts natural-sounding text. 00:01:10.669 --> 00:01:16.003 A system prompt is the set of instructions that sets a chatbot's role, tone, scope and limits. 00:01:16.353 --> 00:01:21.067 The knowledge base is the documents it answers from, such as FAQs and policies. 00:01:21.417 --> 00:01:26.708 Escalation means handing a conversation to a human when the bot cannot or should not answer. 00:01:27.058 --> 00:01:33.927 Finally, the transparency duty is the EU AI Act rule to tell people clearly when they are dealing with AI. 00:01:35.427 --> 00:01:37.411 Customer service never sleeps. 00:01:37.761 --> 00:01:41.900 A chatbot answers routine questions in plain language, at any hour. 00:01:42.250 --> 00:01:45.023 That means staff can focus on the hard cases. 00:01:45.373 --> 00:01:47.080 And this is growing fast. 00:01:47.430 --> 00:01:57.755 According to Eurostat figures published in 2025, 55 per cent of EU businesses used AI, cloud or analytics in 2023. 00:01:58.105 --> 00:02:01.241 So chatbot skills are in demand, and still quite rare. 00:02:01.591 --> 00:02:03.149 They could be your skills. 00:02:04.649 --> 00:02:07.529 A customer-service chatbot does three main jobs. 00:02:07.879 --> 00:02:10.012 First, it answers common questions. 00:02:10.362 --> 00:02:14.458 Think of opening hours, order status, returns, or how do I do something. 00:02:14.808 --> 00:02:16.877 The bot answers these in seconds. 00:02:17.227 --> 00:02:19.190 Second, it works around the clock. 00:02:19.540 --> 00:02:24.809 A customer can get help at midnight or on a bank holiday, with nobody waiting by the keyboard. 00:02:25.159 --> 00:02:27.677 Third, it frees people for the hard cases. 00:02:28.027 --> 00:02:34.149 Staff stop repeating the same answers and focus on complex, sensitive or valuable conversations. 00:02:34.499 --> 00:02:36.483 Remember, the bot supports staff. 00:02:36.833 --> 00:02:38.433 It does not replace them. 00:02:38.783 --> 00:02:42.837 Let the bot handle the routine, so people can handle the tricky questions. 00:02:44.337 --> 00:02:46.470 There are two main kinds of chatbot. 00:02:46.820 --> 00:02:48.783 The older kind is rule-based. 00:02:49.133 --> 00:02:51.245 It follows fixed buttons and scripts. 00:02:51.595 --> 00:02:53.515 It is predictable, but limited. 00:02:53.865 --> 00:02:56.723 If your question is not on its menu, it gets stuck. 00:02:57.073 --> 00:03:00.849 The modern kind is based on AI, using a large language model. 00:03:01.199 --> 00:03:06.127 It understands questions written in natural language and answers from the knowledge you give it. 00:03:06.477 --> 00:03:12.600 It is far more flexible, but it can hallucinate, which means it can give a confident answer that is simply wrong. 00:03:12.950 --> 00:03:15.553 The best practice is a knowledge-grounded bot. 00:03:15.903 --> 00:03:18.868 Grounded means it answers only from your own documents. 00:03:19.218 --> 00:03:21.821 That keeps its replies accurate and on-brand. 00:03:23.321 --> 00:03:26.649 A modern AI chatbot works in four simple stages. 00:03:26.999 --> 00:03:28.535 One, you feed it knowledge. 00:03:28.885 --> 00:03:32.213 You upload FAQs, policies and product information. 00:03:32.563 --> 00:03:34.461 This becomes its source of truth. 00:03:34.811 --> 00:03:36.582 Two, you set its behaviour. 00:03:36.932 --> 00:03:41.263 You give it a name and a system prompt that defines its tone, scope and limits. 00:03:41.613 --> 00:03:43.447 Three, it answers customers. 00:03:43.797 --> 00:03:48.747 It matches each question to your content and replies in seconds, at any time of day. 00:03:49.097 --> 00:03:54.387 Four, you review and improve: check where it fails, improve the prompt and fill gaps in its knowledge. 00:03:54.737 --> 00:03:58.684 The quality of the bot comes mostly from the knowledge and the prompt you give it. 00:03:59.034 --> 00:04:02.234 You will do all four stages in exercise 4.1. 00:04:03.734 --> 00:04:06.401 The system prompt is the bot's job description. 00:04:06.751 --> 00:04:09.567 Write it like a briefing for a new employee on day one. 00:04:09.917 --> 00:04:11.453 Take a small bakery. 00:04:11.803 --> 00:04:17.989 First, role and tone: you are the friendly support assistant for a small bakery, be warm, short and polite. 00:04:18.339 --> 00:04:25.827 Second, scope and limits: only answer questions about our products, opening hours and orders, and for anything else, offer a human. 00:04:26.177 --> 00:04:28.929 Third, the fallback, for when the bot does not know. 00:04:29.279 --> 00:04:33.034 If unsure, say so, suggest contacting the team, and never guess. 00:04:33.384 --> 00:04:35.261 Fourth, AI disclosure. 00:04:35.611 --> 00:04:39.238 Start each chat with: you're chatting with our AI assistant. 00:04:39.588 --> 00:04:42.361 A clear prompt separates a good bot from a bad one. 00:04:43.861 --> 00:04:47.083 Your chatbot is only as good as the information you give it. 00:04:47.433 --> 00:04:49.459 So what goes into the knowledge base? 00:04:49.809 --> 00:04:51.239 Cover four areas. 00:04:51.589 --> 00:04:53.317 First, frequent questions. 00:04:53.667 --> 00:04:58.595 These are the real questions customers ask most often, with clear and correct answers. 00:04:58.945 --> 00:05:03.041 Second, policies, such as returns, refunds, delivery and privacy. 00:05:03.391 --> 00:05:06.143 These are the rules customers need to understand. 00:05:06.493 --> 00:05:12.786 Third, products and services: what you offer, the key details, the prices, and what makes each option different. 00:05:13.136 --> 00:05:18.832 Fourth, contact and escalation: how and when to reach a human, so the bot can hand over smoothly. 00:05:19.182 --> 00:05:22.297 Tip: write your FAQ the way customers really ask. 00:05:23.797 --> 00:05:27.679 You can build a working chatbot in under an hour, with no coding at all. 00:05:28.029 --> 00:05:29.544 One, create the bot. 00:05:29.894 --> 00:05:33.969 Sign up to a free no-code tool such as Chatbase and start a new project. 00:05:34.319 --> 00:05:36.004 Two, upload your knowledge. 00:05:36.354 --> 00:05:41.965 Add an FAQ document, your policies or a website link, so the bot answers from real content. 00:05:42.315 --> 00:05:44.128 Three, name it and prompt it. 00:05:44.478 --> 00:05:49.534 Give it a friendly name and a system prompt: its role, tone, and what it should and should not do. 00:05:49.884 --> 00:05:51.676 Four, test, then deploy. 00:05:52.026 --> 00:05:58.213 Ask real customer questions, fix weak spots, then add it to a website with the code the tool gives you. 00:05:58.563 --> 00:06:01.891 This is exactly what you will do in exercise 4.1. 00:06:03.391 --> 00:06:06.591 Test your chatbot like any product, before customers do. 00:06:06.941 --> 00:06:09.778 First, try everyday questions and edge cases. 00:06:10.128 --> 00:06:12.197 Run the ten most common questions. 00:06:12.547 --> 00:06:15.299 Then try awkward, vague or off-topic ones. 00:06:15.649 --> 00:06:17.548 Second, hunt for wrong answers. 00:06:17.898 --> 00:06:22.442 A confident but incorrect reply does the most damage, so fix these first. 00:06:22.792 --> 00:06:24.435 Third, check the hand-off. 00:06:24.785 --> 00:06:29.094 Make sure the bot offers a human when it should, and never invents information. 00:06:29.444 --> 00:06:30.703 Here is a tip. 00:06:31.053 --> 00:06:34.018 Keep a simple test log of the questions it failed. 00:06:34.368 --> 00:06:36.608 That list becomes your improvement plan. 00:06:38.108 --> 00:06:41.351 The same technology can delight customers or annoy them. 00:06:41.701 --> 00:06:43.813 What makes the difference is behaviour. 00:06:44.163 --> 00:06:47.811 A good chatbot answers accurately from real company information. 00:06:48.161 --> 00:06:50.614 It clearly says it is an AI assistant. 00:06:50.964 --> 00:06:52.799 It admits when it does not know. 00:06:53.149 --> 00:06:55.602 And it offers a human for complex problems. 00:06:55.952 --> 00:06:58.981 A bad chatbot invents confident but wrong answers. 00:06:59.331 --> 00:07:01.038 It pretends to be a human. 00:07:01.388 --> 00:07:03.607 It traps customers in endless loops. 00:07:03.957 --> 00:07:06.261 And it never lets you reach a real person. 00:07:06.611 --> 00:07:08.296 Aim to build the good kind. 00:07:08.646 --> 00:07:10.993 It builds trust instead of frustration. 00:07:12.493 --> 00:07:16.717 Now for something important: your legal duty to disclose the AI. 00:07:17.067 --> 00:07:25.429 Under Article 50 of the EU AI Act, people must be told clearly when they are talking to an AI system, unless it is already obvious. 00:07:25.779 --> 00:07:29.107 This rule applies from 2 August 2026. 00:07:29.457 --> 00:07:31.633 This is the law, not just good manners. 00:07:31.983 --> 00:07:35.973 Add a visible line such as: you're chatting with our AI assistant. 00:07:36.323 --> 00:07:38.648 Never let your bot pretend to be a human. 00:07:38.998 --> 00:07:43.329 And build the habit of disclosing AI into everything you make, starting now. 00:07:44.829 --> 00:07:46.663 Time to check your understanding. 00:07:47.013 --> 00:07:49.936 Pause the video and try to answer these four questions. 00:07:50.286 --> 00:07:54.638 First, what is the difference between a rule-based chatbot and an AI chatbot? 00:07:54.988 --> 00:07:59.383 Second, why does grounding a chatbot in your own documents reduce wrong answers? 00:07:59.733 --> 00:08:03.295 Third, what three things should a good system prompt always define? 00:08:03.645 --> 00:08:08.467 Fourth, under the EU AI Act, what must you tell customers who chat with your bot? 00:08:08.817 --> 00:08:11.313 Take your time, then press play when you are ready. 00:08:12.813 --> 00:08:16.162 If you want to explore further, there are two useful resources. 00:08:16.512 --> 00:08:20.523 Chatbase is the no-code chatbot builder you will use in the exercise. 00:08:20.873 --> 00:08:26.035 And there is Article 50 of the EU AI Act, which sets out the transparency rules. 00:08:26.385 --> 00:08:28.604 You will find the links on the lesson page. 00:08:30.104 --> 00:08:31.832 Let's sum up the key points. 00:08:32.182 --> 00:08:36.619 Modern AI chatbots answer natural questions using the knowledge you give them. 00:08:36.969 --> 00:08:40.980 Ground your bot in your own documents to keep answers accurate and on-brand. 00:08:41.330 --> 00:08:45.959 A clear system prompt, with role, scope and fallback, is what makes a bot reliable. 00:08:46.309 --> 00:08:49.936 Always test for wrong answers and for a clean hand-off to a human. 00:08:50.286 --> 00:08:56.302 And by law, under Article 50 of the EU AI Act, you must tell users they are talking to AI. 00:08:57.802 --> 00:09:00.341 Now it is your turn to put this into practice. 00:09:00.691 --> 00:09:04.979 On the platform, start with the activity: build your chatbot's system prompt. 00:09:05.329 --> 00:09:10.065 Then do exercise 4.1, where you build and test a live AI chatbot. 00:09:10.415 --> 00:09:12.548 It takes about one and a half hours. 00:09:12.898 --> 00:09:14.903 Finally, take the lesson quiz. 00:09:15.253 --> 00:09:18.709 It has five questions, and you need 70 per cent to pass. 00:09:19.059 --> 00:09:21.641 You can retake it as many times as you like. 00:09:21.991 --> 00:09:23.185 Good luck!