A modern AI chatbot is a different species from the clunky "Can I help you?" widget you remember from 2018. Chatbots built on GPT-class language models understand context, handle multi-turn conversations, and take action — they don't just answer questions. For a small or mid-size company, a well-built chatbot typically pays for itself in under three months. A badly built one is an expensive facade. The gap between the two is wider than most people realise. What a Modern Chatbot Actually Does The difference from a 2018 chatbot is the difference between a Nokia 3310 and an iPhone. The old bot matched keywords to canned scripts — you knew it by the fact that "talk to a human" was the only useful button. A modern AI chatbot runs on GPT-4- or Claude-class models, understands language natively, and reads your company's documentation through vector search. It answers based on context and brand voice, remembers the conversation, and can call APIs when it needs to — book a meeting on a rep's calendar, create a CRM lead, or check an order status. Three Use Cases That Pay Back the Fastest First: lead qualification. A chatbot on a landing page asks a handful of precise questions (company size, industry, budget, timeline) and scores the lead before sales ever touches it. Typically 60–80% of inbound contacts arrive pre-qualified, which doubles the conversion rate and cuts response time to seconds. Second: 24/7 support for common questions. 60–75% of tickets are repetitive — "how do I return this", "when does my order arrive", "how do I reset my password". An AI chatbot handles those instantly, around the clock. Your support team only sees the complex cases where human judgement actually matters. Third: booking and sales assist. A chatbot that can book a meeting straight into a rep's calendar beats a contact form every time. Less friction, and the booking happens at the moment buyer intent is highest — not two days later when someone gets around to calling. The Mistakes That Kill Chatbot Projects The biggest mistake is dropping in a generic out-of-the-box widget that knows nothing about your company. The result: a customer asks "what's your return policy" and the bot says "I can't find an answer." The fix is a real knowledge base (docs, FAQs, product guides) served through vector search. The second classic mistake is skipping the human handoff — every chatbot needs a clean escalation path once it hits the edge of its knowledge. The third is measuring the wrong things: forget message volume, track resolution rate, qualified leads, and CSAT. What a Chatbot Costs in 2026 Off-the-shelf SaaS chatbots (Intercom Fin, Zendesk AI) typically run €200–800/month per seat, but they cap out on brand voice, knowledge base, and integrations. A custom AI chatbot with its own knowledge base, CRM integrations, and analytics runs €6,000–18,000 upfront and €150–400/month to operate (API calls, hosting, upkeep). For a mid-size company, over a 12-month horizon the custom build is usually cheaper and delivers materially better results. How to Measure Success Don't measure conversations — measure outcomes. The three metrics that matter: (1) resolution rate — how many conversations end without a human escalation; (2) qualified leads per month — how many bot conversations turn into a meeting or a proposal; (3) CSAT — ask for a rating at the end of the conversation. A well-built chatbot hits 70–85% resolution and 4.2+/5 CSAT by month three. A chatbot is no longer a gimmick — it's a sales channel and a support resource that never sleeps, never calls in sick, and never forgets the brand voice. But like any tool, the payoff comes from how it's built. Pick a partner who understands your industry and builds the knowledge base with care — not one who sells you a widget and disappears.