[{"data":1,"prerenderedAt":199},["ShallowReactive",2],{"blog-ai-chatbots-for-customer-support":3,"blog-related-\u002Fblog\u002Fai-chatbots-for-customer-support":169},{"id":4,"title":5,"author":6,"authorRole":7,"body":8,"cover":151,"coverAlt":152,"date":153,"description":154,"draft":155,"extension":156,"meta":157,"navigation":158,"path":159,"readingTime":160,"seo":161,"stem":162,"tags":163,"updated":167,"__hash__":168},"blog\u002Fblog\u002Fai-chatbots-for-customer-support.md","AI chatbots for customer support: what actually works","William Ali","Founder & AI Specialist",{"type":9,"value":10,"toc":142},"minimark",[11,15,20,28,45,49,52,56,59,70,73,77,80,132,136,139],[12,13,14],"p",{},"Most support bots fail for one of two reasons: they answer confidently from nothing, or they answer nothing at all. The teams that get real value treat the assistant as a system with three parts (knowledge, actions and hand-off) and measure it honestly.",[16,17,19],"h2",{"id":18},"_1-ground-it-in-what-you-actually-know","1. Ground it in what you actually know",[12,21,22,23,27],{},"A large language model on its own is a very persuasive guesser. Retrieval-augmented generation (RAG) fixes that by fetching the relevant help-centre article, policy or order record first and asking the model to answer ",[24,25,26],"em",{},"from"," it. Two rules we follow:",[29,30,31,39],"ul",{},[32,33,34,38],"li",{},[35,36,37],"strong",{},"Cite or decline."," If nothing relevant is retrieved above a confidence threshold, the assistant says so and escalates. Silence beats hallucination.",[32,40,41,44],{},[35,42,43],{},"Keep the index fresh."," Nightly re-ingestion of your knowledge base is not optional; stale answers erode trust faster than no answers.",[16,46,48],{"id":47},"_2-give-it-hands-carefully","2. Give it hands, carefully",[12,50,51],{},"The difference between \"your order usually ships in two days\" and \"your order shipped yesterday, here's the tracking link\" is a tool call. Connect the assistant to read-only APIs first (orders, subscriptions, tickets), then add write actions behind confirmation steps: start a return, reschedule a delivery, open a ticket.",[16,53,55],{"id":54},"_3-design-the-hand-off-as-a-feature","3. Design the hand-off as a feature",[12,57,58],{},"Every conversation the bot cannot resolve should reach a human with:",[29,60,61,64,67],{},[32,62,63],{},"the full transcript,",[32,65,66],{},"the retrieved sources,",[32,68,69],{},"a suggested reply the agent can edit and send.",[12,71,72],{},"Agent-assist alone often saves as much time as full automation, with none of the risk.",[16,74,76],{"id":75},"_4-measure-resolution-not-deflection","4. Measure resolution, not deflection",[12,78,79],{},"\"Deflection\" counts customers who gave up. Track instead:",[81,82,83,96],"table",{},[84,85,86],"thead",{},[87,88,89,93],"tr",{},[90,91,92],"th",{},"Metric",[90,94,95],{},"Why it matters",[97,98,99,108,116,124],"tbody",{},[87,100,101,105],{},[102,103,104],"td",{},"Resolution rate",[102,106,107],{},"Conversations that ended without a human and without a re-contact within 7 days",[87,109,110,113],{},[102,111,112],{},"Hand-off quality",[102,114,115],{},"Time saved per escalated ticket",[87,117,118,121],{},[102,119,120],{},"CSAT on bot-handled conversations",[102,122,123],{},"Compared with human-handled, not with the old bot",[87,125,126,129],{},[102,127,128],{},"Cost per conversation",[102,130,131],{},"Tokens + infrastructure ÷ resolved conversations",[16,133,135],{"id":134},"_5-ship-an-evaluation-set","5. Ship an evaluation set",[12,137,138],{},"Collect 200–500 real questions with approved answers. Score every release against it. When a prompt tweak or model upgrade lands, you'll know within minutes whether quality moved.",[12,140,141],{},"Done this way, we routinely see assistants resolve well over half of first-line contacts within a quarter and, more importantly, support teams that trust the system enough to keep expanding it.",{"title":143,"searchDepth":144,"depth":144,"links":145},"",2,[146,147,148,149,150],{"id":18,"depth":144,"text":19},{"id":47,"depth":144,"text":48},{"id":54,"depth":144,"text":55},{"id":75,"depth":144,"text":76},{"id":134,"depth":144,"text":135},"\u002Fcovers\u002Fblog\u002Fai-chatbots-for-customer-support.svg","Line-art illustration of a support conversation in chat bubbles ending in a check mark, with a hand-off line to a human agent.","2025-11-12","Beyond the demo: how to design a support assistant that resolves issues, hands over gracefully and earns its keep.",false,"md",{},true,"\u002Fblog\u002Fai-chatbots-for-customer-support",7,{"title":5,"description":154},"blog\u002Fai-chatbots-for-customer-support",[164,165,166],"Conversational AI","LLMs","Product",null,"7swKvaHJAkL_ELvfnX3dxSQLcYMqbnkitNr2agHXt9M",[170,180,190],{"path":171,"title":172,"description":173,"date":174,"tags":175,"cover":178,"coverAlt":179,"readingTime":160},"\u002Fblog\u002Fmeasuring-ai-roi-honestly","Measuring AI ROI honestly: a method that survives the finance review","How to measure the return on an AI project without fooling yourself: baselines, full costs, holdout attribution, leading indicators and inflated-ROI red flags.","2026-08-11",[176,177],"Strategy","Analytics","\u002Fcovers\u002Fblog\u002Fmeasuring-ai-roi-honestly.svg","Line-art illustration of a ruler measuring a stack of coins beside a rising bar.",{"path":181,"title":182,"description":183,"date":184,"tags":185,"cover":188,"coverAlt":189,"readingTime":160},"\u002Fblog\u002Fmultimodal-ai-business-workflows","Multimodal AI in real business workflows: where it earns its place","Vision-language and audio models now read forms, photos, screenshots and calls directly. Where multimodal AI beats OCR-then-LLM, with examples, limits and cost.","2026-06-30",[186,165,187],"Computer Vision","Automation","\u002Fcovers\u002Fblog\u002Fmultimodal-ai-business-workflows.svg","Line-art illustration of an eye, an ear and a speech bubble converging into a single processing node.",{"path":191,"title":192,"description":193,"date":194,"tags":195,"cover":197,"coverAlt":198,"readingTime":160},"\u002Fblog\u002Fmlops-basics-first-model","MLOps basics for teams shipping their first model","The minimum MLOps a team needs to put a first model into production safely: versioning, reproducible training, a registry, serving, monitoring and rollback.","2026-05-19",[196,166],"MLOps","\u002Fcovers\u002Fblog\u002Fmlops-basics-first-model.svg","Line-art illustration of a looping pipeline with gauges, from data to model to monitoring and back.",1792973685026]