[{"data":1,"prerenderedAt":32},["ShallowReactive",2],{"article-customer-service-automation":3},{"id":4,"slug":5,"title":6,"description":7,"authorId":8,"categoryIds":9,"status":11,"publishedAt":12,"updatedAt":12,"heroImage":13,"heroImageAlt":14,"content":15,"author":16,"categories":26,"readingTimeMinutes":31},"article-1794228486531","customer-service-automation","Customer service automation: teach the AI your business first","Customer service automation from a real build: what the AI learns first, where a person stays in, how long it took and the two numbers we measure.","author-andy-slack",[10],"category-ai-development","published","2026-11-09","\u002Fassets\u002Farticles\u002Fcustomer-service-automation\u002Fhero.png","Past customer messages, client data and process documents flow into an AI brain that learns the business's processes, products and communication style, while a person confirms what it knows and fills the gaps","\u003Cp>Customer service automation means letting software take on some of the work your team does by hand when customers get in touch: sorting messages, sending each one to the right person, answering the common questions, or drafting a reply for someone to send. A list of features won't tell you where to start. This article is about the step that comes before any of them: what the AI has to learn about your business before it answers anyone. It draws on one ecommerce build of ours and on how we run our own inbox.\u003C\u002Fp>\u003Cp>Two neighbouring questions have their own pages. How to pick a first process in general is in \u003Ca href=\"\u002Farticles\u002Fbusiness-process-automation-what-to-automate-first\">what to automate first, and when to build\u003C\u002Fa>. What an AI step may read and write, and where a person approves, is in \u003Ca href=\"\u002Farticles\u002Fai-integration-services\">AI integration services\u003C\u002Fa>.\u003C\u002Fp>\u003Ch2>Three routes, one question\u003C\u002Fh2>\u003Cp>You have three routes to choose from. Use the AI features in the helpdesk you already have, if it offers them. Put a chatbot on your site. Or have AI built around your own systems and records.\u003C\u002Fp>\u003Cp>Ask each route the same question: what will it learn from, and will it learn how your business works and how your team writes? An answer is only as good as what the AI knows about your business. If the choice in front of you is a connector, shared tasks or software written for the job, we set that out in \u003Ca href=\"\u002Farticles\u002Fworkflow-automation-software\">workflow automation software\u003C\u002Fa>.\u003C\u002Fp>\u003Ch2>What we automated first: the AI’s knowledge\u003C\u002Fh2>\u003Cp>For an ecommerce client, the first step wasn’t a chatbot or an auto-reply. We had AI read all of the business’s past customer communication, to fill the business’s own “AI brain”: its memory of how it deals with customers. We also fed it all of the client’s data, process documents included.\u003C\u002Fp>\u003Cp>From that, it learnt the business processes, the product information and the communication style: how this team writes to its customers. Starting there means every later answer draws on the business’s own record, not on general knowledge.\u003C\u002Fp>\u003Ch2>Where a person stays in\u003C\u002Fh2>\u003Cp>In that build, the people’s part was teaching. They confirmed what the AI had learnt and answered its questions about the business, so the training stuck and the gaps in what it knew were filled.\u003C\u002Fp>\u003Cp>Plan time for that. A team that answers the AI’s questions is doing the work that makes its later answers right. Where an AI step should stop and hand a case to a person, and what that person should see, are boundaries we set out in \u003Ca href=\"\u002Farticles\u002Fai-integration-services\">AI integration services\u003C\u002Fa>, so we don’t repeat them here.\u003C\u002Fp>\u003Ch2>How long it took\u003C\u002Fh2>\u003Cp>It took about three months to go from no AI at all to AI handling 80% of the client’s cases (our project record, confirmed 9 November 2026). Plan for a period of teaching and checking like that, not a launch day. Your own timeline depends on your business, and we don’t promise one.\u003C\u002Fp>\u003Cfigure>\u003Cimg src=\"\u002Fassets\u002Farticles\u002Fcustomer-service-automation\u002Ffig-1.png\" alt=\"The build in order: AI reads all past customer communication, is fed the client's data and process documents, and learns processes, products and communication style. People confirm what it has learnt and answer its questions. After about three months, AI handles 80% of cases\">\u003Cfigcaption>The AI learnt the business first, people filled its gaps, and about three months on it handled 80% of cases.\u003C\u002Ffigcaption>\u003C\u002Ffigure>\u003Ch2>The two numbers we measure\u003C\u002Fh2>\u003Cp>We judge a customer service automation by two numbers, taken before and after: average resolution time, and cNPS, which is how likely customers are to recommend the business, out of 10. For the ecommerce client, average resolution time went from 29.5 hours to 10.9 hours, and cNPS from 7.1 to 7.9. Both are listed with our \u003Ca href=\"\u002F\">client results\u003C\u002Fa>.\u003C\u002Fp>\u003Cp>Take your own before readings before anything is switched on, or you won’t be able to tell what changed. And read the two together: a faster answer that leaves customers less likely to recommend you is not the result you were after.\u003C\u002Fp>\u003Ch2>The tools behind it\u003C\u002Fh2>\u003Cp>For that client we automated operations with no-code workflows. The tools in the client’s work include n8n, a workflow tool, and OpenAI. Our records don’t say which tools the customer service part used, so we don’t name any others.\u003C\u002Fp>\u003Ch2>If your support inbox is small\u003C\u002Fh2>\u003Cp>Start with the sorting, and leave the replies with a person. In our own inbox, an agent labels new mail every 15 minutes and only escalates what needs a person, and no reply goes out without a person’s approval.\u003C\u002Fp>\u003Cp>In a small support inbox, sort and route first, and measure how many hours a week it saves. Let AI draft answers only once you’ve checked it against your own past messages: the same record that taught the AI in the build above is the test its drafts have to pass.\u003C\u002Fp>\u003Cfigure>\u003Cimg src=\"\u002Fassets\u002Farticles\u002Fcustomer-service-automation\u002Ffig-2.png\" alt=\"A small support inbox in three steps: sort each new message, route what needs a person to them, and a person replies, with no reply going out without approval. Below, AI drafting answers comes later, once it has been checked against your own past messages\">\u003Cfigcaption>Sort and route first, keep the replies with a person, and let AI draft only once it passes your own past messages.\u003C\u002Ffigcaption>\u003C\u002Ffigure>\u003Cp>Past customer messages carry customers’ names and details. Before an AI reads them, check your data protection duties for that use.\u003C\u002Fp>\u003Ch2>What you keep\u003C\u002Fh2>\u003Cp>Our usual handover is full training, so your own team can run and change the automation with no vendor lock-in, and a written summary of what was built and how it runs. We describe that practice in \u003Ca href=\"\u002Farticles\u002Fbusiness-process-automation-what-to-automate-first\">what to automate first\u003C\u002Fa>. For an AI that learns from your records, ask whoever builds it to say in that summary what the AI learns from, so your team can keep teaching it.\u003C\u002Fp>\u003Cp>\u003Cem>If you’re a founder with a product and want technical leadership to plan work like this, \u003Ca href=\"\u002F#lead\">book a founder fit call\u003C\u002Fa>. There is more in our \u003Ca href=\"\u002Farticles\">articles\u003C\u002Fa>.\u003C\u002Fem>\u003C\u002Fp>",{"id":8,"name":17,"slug":18,"title":19,"bio":20,"avatar":21,"featured":22,"social":23},"Andy Slack","andy-slack","Founder","Experienced Entrepreneur, Fractional CTO and Business Bootstrapper.","\u002Fassets\u002Fimages\u002Favatars\u002FAndy.jpeg",true,{"x":24,"linkedin":25,"website":24},"","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fandrewslack\u002F",[27],{"id":10,"name":28,"slug":29,"description":30},"AI Development","ai-development","Unlock the power of intelligent software. Our AI development category breaks down complex concepts into actionable insights. Discover how automation, predictive analytics, and custom AI solutions are transforming industries. Read case studies and guides on implementing enterprise AI effectively.",5,1795013445470]