## [AI development and implementation](/services/development/ai-development-and-implementation) Once a business has worked out what AI should be used for, the solution has to be built and put into production. That is the part we call development and implementation. Here we explain what it involves, why many solutions stop working after the first demo, and how we work. Have a task you would like solved? [**Write to us, and we'll take a 20-minute conversation →**](/contact) ## What does it mean to implement AI? To implement means to put a solution into use so it becomes a fixed part of everyday work. For an AI solution, that means it runs on your own data, is connected to the systems you already use, and keeps working without anyone having to watch it every day. A language model such as ChatGPT or Claude is only a small part of such a solution. The rest is ordinary software: it fetches the right information, sends it to the model, checks the answer, stores the result in the right place and raises the alarm if something goes wrong. Most of the work lies in that part. ## Why do so many solutions stop working? It is fairly easy to make a prototype that looks good in a meeting. It is harder to get the same solution to run for a year. When a server restarts at night, a password expires or another system changes, the solution can stall, and often nobody notices until a customer or a colleague complains. Those failures rarely have anything to do with AI. They are about operations: what happens after a restart, who gets notified when something stops, and whether you can see that the solution is running right now. We build the answers to those questions in from the start, and we restart the solution ourselves and watch it come back up before we call it finished. ## What we build The most common is a solution that can answer questions from your own documents, such as handbooks, contracts, product sheets or previous quotes. The technique is called RAG. The solution first finds the documents that match the question and gives them to the language model, so the answer is based on your knowledge and can show where it comes from. We also build automated workflows, where a task done by hand today runs by itself. That might be sorting incoming emails, drafting replies, extracting information from invoices or producing a daily summary. A human approves anything that cannot be undone before it is passed on. We connect systems that do not talk to each other today, so information does not have to be entered in two places. And we make your website and product data easy for AI assistants to read. More and more customers ask an AI assistant before they buy. The assistant reads your website, and if the texts are missing, outdated or in the wrong language, it gives a poor answer about you. ## How a project runs We start with a small prototype that solves a limited part of the task on a sample of your real data. You can see it working within a few weeks, and it shows whether the idea holds before more money is spent. Then we finish building the solution, test it on the full data set and connect it to your systems. We agree in advance how you can measure whether it does what it should, for example time saved per case or fewer errors. Finally, we put the solution into production with monitoring, so someone is notified if it stops. You get a description of how it is built, and you own the code and your data. ## What you need to contribute We need access to the systems and data the solution will use. And you need to appoint a person who knows the workflow and can make decisions along the way. That is the precondition that most often makes a schedule slip, so we write it into the proposal. A solution also has to be maintained once it is in production. Models are updated, and the systems around it change. You can do that yourselves, or you can let us do it. ## How to get started - [AI Starter Kit](/packages/ai-starter-kit), €10,000. A workshop and a review of your business that ends with a prioritised list of concrete AI projects. For those who do not yet know what should be built. - [AI Implementation package](/packages/ai-implementation-package), from €50,000. We design, build and put the solution into production on your own data. The price depends on the scope. - [Ongoing AI support](/packages/ai-support-retainer), from €2,000 per month. A fixed number of hours each month for maintenance, adjustments and questions once the solution is in production. All prices exclude VAT. If the task is smaller than a full implementation, write to us and we'll give you a price. ## Read more (in Danish) - [Works in the demo, dead after a restart: what automation is really about](/da/node/233) - [Systemd and a JSONL file: how I automated folkets.dk](/da/node/230) - [What one person can build in a week](/da/node/225) - [Puncturing the AI agent hype](/da/node/78) ## Contact us Tell us briefly what task you would like solved. We'll take a conversation of about 20 minutes and tell you honestly whether it can be solved with AI and what it will take. [**Write to us**](/contact)