What you want to know before the first call.

The AI workflow analysis costs €4,900 at a fixed price. The build per workflow costs €12,000 to €40,000 and becomes a binding fixed price after the analysis. Ongoing operation starts at €2,900 a month. A typical first year with two workflows therefore comes to between €60,000 and €100,000 net, plus VAT. Whether it pays off for you is set out in the analysis, with figures from your business. All stages and what they include can be found under Approach and prices.

That depends on how much work the workflow involves today, and that is exactly what the analysis measures: volumes, hours, errors, people involved. From this, each workflow is given a projected saving based on figures from your business, right next to the fixed price for the build. Workflows where the numbers do not add up go in a separate "does not pay off" column. The cost calculator gives you a rough estimate in advance. How the costs of an AI project break down is explained in the article What an AI project costs a mid-sized company.

No, and that is deliberate. Code, rules, prompts and documentation belong entirely to you after payment; there is no platform and no licence that ties you to us. We document every workflow so that your IT or another service provider can take it over. Ongoing operation can be cancelled monthly, and the workflows keep running afterwards. Rights of use are governed by our terms and conditions, and what we hand over is described under Systems.

You do: code, rules, prompts and documentation pass entirely into your ownership after payment, and the details are governed by our terms and conditions.

Monthly, to the end of the month, with no minimum term. The workflows keep running afterwards, because they belong to you. You only lose the monitoring, the model updates, the response time and the allowance for extensions. What ongoing operation includes in detail is set out under Ongoing operation.

Model costs depend on volume, languages and operating mode, and the analysis estimates them for each workflow. We do not pass them on with a mark-up: they run through your own account with the provider or are billed transparently by usage. With open models on your own servers, you pay for set-up and computing power instead of fees per request. What else determines the price of a build is shown in the table under What determines the price.

Yes. You decide for each workflow where the model runs: on your own servers with open models on your hardware, in an EU data centre or in the cloud under a data processing agreement. For personal data, contracts and anything confidential we recommend running on your own servers; then no data leaves your network. Every contract with us contains this sentence: No customer data leaves your systems without your written decision, and none of your data ever trains a third-party model. Which data flows where for which service is shown on the Data sovereignty page. The trade-offs between the three operating modes are explained in the article On your own servers, in Europe or in the cloud.

In your systems. We build into them, not out of them. For AI processing you choose for each workflow between open models on your own servers, EU data centres and cloud models under a data processing agreement. Our access runs through named accounts with two-factor authentication, logged and revocable by you at any time. The data flow for each service is on the Data sovereignty page.

The EU AI Act, Regulation (EU) 2024/1689, has been in force since August 2024 and classifies AI systems by risk. Most of our workflows, such as product texts, data clean-up or classifying enquiries, are minimal-risk systems. Where obligations arise, for example transparency under Article 50, documentation or human oversight, the workflow fulfils them through rules, a log and approval, and we document this. Since February 2025, Article 4 has also required AI literacy among employees who work with AI, and the handover of each workflow contributes to this. Details are in the article The EU AI Act for mid-sized companies and in the checklist for introducing AI. Neither is legal advice.

Language models make mistakes; we do not pretend otherwise. That is why every workflow has rules, a check and a log, and anything the rules do not clearly decide is referred to a person on your side. What we owe is a workflow that checks, logs and refers according to the agreed rules. Approval of the content of texts, data and customer contacts that are published or passed on stays with you. Exactly how this is divided is governed by sections 4 and 11 of our terms and conditions.

A data processing agreement under Article 28 GDPR before we access any data, and a list of sub-processors for each workflow with model provider, data centre and, where necessary, the legal basis for a third-country transfer. In addition, the data flow sketch for each workflow from the analysis as the basis for your record of processing activities, and deletion rules with proof of deletion at the end of the project. Access runs through named accounts with two-factor authentication, and credentials are never sent by email. The complete list is on the Data sovereignty page, and the points to check on your side are in the checklist for introducing AI.

We build workflows, not surveillance. What a workflow logs is defined in advance, and employee data is not a management target: the log records what happened to a record or text, not how fast someone worked. Because AI systems are regularly subject to co-determination, for example under section 87(1) no. 6 and section 90 of the German Works Constitution Act (BetrVG), we supply the documents for a works agreement: purpose, data flow, log content, access rights. Ideally the works council is already at the table during the analysis. How this works in practice is described in the article Introducing AI with the works council, which is no more legal advice than this answer.

That the second shift takes over the work nobody likes doing: copying, checking, transferring, reconciling data at night. The decisions stay with the people, and everything the rules do not clearly settle lands on their desk. We introduce it together with the people affected, not over their heads: in the analysis we talk to those who do the work today, and in the trial run they check the results. Their corrections become the rules of the workflow. How we approach this together with the works council is described in the article Introducing AI with the works council.

On the workflow itself. The handover in week 6 is part of the build: your team receives the documentation, learns to adjust the rules and knows what to do when the workflow puts something forward. A workflow running in the background needs no operating, just clear responsibilities, and a custom tool uses your company's own terminology. Training days are therefore not part of the package; anyone who wants more uses hours from the ongoing operation allowance. You can find the schedule under Approach and prices.

That is exactly why we build per workflow and not from a catalogue. In the analysis we spend three days watching where work arises in your company and write down your company's rules before anything is built. Where it does not work, the analysis says so in the "does not pay off" column, with reasons. Where standard software does not map your workflow, we build a custom tool. The criteria we use to decide are in the article Which workflows pay off with AI.

Often, yes, and that is not an obstacle but the natural first step. Putting data in order is a separate service of ours, at a fixed price: merging duplicates, filling mandatory fields, standardising spellings, with rules you help decide and a log for each record. We have cleaned up data with decades of sprawl, for example at a spare parts dealer with millions of parts in Salesforce. The potential check also asks about the state of your data. More under Putting data in order and in the article Before AI is let into the CRM.

Good, then your people know the difference from their own experience. A chat window answers questions, one after another, and someone has to copy the result, check it and enter it into the system. A workflow does work at volume, at night and on a recurring basis, directly in your CRM, ERP or PIM, with rules, a check and a log. The two are not mutually exclusive: Copilot stays for individual work at the desk, and the second shift takes over the bulk work. The difference is explained in the article Why ChatGPT has changed nothing in the business.

No. You need someone who grants us access and takes decisions, such as which rules apply and who approves. We take care of the rest, from setting up the environment to running it under ongoing operation. We document everything so that your IT or another service provider could take over at any time. What is needed from you in which week is in the schedule under Approach and prices.

The right one for each task and level of sensitivity: models from Anthropic, OpenAI and Google for language at volume, open models such as Llama and Mistral when data must not leave the company. The choice is set out in the analysis, with reasons, and you take the decision. We are not tied to any provider and do not earn anything from model costs. The operating modes are described on the Data sovereignty page.

Then we swap the model. Our workflows are built so that the model is an interchangeable component; rules, checks and the log stay as they are. Before a switch, the workflow runs with the new model against the same checks, and only when the results are right do we switch over. Such model updates are part of ongoing operation. How we monitor workflows in operation is described in the article AI operation and monitoring.

Every workflow has rules, checks and a log. Whatever the rules cannot clearly decide, the workflow puts before a person on your side instead of guessing. Before a workflow works on its own, it runs in parallel operation alongside the existing work, and its results are compared. Where a workflow writes back into your systems, we build in an undo function that restores the previous state. This is not small print but the difference between a prompt and a workflow; what all our workflows have in common is described under Services.

Where your workflow matches the standard, that is often right, and we connect the software. Where your workflow runs differently, you pay per head per month for a tool your company has to adapt to, and maintain the rest in Excel. In that case we build a tool that fits you, built with AI and therefore affordable. Which workflow belongs where is decided by the analysis, and the cost calculator shows beforehand how licence and custom build compare over five years. More under Tools that fit you and in the article Custom software at mid-sized companies.

The ones you already have: CRMs such as Salesforce, HubSpot and Pipedrive, ERPs such as NetSuite, SAP, Microsoft Dynamics and weclapp, HR systems such as Personio, SuccessFactors and Workday, PIMs such as Akeneo and Pimcore, plus Microsoft 365 and Google Workspace. We get to grips with any system that has an interface; that has been our trade since 1998. Where a system sets limits, for example with request limits, we build in caching and throttling so that you stay within your licence limits. The complete list and what we typically build for each family of systems is under Systems.

It lasts 30 minutes and is free and without obligation. You tell us about the workflow that costs you the most time, and we ask about volumes, systems and who works on it today. At the end we tell you honestly whether AI pays off there and what the next step would be, even if the answer is that we are not the right partner. You talk to the same specialists who will later build and run the workflow. You choose a free slot directly on the Book a call page and receive the invitation for your calendar.

No, but three things help: which workflow costs you the most time, roughly what volumes lie behind it (records, enquiries, texts, languages) and which systems are involved. We do not need documents, data exports or access for the first call. If you would like to sort out your key figures beforehand, take three minutes for the potential check. You can book the slot under Book a call.

Ten questions on size, systems, volumes and data quality, answered in under three minutes. The result appears immediately on the page: three to five workflows, each rated very likely to pay off, worth an analysis or unlikely to pay off, plus the suitable operating mode for your data. You only get a summary by email if you request it, and we do not store your answers on the server. If we are not the right partner, the check says so too. Go to the potential check.

Six sections of checkpoints that should be settled before the first workflow: preparation, data protection, works council and employees, EU AI Act, technology and operation, and the contract with the service provider. Each point has a short explanation and, where appropriate, a link to the detailed article. The checklist can be read in full on the page, without a form, and can be printed as a PDF. It is written so that you can pass it on to your data protection officer or works council, but it is not legal advice. Go to the checklist for introducing AI.

For a single workflow, often yes, and we say so. What we bring is experience from dozens of workflows with checks, logs and operation: where models stumble, which rules are missing, what a system can withstand under load. On top of that, we have worked on projects since 1998 in precisely the systems the AI has to work in. We hand over with code, rules and documentation so that your IT can carry on, and a division of labour in which your IT takes over operation is expressly possible. How we work is described under About us.

Less than most people expect. The analysis takes three days with the people who do the work today, in conversations at their workplace, not in workshops. The build needs one contact for rules and approvals, not a project team. In addition, you grant access in week 2 and check the sample during the trial run in week 3. You can find the week-by-week schedule under Approach and prices.

Then you keep the analysis and owe us nothing further. The list of your workflows with effort, saving, fixed price, data flow sketch and model recommendation belongs to you, and you can use it to build yourself or commission another provider. After each stage you decide afresh whether and how to continue. What the analysis contains exactly is set out under Approach and prices.

For B2B companies with 50 to 2,000 employees in German-speaking countries that have data and texts at volume and no AI development of their own. Typical sectors are industry and engineering, distribution, energy technology and cleantech, logistics, and professional services and consultancies. What matters is less the industry than the volume: thousands of products, dozens of enquiries a day or a CRM that nobody trusts. You can find examples under References.

For English-speaking decision-makers in companies based in Germany, Austria or Switzerland: an international management team, a parent company abroad, a works council that reads English. We work in English or German, as you prefer. Invoices are issued in euros by WDM Webdesign München GmbH, a German company; prices are net, plus VAT. Our working hours are German time (CET/CEST). The sections on the works council apply to sites in Germany.

For companies with fewer than about 50 employees and without a dataset or text inventory of their own, our analysis rarely pays off. We say so in the first call, not after the invoice. Nor do we build chatbots as decoration, or AI where an Excel formula is enough. The dividing line is volume, not size alone: a spare parts dealer with around 50 employees and millions of parts in the CRM is one of our cases. The potential check shows you in three minutes whether you are a good fit.

We have been in business since 1998, as a limited company (GmbH) in Munich, and have looked after a listed fuel cell manufacturer since 2017. The same specialists who carry out your analysis build the workflow and look after it afterwards. Anyone who has been building and running systems for over 25 years documents everything so that nothing depends on a single person. And because code, rules and documentation belong to you, your workflow does not depend on our continued existence anyway. More under About us and References.