Our AI Blog
Articles on AI agents, chatbots and automation for mid-sized companies. We write about what we build ourselves: GDPR-compliant systems for customer service, sales and back office - with concrete figures, legal context and real examples instead of futurism.
Zapier, Make or n8n - or a Custom AI Agent? The 2026 Comparison
Three toolkits and one custom build compete for the same question: who automates the process? This comparison shows why the billing unit matters more than the list price, where Zapier, Make and n8n each hit their limits - and at which point a custom AI agent becomes the honestly cheaper calculation.
Who Is Liable When AI Makes a Mistake? New Product Liability from December 2026
From 9 December 2026, manufacturers of AI software are strictly liable - the new Product Liability Directive counts software and AI systems as products for the first time. For users, less changes than the headlines promise: business damage remains a matter of contract. The real novelty is the manufacturer definition - anyone who offers AI under their own name or substantially modifies it becomes liable like a producer.
Which AI Agency Fits a Company Without Its Own IT Department?
Four provider types compete, but only one reliably works without in-house IT staff: the agency with contractual operation. What consulting, handover and managed operation each mean, what the models cost in 2026 and which five points reveal in a first call whether the provider actually runs the system.
What Does AI Automation Cost in 2026? Pricing Models, Hidden Costs and the Calculation for Mid-Sized Companies
Three routes to AI automation with real figures: toolkits from 20 euros a month, agency day rates of 1,200 to 2,500 euros and fixed prices per process. Plus the seven cost types, the items quotes leave out, and the calculation based on the German statistical office's hourly labour cost.
AI Policy in the Company: What It Must Contain to Work
42 percent of companies observe or suspect private AI accounts at work, only 23 percent have binding rules. What belongs in an AI usage policy, how the approval matrix for data and tools works, when the works council has a say and why a total ban fuels shadow AI rather than stopping it.
Works Council and AI Adoption: What § 87 BetrVG Requires Before Go-Live
Do I have to involve the works council before the AI assistant goes live? Usually yes, and not because of the AI but because of the data it generates about employees. Which systems are subject to co-determination, what belongs in the works agreement and how the procedure stays short.
ChatGPT in the Company: Trade Secrets, the AI Policy and What Works Technically Instead of a Ban
A ban stops nobody from pasting customer lists into ChatGPT; it only makes sure it happens in secret. What the Trade Secrets Act really requires, which version of the service tolerates which data and what a one-page AI policy looks like.
Position 1, Not a Single Click: How to Spot Machines Searching for Your Website
The Search Console for this website contains a query with eleven exclusion operators; the Bing data holds one with encoded line breaks lifted from a chat transcript. Neither could have been typed into a search box. This class of query grew from 2.95 to 23.1 impressions per day in five weeks, sits at an average position of 6.6 instead of 10.0 - and produced exactly one click across 218 impressions. Why click-through rate is becoming useless as a measure of success, and what replaces it.
The Cache Beats the Model: What Really Drives an AI Bill
Anthropic did not cut the token price of Fable 5.1 on 1 September. It cut the price of cache hits, from 1.00 to 0.25 US dollars per million. In a worked monthly bill that single line weighs more than the choice of model.
Claude Fable 5.1 Is Here: What the Model Means for Businesses
The strongest evidence for the release is checkable: on the Amazon Bedrock API the identifier claude-fable-5-1 flipped on 31 August from a 400 error to a 404 - from a name that does not exist to a registered entry.
Your Customers No Longer Search, They Brief the AI
Anyone looking for a provider today rarely types three words. They dictate a paragraph: situation, problem, instruction, usually ending with 'name specific providers'. These briefs appear verbatim in Google Search Console. In our own four-week data there are 16 of them, 211 impressions, not a single click. The reason sits in the positions: 9.8 for short terms, 61.2 for briefs. What causes that gap, how much of it is real buyers, and what a page must contain to survive a 43-word brief.
Where May the Model Run? LLM Hosting and Professional Secrecy
Doctors, lawyers and tax advisers in Germany have been allowed to use AI since Section 203 of the Criminal Code was reformed in 2017. What decides the matter is not the data processing agreement but whether every person with access to the inputs was bound and given notice of criminal liability beforehand. That is where large cloud offerings fail in their default configuration: abuse monitoring surfaces flagged inputs to a human reviewer. Four routes compared, six questions for the vendor call, and why open model weights change the picture.
Ox Alpha Was GLM-5.3-Flash: The Reveal and What It Means for Companies
For six days Ox Alpha was the most-used model on OpenRouter without anyone knowing the provider. On 26 August 2026 Zhipu identified itself: it was GLM-5.3-Flash, 320 billion parameters, served on Chinese accelerators, weights under an MIT licence. The benchmarks, the open data protection question from the test week, and why the licence is the real story for mid-sized companies.
Buying AI Visibility: What a Provider Must Prove Since the IAB Standard
More than 20 firms sell tools for measuring AI visibility, and the same brand gets different numbers from each. Since 3 August 2026 there is a yardstick from the industry body: four metric categories, two quality tiers and disclosure obligations. What the standard requires, which promises do not hold up technically, and the six questions that let you evaluate a quote.
AI Phone Assistants in Medical Practices: What the Law Requires
Practices usually test whether the provider of an AI phone assistant is GDPR-compliant. The rule that applies first sits in the criminal code and is aimed at the practice itself: failing to bind the provider to secrecy under Section 203 is an offence. On top comes the EU AI Act classification, which can tip where calls are sorted by urgency; following the Digital Omnibus the high-risk obligations apply from 2 December 2027.
Why Your Company Does Not Appear in AI Answers
Not appearing in AI answers is a different problem from being described wrongly. A study across 38 models shows there is a measurable threshold below which a company does not exist for the model, and that larger models change almost nothing about it. What works instead.
The AI Literacy Duty: What Article 4 Requires and What Does Not Threaten
Since 2 August, market surveillance authorities have been monitoring the EU AI Act, with the Bundesnetzagentur responsible in Germany. One obligation has applied for a year and a half already: AI literacy in the company. What it requires, why the circulating fine figures are wrong, and how to document compliance in half a day.
Why AI Describes Your Company Vaguely - and What Helps
The description is not wrong, yet it fits every second company in the industry. Why that comes down to attribution and omission rather than invention, and how to build a self-description that survives compression.
Sources Behind AI Answers: Where Mid-Sized Companies Can Actually Get In
AI systems answer predominantly from other people's sources. Why the common advice to get a Wikipedia entry leads nowhere for most companies, why the Reddit figures are overstated, and which sequence actually works.
AI in the Tax Firm: What § 203 StGB Permits and What It Does Not
The dividing line runs not between permitted and forbidden tools but between use with and without client data. What § 203 StGB and § 62a StBerG require, who is liable in a breach, and which tasks can be automated with legal certainty.
Measuring AI Visibility: Figures Instead of Gut Feeling
The board wants figures before releasing budget. Why a single ChatGPT query says nothing, how many repetitions a defensible mention rate needs, and which free measurement channel no monitoring tool shows you.
False Claims About Your Company in ChatGPT: What You Can Do
A prospect asks ChatGPT about your prices and gets a figure that never existed. Why there is no direct correction route, what actually works, how long it takes, and when legal claims apply.
AI Phone Assistant for Trades: Reachable Without Office Staff
Nobody can answer the phone on site, and office staff cannot be recruited. Why the circulating figures on missed calls are unsourced, how to measure your own rate in two weeks, and what an AI phone assistant can and cannot take over in the trades.
How Does an AI Agent Know What It May Answer?
A language model knows nothing about your company, it only phrases plausibly. How a connected knowledge base changes that, how high hallucination rates really are, and the four rules that prevent false statements.
AI Disclosure Duty: What Has Applied Since 2 August 2026
Since 2 August 2026, Article 50 of the EU AI Act applies: chatbots, voice bots and AI assistants must identify themselves. What that requires in practice, which deadline applies to existing systems, and why the Omnibus postponement does not cover it.
Hiring an AI Agency: 12 Questions for the First Meeting
Hiring an AI agency: 12 questions to ask in the first meeting - on integration, data protection, cost and operation. With warning signs and a checklist.
AI Phone Assistant for Medical Practices: 2026 Guide
AI phone assistant for practices: appointments and prescriptions automated, GDPR-compliant with EU hosting. One practice cut phone load by 42%.
Which AI Agents Exist? The 6 Types for Businesses in 2026
Which AI agents exist for businesses? The 6 types of 2026 with use case, channel and delivery time - and the question that decides which one fits.
Automating Back-Office Processes with AI: 7 Processes with the Fastest ROI, and What GoBD, E-Invoicing and the GDPR Demand
Incoming invoices, inbox, quotes, master data, appointments, reporting, onboarding: which process pays off first, which rule plays a part and what the ROI calculation looks like beforehand.
WhatsApp Business API: Costs & GDPR 2026 - The Practical Guide
What does the WhatsApp Business API cost and is it GDPR-compliant? Meta's message fees, the two cost blocks and why the Business App is not enough.
AI Phone Assistant vs. Answering Service vs. Voicemail (2026)
AI phone assistant, answering service or voicemail? The honest 2026 cost comparison for SMBs and practices - with cost models, table and decision guide.
What Does an AI Chatbot Cost for SMBs in 2026? Pricing Models and Calculation
What does an AI chatbot cost for SMBs in 2026? Three pricing models, six cost types, items quotes often leave out, and a calculation with your own figures.
AI Agents in HR & Recruiting: The Future of HR Process Automation for SMEs
Candidate screening, scheduling, faster replies - and the line at which the EU AI Act classifies applicant selection as a high-risk use case.
Standard Chatbot or Custom AI Agent: What Decides the Choice
No-code chatbot or custom-built AI agent? Five criteria SMBs can use to decide, and three questions every provider should answer in writing.
Introducing AI in a Mid-Sized Company: What the Managing Director Must Decide, Check and Answer For
Where AI pays off in mid-sized companies, which duties have applied since 2025, when the managing director is personally liable and how adoption succeeds in 90 days. The guide for decision-makers.
AI-Powered Software Development for SMEs
What AI-powered software development changes for custom software in SMEs, where it saves time and what matters for architecture, testing and maintenance.
AI Agents Alone Don't Drive Growth: Strategy, Brand & AI
A chatbot or voice agent rarely lifts revenue on its own. Why real growth only emerges when strategy, brand and AI work together as a system.
Claude Opus 4.8 Is Here: Release, Benchmarks, and What Anthropic's New Model Means for Businesses
Anthropic released Claude Opus 4.8 on May 28, 2026 - its most capable model yet, at the same price as 4.7. All the facts, benchmarks (SWE-Bench Pro 69.2%), new features, and what the release means for businesses.
GEO Beyond SEO: How Mid-Market Companies Win Visibility in ChatGPT, Perplexity and Google AI Mode in 2026
AI answers are changing how customers search: when an AI summary appears, far fewer people click on results. What Generative Engine Optimization is, which levers are backed by studies and which widespread advice does not stand up to scrutiny.
EU AI Act for AI Agents in Mid-Sized Companies: Which Duty Applies to Which Use?
Since 2 August 2026 the transparency duties of the EU AI Act apply. One table maps ten typical use cases in mid-sized companies to risk class, duty, deadline and responsible role.
AI Automation for SMEs: A Guide to Boosting Efficiency in 2025
36 percent of German firms cannot fully fill open positions, according to the DIHK. Which processes can be automated first, how to recognise them, and why most projects fail.
WhatsApp Business API & AI: How Companies Generate More Revenue
WhatsApp reaches customers more reliably than email. But without documented consent under Section 7 UWG, every promotional message is a legal risk. What has to be settled first.
AI in Marketing: What Personalization Really Means for a Mid-Sized Company
Personalization in mid-sized firms comes from rule-based segmentation using data already sitting in the CRM. Which three use cases pay off, what the legal position is, and why most projects fail.
Discover AI potential for your business
Let us find out in a short call how AI agents can optimize your processes.