Local language models in the office: performance and cost
What your own AI server in the office can do, where it falls behind large cloud models, and what it costs in the first year in an example calculation.
Knowledge
Articles for design, construction and existing buildings
Briefly explained, with sources and without sales talk. The articles describe the situation in Germany and are translations of the German originals.
-
-
Receiving e-invoices: invoice intake in a construction company
How an e-invoice gets from the mailbox to the archive via allocation to the building project and site management approval, and who clarifies what.
-
Cost estimate to DIN 276 using your own benchmark figures
How a DIN 276 cost estimate is built from your own benchmark figures: cost groups, price level, construction price index, assumptions list and range.
-
Interim and final invoices and measurement in e-invoicing
What the BMF letters say on interim invoices, final accounts, the bill of quantities and site measurement as an attachment, and the relief until 2030.
-
Running AI in the office in line with the GDPR: three options
Cloud in the EU, a German provider or your own server: what data processing agreements, place of processing and training mean, and what risk remains.
-
XRechnung and ZUGFeRD: what the data set holds, how to read it
XRechnung is an XML file, ZUGFeRD a PDF with embedded XML. What the data set contains, what validation checks and how to view it for free.
-
Checking a bill of quantities: common errors and AI support
Common errors in a bill of quantities and what an AI-assisted check finds: notes rather than an expert opinion, with calculation done in a fixed program.
-
E-invoice deadlines in construction: which date applies to you
You must be able to receive e-invoices since 2025 and issue them from 2027 or 2028. Which deadline applies and why a PDF by email is not enough.
-
Retaining e-invoices: data set, period and archive
The data set is the original: what to archive for e-invoices, for how long, what the GoBD require and how to assign invoices to a building project.
-
Issuing e-invoices: preparing a design practice
What design practices and small firms should settle before the deadline: routes to an e-invoice, master data, fee and interim invoices, test run, sign-off.
-
Introducing AI in a small office: the first 90 days
Introducing AI in a small office in 90 days: measure first, then pilot one construction process with a contact person, training per role and re-measuring.
-
Is AI worth it in a small office? A calculation with assumptions
Is AI worth it in a small office? Costs set against hours gained, with invented example data, open assumptions and the checking time included.
-
EU AI Act: which dates matter for offices that only use AI
Which stages of the EU AI Act affect an office that only uses AI: AI literacy, transparency and high risk, with the status of the Digital Omnibus.
-
Preparing acceptance: defects list, documents and minutes
Preparing acceptance in a building firm: what to assemble before the date, how defects are recorded by voice and photo, and what the minutes record.
-
Searchable building files: inspection dates and contracts
Find maintenance contracts, inspection reports and as-built records with a source reference, prepare a deadline list. What stays out, such as tenant data.
-
Checking variations: variation, BoQ and contract side by side
Checking a variation means comparing it with the bill of quantities and the contract. How a deviation list prepares the check and what stays with you.
-
Project mailbox: sorting and filing emails with AI
What an AI can prepare in a project mailbox: classify, summarise threads, draft replies. And what you should not let happen automatically.
-
Cost monitoring under DIN 276: budget, orders, forecast
Cost monitoring under DIN 276: budget, orders, variations and forecast per cost group in one overview. Who calculates, who matches and who decides.
-
Searching a contract: deadlines and duties as an AI-built list
A list of deadlines and duties with source references from a design or construction contract: what AI finds, what stays with you, how sampling helps.
-
Shadow AI: when staff use private chat services
What shadow AI is, how widespread it is per Bitkom and the IW, which risks it poses for project data and what a company can offer instead of a ban.
-
Capturing knowledge: old projects as a reference library
Preventing knowledge loss in construction: which old-project records suit a searchable reference library, which do not, and what retention periods mean.
-
Preparing delay and concern notices under the VOB/B
Preparing a delay notice and a notice of concerns in a building firm: what sections 6 and 4 VOB/B provide and how AI drafts from your records.
-
Checking interim payment applications: invoice, BoQ, measurement
Checking an interim payment application means comparing invoice, BoQ and measurement. A deviation list with source references prepares it.
-
Prepare a monthly client report with AI in project management
Prepare the monthly report to the client with AI: structure, sources per chapter, checking steps and what stays with the project lead.
-
AI training by role: content for the office and the site
What managing directors, project leads, site managers and office assistants should know about AI: content by role, exercises, duration, record.
-
AI policy for the office: what it should contain
An AI policy for the office in five steps: approve tools, classify data, check results, label use, train staff. With the current position on Art. 4 AI Act.
-
Why AI answers need a source reference
Language models invent answers when they guess. How source references, spot checks and review steps keep wrong statements out of your contracts.
-
Preparing a price comparison: tenders side by side
Preparing a price comparison: a fixed program checks sums and compares prices, the AI flags text anomalies, and the assessment stays with you.
-
Data processing agreements for AI services: what to check
Using a cloud AI service usually calls for a contract under Art. 28 GDPR. What to look for in it, plus a list of questions to put to the provider.
-
Site diary and defects: record them by voice and photo
Record the site diary and defects list on site by voice note and photo, check in the evening instead of typing. What the AI prepares, what stays with you.
-
GAEB file explained: X81, X83, X84, X86 and what is inside
What a GAEB file contains, what the exchange phases X81, X83, X84 and X86 mean, and why structured data makes checking bills of quantities easier.
-
Old plans and poor scans: what OCR can and cannot do
What OCR achieves on scans: my own invoice test (99.85 per cent against 64.2 per cent) and limits with handwriting, title blocks, tables, dimension chains.
-
Tracking open items: a list from meeting minutes with AI
How meeting minutes become a list of open items with owner and date, what the AI proposes and what the project lead confirms.
-
Recording meetings: what to clarify first
Recording meetings for AI minutes: consent of all, section 201 StGB, GDPR, notice in the invitation, deletion period and works council at a glance.
-
Writing AI work instructions: prompts in the building process
How to give an AI a work instruction: role, documents, task, form and a check step, with one running example and the typical mistakes to avoid.
-
Project folder structure and file names: order before technology
Consistent folders, file names with date and drawing revision, clear versions: why search and an AI knowledge base depend on them and how to begin.
-
Searchable project files: a knowledge base with AI
How an AI knowledge base searches your project documents: find passages first, then answer with a source. And why order comes before technology.
-
Meeting minutes with AI: from recording to sign-off
How a recording becomes signed-off minutes, where AI saves time, where it does not, and why responsibility stays with the person who writes them.
-
Measure workflows before introducing AI: log one week
Where the time goes in an office shows after one week of logging: frequency times minutes per workflow. How to spot a workflow that suits AI.
-
Language models: what they can and cannot do, for construction
A language model reads, sorts and drafts text, but it does not calculate reliably and can invent facts. What that means for who does what.
-
MCP and AI agents: what is behind them and what they may do
An AI agent chooses its own steps, and MCP is the standard for its tools. What an agent may do in a small business and what needs approval first.
-
E-invoices in the trades: automating invoice intake
Trades businesses have had to receive e-invoices since 2025. How intake, checking, matching and filing can run as a workflow, and what people still decide.
-
Knowledge assistant on your own documents: local or EU cloud
A knowledge assistant answers questions from your own documents and names the source. What vendors say about local and EU-cloud processing.
-
Following up quotes and appointment requests: a workflow
Following up quotes and appointment requests in a trades office: a workflow with a list and a draft, section 7 UWG in its wording, and a worked example.
-
What does an automation project cost? Prices with date
What does an automation project cost? Published fixed prices and day rates with source and date, November 2026, plus cost types and funding.
-
Handing over an automation: a checklist for your business
What a business must own after an automation project: workflows, access, key and documentation, based on n8n docs and the BSI, with a checklist.
-
AI texts people enjoy reading: minutes without filler
How to use AI for minutes, reports and emails that people enjoy reading: write the brief, cut the draft, name who does what by when and read it aloud.
-
n8n updates and backups: who looks after them?
Who updates n8n and backs up the credentials? Security advisories read at the source, what a complete backup contains, and a simple rule to adopt.
-
Silent failures: how to tell that an automation has stopped
How a business notices that an automated workflow has failed: error alert, sign of life, count check, and the question of who reads it.
-
Data zones: which data may go into which tool?
Data zones for automation: which data may go into which tool, when Article 28 GDPR requires a contract, and what the n8n contract does not cover.
-
Sorting enquiries in a trades office: what can be automated
Sort enquiries from form, email and phone note in advance: what a workflow takes over, what the office decides and why emails are untrusted text.
-
AI in a workflow: one AI step, fixed rules, human approval
How do you keep AI under control in an automated process? A fixed workflow, a single AI step and human approval: the pattern, with an invented example.
-
n8n licence: what a business and a service provider may do
The n8n licence against its wording: what a business and a service provider may do with self-hosted n8n, what the vendor FAQ says and what stays open.
-
Cloud or self-hosted: what automation really costs
Run automation in the cloud or host it yourself? Prices of n8n, Zapier, Make and Power Automate with date, billing units and the effort of running it.
-
Incoming invoice processing time: where the real lever is
An incoming invoice takes 6 to 9 minutes to book, says an IW survey. Why capturing it saves little and where the real lever lies in checking the invoice.
-
AI in the skilled trades: only 4 per cent of firms use it
AI in the German skilled trades per Bitkom 2025: 4 per cent use it, 85 per cent offer digital services. What the figures measure and what follows.
-
Measuring time savings: cases times minutes, before and after
How much time does automation really save? Calculate cases times minutes, measure before and after, and read two studies correctly.
-
Red tape and working time: what 3, 7 and 23 percent measure
Depending on the study, red tape eats 3, 7 or 23 percent of working time. What KfW and IW each asked, and why the figures should not be mixed.
-
When you should not automate a process
Rare, judgement-based or legally sensitive tasks, and tasks full of exceptions, are better left to people. With a worked calculation and check questions.
-
Process automation in a small business: the basics
What process automation means in a small business: trigger, steps and connections, four tools described neutrally, and how to recognise a suitable process.