An AI assistant for the HOAI and VOB construction regulations
Project management in construction, German market
Industry and type of organization
A company running project management in construction on the German market. It bases its estimates and settlements on two sets of rules: HOAI, the fee regulation for architects and engineers, and VOB, the rules for awarding construction contracts.
Business problem
The knowledge needed for a decision is usually already on hand within the company. It sits in contracts, manuals, documentation or in the regulations the company works with. The trouble is access: you have to know where to look and have the time to search. Until that moment, the decision stalls.
At this client it looked like this. The dispute began on the construction site: the site manager put forward a claim, and to confirm or refute it you needed a commentary on the HOAI at hand. No one had it, so the conversation broke off.
Next came the return to the office, about an hour and a half of leafing through commentaries in search of the right explanation, and only then the further debate. With more serious doubts, a call to a lawyer was added and another day of waiting. A mistake in the fee basis raises the cost of the whole project.
The industry name is interchangeable here. The same pattern repeats with supplier contracts, quality procedures, insurance terms and internal instructions. The document changes, not the mechanism.
Approach and scope of work
We built an assistant that answers questions about HOAI and VOB based on the regulations themselves and their interpretation, available wherever there is coverage. Instead of going back for the commentary, the employee asks about their case on the spot and gets the answer together with the basis it follows from.
This is the RAG methodology, retrieval-augmented generation, one of the applications of generative AI. The model does not base the answer on its memory alone; it retrieves the right passage of the source and builds the answer on it. This makes every answer verifiable, and that is the necessary condition when it comes to the fee basis.
The assistant takes one more step. Beyond the answer itself, it proposes the next best action: which paragraph to apply in a given setup, what else to check and where in the documentation to look for confirmation. The employee immediately knows what to do with the basis they have, instead of being left with the regulation alone.
Business effect
10 minutes
instead of 1.5 hours
reaching an answer together with its justification
What changed most is the place where the answer is produced. It is produced where the question is asked, so the conversation does not break off and start over the next day at the office. The reduction from an hour and a half to about ten minutes is the result of that.
The character of the discussion itself changed too. Earlier it was settled by whoever remembered more or spoke more confidently. Now each side can call up the regulation and its interpretation in a few minutes, so the dispute moves from the level of beliefs to the level of facts.
Knowledge of HOAI and VOB stopped being the asset of a few people in the company. A new employee asks the system instead of looking for someone experienced, and the experienced ones are no longer a bottleneck.
Scale of the implementation
The source base runs to several thousand pages. The assistant covers HOAI and VOB together with interpretations and is available when the question comes up, including away from the office.
We maintain it and update the base when regulations change, because an answer based on an outdated wording is worse than no answer.
Security and confidentiality
The assistant runs in the AWS cloud in the Frankfurt region, so the data stays within the European Union. For a client in a regulated industry this is the condition for putting a solution into operation.
We state one limitation plainly: the tool shortens the way to the right question, cites the legal basis and suggests the next step, but it does not replace legal advice. Where the matter is disputed, the decision still belongs to the lawyer.
If the answers your team needs sit in documents that no one has time to search, describe your case to us. We will tell you whether this path can be shortened, including when, in our view, it is not worth automating.
Related articles in the cm-opti knowledge base
Service area: AI agents and RAG systems on company documents.