German construction software company Phase0 has raised €2.6 million to expand AI agents designed to take over administrative work inside architecture and engineering practices, placing automation directly into tendering, costing, invoicing, and construction-site workflows rather than adding another general-purpose chatbot beside them.

The seed extension was led by construction and real-estate investor Realyze Ventures, with ATP architekten ingenieure chief executive Albert Achammer and former ServiceTitan chief growth officer Ershad Jamil joining as business angels. Existing investors include LocalGlobe and Atlantic.

Phase0, founded in Berlin in 2020 and previously known as Compa, provides browser-based software for small and mid-sized architecture and engineering businesses. Its platform covers cost estimation, bills of quantities, tendering and awarding, invoice verification, fee proposals under Germany’s HOAI framework, time recording, project controlling, and construction-site documentation.

The funding will be directed towards AI agents operating across those workflows. Phase0 already offers an AI tool that generates bills of quantities from project data and checks specifications against German VOB procurement and contracting requirements and DIN standards, while its site-diary system turns photographs and voice messages submitted through WhatsApp into reports, tickets, and emails.

Vertical AI moves into the back office

The attraction of this model is that the software already contains much of the structured information needed to automate administrative tasks. An architecture practice using Phase0 to prepare tenders, track costs, record time, and produce invoices does not have to move information manually into a separate AI service before asking it to carry out work.

That distinguishes vertical AI products from general-purpose assistants, at least in theory. The competitive advantage is less about access to a uniquely capable underlying model than about the workflow surrounding it: domain-specific data, integrations, permissions, audit trails, regulatory rules, and the point at which generated work is checked by a professional.

Phase0 founder and managing director Matthäus Kerres summarised the operational frustration behind the product: “Anyone managing million-euro projects shouldn’t be writing invoices in Word on a Sunday afternoon.” The company says more than 1,000 architecture and engineering practices now use its platform across Germany, Austria, and Switzerland.

Its target market illustrates why industry-specific software companies are becoming an important route for AI adoption. Phase0 estimates that Germany alone has around 110,000 small and mid-sized architecture and engineering practices, many of which still combine specialist desktop applications with spreadsheets, email, documents, and manually transferred project information.

Replacing those systems outright can be harder than adding AI to them. Construction and professional-services workflows are shaped by industry standards, contractual requirements, established data formats, and national rules, while smaller companies rarely have large internal IT teams able to stitch multiple automation products together.

Automation earns its place by understanding the job

Phase0’s approach is built around tasks with clearly defined inputs and outputs. Preparing a bill of quantities, extracting information from construction-site records, checking invoices, creating meeting minutes, or updating project cost information all involve significant administrative effort, but they also operate within existing processes against which outputs can be checked.

Those tasks are not risk free. A generated specification that omits a requirement, a wrongly categorised cost, or an inaccurate site record can flow into procurement, contractual, or financial decisions. As the software takes on more of the work between raw project data and a finished document, architecture and engineering businesses will need controls establishing who reviews AI-generated outputs, which source information has been used, and how corrections are recorded.

The problem becomes more pronounced as vendors move from assistants that prepare drafts to agents that complete sequences of tasks. Phase0 says future features will automate more work around costs, billing, documentation, and project management, bringing its software closer to acting on project information rather than simply presenting it.

For smaller professional practices, that could alter the economics of software adoption. Administrative roles and external support are expensive relative to the size of many architecture offices, while senior architects and engineers can lose billable or design time to reporting and paperwork. Automation only produces a productivity gain, however, if reviewing machine-generated work takes materially less time than completing the original task.

Realyze Ventures’ investment reflects a wider bet that sector-specific software providers can capture that value because they already sit inside the relevant process. Instead of asking construction businesses to adopt AI as a separate technology category, vendors can attach it to software used for tenders, invoices, project controls, or site records and charge for the resulting reduction in administrative work.

Phase0’s €2.6 million round is modest beside the capital flowing into model developers and AI infrastructure, but that is partly the point. The next phase of enterprise adoption will depend on thousands of comparatively narrow applications where AI has to understand invoices, contracts, technical standards, and established operating processes rather than demonstrate general intelligence.