The workload is the whole story
Engineering consultancies have a computing profile almost nothing else in small business shares. A single Revit model can be gigabytes. Opening it, synchronising to a central file, and running clash detection move data at volumes that expose every weakness in a network and storage design.
An environment that is entirely adequate for a business running Office and a browser will make engineers wait several minutes for something that should take twenty seconds. That waiting is charged time, and it accumulates into a genuine cost that nobody attributes to the infrastructure causing it.
Specifying workstations properly
The most common expensive mistake is buying on core count.
Much CAD and Revit work is single-threaded. A processor with a high clock speed and fewer cores frequently outperforms a many-core chip running slower, for a lower price. Rendering and finite element analysis are the exceptions and do scale across cores — but they are a minority of the working day for most engineers.
The practical priority for typical Autodesk workloads is clock speed, then RAM, then a certified professional GPU, then fast local storage. Getting that order right means the money buys performance rather than specification sheet.
This is why procurement matters more here than in an office business: a standard fleet specification applied across an engineering firm either overspends on administrative staff or underserves the designers.
Where the models should live
Not a question with one answer.
A single-office firm with everyone on the local network is usually better served by a local server. Model synchronisation is bandwidth-heavy and latency-sensitive, and a well-configured local file server beats a cloud round trip.
A firm with multiple offices, remote staff, or heavy collaboration with architects and other consultants usually does better on a cloud collaboration platform. The alternative — synchronising large models across a VPN — is a reliable source of complaints.
The wrong approach is deciding on principle rather than on how your teams are actually distributed.
Intellectual property
Design work is the firm’s asset, and it walks out on laptops and in downloads. Complete prevention is not achievable while people need legitimate access to do their jobs.
What is achievable: access scoped to the projects someone actually works on, logging so that bulk file access is visible rather than invisible, and an offboarding process that revokes access promptly and looks at what was accessed in the weeks beforehand.
That converts IP loss from undetectable to detectable, which is a meaningfully different position, particularly if it ever becomes a legal matter.
Archives that still open
Professional liability means models and calculations are retained for years. Two things go wrong.
The archive stops being findable, because it was moved to cheap storage with no index and the person who filed it left. And the files stop being openable, because the software moved on several major versions.
Keeping native files alongside a neutral format — IFC, PDF, or a plotted set for the record — addresses the second. Maintaining structure and metadata addresses the first. Both are decisions to make when archiving, not when the file is eventually requested.