Project-based firms have a billing problem hiding in plain sight. Engineers and architects bill at CHF 120–180 per hour. Between a fifth and a third of their working week — the share varies by role and firm size — goes on tasks that never appear on a timesheet: scanning procurement portals for new tenders, assembling compliance checklists from regulation PDFs, cleaning up meeting notes into action lists, and chasing half-finished RFP sections across email threads.
None of that is billable. All of it is necessary. And it is precisely the kind of structured, repeatable, document-heavy work that AI agents handle well.
This article covers where AI agents for engineering firms actually move the needle — and where they still fall short.
Why Utilisation Rate Is the Right Metric
Before jumping to use cases, it helps to frame the economics correctly. For a professional services firm, the number that matters most is the billable utilisation ratio — what percentage of paid working time is logged against client projects. A typical mid-sized engineering firm might run at 65–80% billable utilisation — the range is wide and role-dependent. The goal is not to make people work faster. The goal is to shrink the non-billable hours so utilisation climbs.
Even a modest recovery matters. Take a 12-person team of senior engineers billing at CHF 150/hour. If an agent workflow saves each person 90 minutes a week on admin tasks, that is 18 hours per week across the team — around CHF 2,700 in recaptured billing capacity, every week, without hiring anyone.
That is illustrative math, not a guarantee. Real recovery depends on which tasks you automate, how cleanly your document workflows are structured, and whether your team actually redirects saved time to billable work rather than longer lunches. The point is that the lever is utilisation, not headcount.
RFP Triage: The First Win Most Firms Leave on the Table
Monitoring procurement platforms, reading 80-page tender documents, and deciding within 48 hours whether to bid — this process absorbs serious senior-staff time, often across multiple portals in multiple languages if you operate across Swiss cantons or European borders.
An AI agent built for RFP triage can do the following without human intervention:
- Pull new publications from configured procurement feeds or portals on a set schedule
- Extract scope, deadline, qualification thresholds, and key requirements from the downloaded documents
- Score bids against a go/no-go rubric you define — project size, geography, technical match, client history
- Draft a one-page pre-qualification summary and route it to the responsible partner with a recommended action
What this does not do: make the final bid decision. That judgment still belongs to a senior person who understands your firm’s capacity, risk appetite, and client relationships. The agent handles the triage so that the 90-minute read-and-discuss becomes a 10-minute partner review.
For firms that track win rates on submitted proposals, there is a secondary benefit: better-filtered bids mean you pursue fewer long-shots, improving the ratio of time invested to contracts won.
Meeting-Minute Processing: The Quiet Time Drain
Every project review, client briefing, and coordination call generates the same follow-up ritual. Someone types up notes. Action items are scattered across a paragraph of prose. The project manager copies them into a tracker — maybe. Three days later, someone asks what was agreed.
A document processing agent connected to your meeting recording or transcript service can compress this cycle to under five minutes:
- Transcript arrives (from your video platform, a dictation app, or a direct recording)
- Agent extracts structured action items, owners, and deadlines
- Output is written into your project management tool, with items linked to the relevant project and assigned to the right team member
- A short summary email is drafted for client distribution with one-click send
The accuracy depends heavily on transcript quality and how consistently your meetings follow a recognisable structure. Technical coordination calls with clear deliverables convert well. Messy ideation sessions with fluid responsibilities are harder. Start where the signal is clean.
Norm and Compliance Checklists: The High-Risk, High-Value Use Case
Architecture and engineering work is governed by a dense stack of norms — SIA standards in Switzerland, EN Eurocodes across Europe, local cantonal building codes, energy performance requirements, fire safety regulations. Keeping track of which version of which standard applies to a given project, and confirming that a design document addresses every relevant clause, is tedious and error-prone.
This is one of the highest-value applications of AI agents for architecture firms — and also one of the most sensitive.
An agent can be trained on a firm’s internal norm library and configured to:
- Generate a project-specific compliance checklist based on building type, use category, and canton
- Cross-reference submitted design documents against the checklist
- Flag gaps or clauses that need manual sign-off
The important caveat: an agent-generated checklist is a first-pass tool, not a professional sign-off. The output needs review by a qualified engineer or architect before it influences design decisions. The value is in the time saved assembling the checklist and doing the initial document pass — not in removing the professional judgment at the end.
Used correctly, this reduces the risk of missing a clause during a busy phase, and gives less experienced team members a structured starting point rather than a blank page.
This kind of multi-step, document-aware workflow is exactly what agentic workflows are designed for — chaining retrieval, reasoning, and output into a single automated process rather than a chain of manual steps.
Where AI Agents Are Not the Right Fit (Yet)
Some tempting use cases do not hold up in practice at current capability levels:
Structural calculations and engineering analysis. Agents can retrieve references and flag inconsistencies, but they cannot replace the verified computation tools that underpin structural design. The liability exposure alone makes this a hard boundary.
Client-facing design feedback. Architectural clients expect nuanced, contextual responses to design questions. A generic AI response that does not understand the project’s history, the client’s aesthetic preferences, or the regulatory context will do more damage than good.
Complex contract negotiations. Agents can summarise terms and flag deviations from a standard contract template, but the negotiation itself requires human relationship management and legal judgment.
The pattern is consistent: agents work where the task is structured, the inputs are defined, and the output needs human review before it creates downstream consequences. See the broader AI agent ROI framework for how to score potential use cases against effort and risk before committing.
What a Realistic Implementation Looks Like
Firms that have deployed these workflows — whether engineering, consulting, or other project-based businesses with similar document workflows — typically start with one well-scoped pilot rather than a full rollout.
A reasonable starting sequence:
- Pick one workflow with clean, structured inputs — the tender feed is often the easiest because the documents are standardised
- Define the output format your team actually uses — an agent that produces output nobody reads is wasted
- Run in parallel with the manual process for four to six weeks to calibrate accuracy and build trust
- Measure the time delta before deciding to expand
The firms that get stuck are usually those that tried to automate too much at once, or that underestimated the integration work required to connect an agent to their existing project management stack. Custom agent development matters here — off-the-shelf tools rarely understand the specific document formats, procurement portals, or norm libraries a mid-sized engineering firm relies on.
For a comparable picture of how project-based professional services firms approach this, the AI agents for consulting firms article covers similar ground from a strategy and advisory angle.
The Business Case in Plain Terms
AI agents for engineering firms are not infrastructure investments that pay off in five years. The operating cost of a well-built agent workflow is typically a fraction of the labour cost it replaces. The implementation effort is the variable — it depends on how structured your existing processes are, what systems need to be integrated, and how much customisation the domain requires.
For a firm serious about recovering billable utilisation, the question is not whether to automate these workflows. The question is which one to start with, and whether to build it in-house or bring in a specialist.
Orange ITS designs and builds custom AI agents for professional services firms across Switzerland and Europe. We do not pitch generic platforms — we scope the workflow, build the integration, and measure the utilisation impact.
If you want to map these use cases against your firm’s specific situation, book a 30-minute call with our team. We will walk through your highest-leverage workflow and give you an honest view of what an agent could realistically deliver.