Where Should an Architect or Engineering Firm Start With AI?

AI adoption in A&E has moved quickly.

 According to the 2026 AEC Inspire Report, 75% of firms are already using AI in some form. But only 29% report high confidence in the data feeding those tools. ACEC is seeing a similar shift: nearly half of engineering firms now report having a formal AI strategy, although adoption varies dramatically by firm size.  

So for most firms, the question is no longer whether to experiment with AI.

It is:

Where do we actually start?

You need a strategy.

You need governance.

You need to make thoughtful platform and technology decisions.

But I wouldn't start with only any one of those things.

Start with the work.

At the end of the day, AI isn't the objective. The objective is helping your A&E business do what it already does—win work, deliver projects, manage people, protect margin, serve clients, and run the firm—better.

So look first at the business processes where the work is repetitive, context is scattered, people spend too much time assembling information, or the outcome has a meaningful impact on growth, margin, efficiency, or risk.

Then let the work inform the rest.

It should help shape your AI strategy by showing where AI can create meaningful business value.

It should inform your governance priorities by revealing where sensitive data, client requirements, risk, and human judgment matter most.

And it should inform your platform decisions by clarifying what capabilities, connections, and workflows you actually need.

That's an important distinction.

The strongest AI solutions will connect enterprise strategy with the reality of how the firm actually works.

Establish the rules of the road

Starting with the work doesn't mean skipping governance.

Before firms begin putting AI into meaningful business workflows, leadership should be able to answer some basic questions:

What AI tools are approved?

What information can employees put into them?

How will confidential and client information be handled?

What do our contracts require?

Where is human review mandatory?

Who remains accountable for the final decision or deliverable?

Those aren't theoretical questions for an A&E firm. They touch client confidentiality, professional responsibility, security, intellectual property, and risk.

ACEC has similarly emphasized AI policies, data classification, governance, risk management, and vendor vetting as important foundations for responsible adoption.

I don't interpret that to mean firms need months of committee meetings before anyone can experiment.

But you should know your boundaries.

A strong policy should give people enough clarity to move confidently, not another reason to stand still.

And once those guardrails are in place, firms can spend much more of their energy on the question that really matters:

Where can AI make our work meaningfully better?

Look for the right kind of process

When I'm evaluating a potential AI use case, there are four characteristics I look for.

It requires context. The process depends on information from multiple places: ERP data, CRM records, project information, email, documents, meeting transcripts, financials, or other business systems.

It is repeatable. This isn't a one-time exercise. People perform some version of the work over and over.

It has clear business value. Improving it could save meaningful time, support growth, protect margin, or reduce risk.

It contains points of human judgment. AI can do more of the gathering, analysis, and preparation while a person remains in the decision seat.

That last point may sound counterintuitive.

We're often attracted to use cases because we want AI to automate the whole thing.

But especially when firms are starting, I think some of the best applications are those where AI does more of the work around a decision while an experienced person remains responsible for making it.

That's not a limitation.

It's often where the value is.

What could this look like in practice?

If you're struggling to identify a first process, here are three examples I've seen resonate with A&E firms.

Opportunity and pursuit signal monitoring

Growth opportunities don't always arrive neatly as a new record in CRM.

They show up in client emails, meetings, association notices, conversations, public information, and relationship activity.

AI can monitor those signals, bring the relevant context together, compare a potential opportunity against the firm's experience and past performance, and surface the ones that deserve someone's attention.

The point isn't for AI to decide what your firm should pursue.

It's to help your business development team recognize potential work earlier and give the right person enough context to decide whether to lean in.

Project handoff from pursuit to delivery

Think about everything a firm learns while winning a project: what the client cares about, commitments made during the pursuit, staffing assumptions, scope discussions, schedule expectations, financial assumptions, risks, and dozens of small pieces of context buried in emails and meetings.

Then the project is won—and too much of that intelligence has to be rediscovered by the delivery team. For some projects, it’s a requirement to get the coveted project code for charging time.

AI can help gather that context and prepare a project-start brief for the PM:

What did we promise?

What assumptions did we make?

What does the client care about?

What risks have already surfaced?

What is still unresolved?

This is a good example of AI improving a process that firms have tried to improve for years.

The goal isn't a more impressive project kickoff document.

It's a cleaner transition from winning the work to delivering it.

Project profitability and emerging risk

A third place to look is project performance.

Rather than simply asking AI for another profitability report, firms can use it to compare current project activity against historical performance and look for meaningful deviations: changing labor patterns, unusual WIP, margin movement, shifts in billable mix, unexpected effort, or other indicators that a project may be moving off course.

AI can surface what looks unusual, explain why it deserves attention, and recommend where a PM or operations leader should investigate next.

Again, the AI isn't making the project decision.

It's helping an experienced person see the signal sooner.

These are very different business processes, but they have something important in common:

The information already exists, gathering and interpreting it takes work, human judgment still matters, and improving the outcome creates measurable business value.

That's the pattern I'd look for.

Reporting may be the easiest place to begin

If those use cases feel ambitious, reporting can be one of the lowest-friction places for an A&E firm to start.

Think about a dashboard you use regularly.

Today, you might open it, click through several screens, apply filters, select a date range, find the relevant project, and eventually get to the number you need.

AI changes that interaction.

Instead of navigating the report, you can ask:

Is this project profitable?

And once you have the answer:

Why is profitability declining?

Which phases are driving the change?

Are we consuming labor faster than planned?

How does this project compare with similar projects we've delivered?

What indicators should I be watching over the next 30 days?

The important innovation isn't simply replacing clicks with a chat interface.

It's the ability to continue the analysis.

As I've written previously, traditional systems have been good at producing the report. AI gives us the opportunity to ask the thousand questions that follow the report.

That makes existing business data much more accessible.

A project manager can investigate budget or schedule variance.

A finance leader can dig into WIP.

An operations leader can examine capacity against backlog and pipeline.

An executive can interrogate a forecast rather than simply receive one.

You don't need to begin with autonomous AI running the firm.

You can begin by making information you already trust dramatically easier to understand and explore.

Personal productivity is a good starting point—as long as it doesn't become the finish line

There's also nothing wrong with starting personally.

In fact, personal productivity can be one of the best ways for a business leader to build confidence with AI before expanding into broader workflows.

I spoke recently with a CFO who is now deploying AI across finance and more broadly throughout his organization.

But his journey didn't begin with a firm-wide transformation initiative.

It began with his own spreadsheets.

He looked at some of the highly manual work he was doing personally and asked where AI could help.

One of those processes happened at month-end. He would compile a series of reports and distribute them to stakeholders across the business. It was necessary, repetitive, and time-consuming.

He automated it.

Then he went on vacation—literally.

While he was at the beach, the reports went out as expected. One of his stakeholders emailed him essentially saying, I thought you were on vacation. Thanks for taking the time to send this.

He hadn't.

That might sound like a small win. But small wins matter.

It gave him time back. More importantly, it built confidence. He had evidence that AI could reliably take meaningful work off his plate.

And this is where I think his experience is particularly instructive:

He didn't stop there.

He used personal productivity as a proving ground. The confidence he built solving his own problems gave him the motivation and experience to begin applying AI to shared processes across finance and eventually the broader organization.

That's the progression firms should be looking for:

Personal productivity → confidence → shared business processes → organizational value.

The trap isn't starting with personal productivity.

The trap is staying there.

If every employee finds five ways to write emails faster, summarize documents, or clean up spreadsheets, the firm may become incrementally more productive without ever changing how the business operates.

Personal productivity can be an excellent starting point. It just shouldn't become the destination.

A good use case still needs an owner

This may be one of the most overlooked parts of AI adoption.

A compelling use case doesn't implement itself.

Take opportunity signal monitoring.

An AI workflow might be capable of monitoring incoming information, identifying potential opportunities, comparing them against the firm's experience, and bringing the most relevant ones forward.

Technically, that's increasingly achievable.

But who owns it?

Who decides whether the recommendations are useful?

Who measures whether it is actually improving opportunity identification?

Who adjusts the workflow when it isn't?

Who teaches seller-doers how to benefit from it?

Who drives the behavior change required for the firm to realize the value?

Without an owner, a promising pilot can remain exactly that: a promising pilot.

One of the biggest lessons I'm taking from firms that are moving beyond experimentation is that AI adoption needs business ownership, not just technical enablement.

The technology can work perfectly and still fail to create value if nobody owns the outcome.

Prove trust before you scale

One of the best examples I've seen recently came from the CTO of a large architecture firm.

Their technology team had already developed a suite of Power BI dashboards used throughout the organization. Finance, leadership, project teams, and others knew those dashboards and relied on the information inside them.

When the firm began experimenting with AI, they had a benchmark.

They could ask the AI questions and compare its answers against information they already trusted.

Was the number right?

Was the analysis accurate?

Did the response point the user in the right direction?

Could people access it easily enough that they would actually use it?

The firm started within the technology organization.

As confidence in the accuracy grew, they expanded into finance, where teams could put additional scrutiny around accuracy and governance.

From there, they began moving into project and operational areas of the business.

That's a useful model.

Innovate. Validate. Build trust. Expand.

The technology didn't earn broader adoption because someone declared it transformative.

It earned adoption because people repeatedly saw it produce useful, accurate answers.

And accessibility mattered too.

The firm's dashboards had already established a familiar place where people knew to go for information. AI needed to meet the same standard.

Intelligence isn't particularly transformative if employees can't easily reach it in the flow of their work.

Don't confuse an AI strategy with a list of AI tools

Over the next several months, A&E firms are going to feel increasing pressure to move from AI experimentation toward something more deliberate.

Some of that pressure is justified.

But buying several AI tools is not the same thing as having an AI strategy.

Neither is running dozens of disconnected pilots.

PSMJ has increasingly framed the industry's AI conversation around moving from hype toward ROI and practical workflows—and I think that's exactly the right direction.

The question shouldn't be:

Where can we put AI?

It should be:

Where does our firm spend time, lose margin, introduce risk, or miss opportunities because people cannot assemble the context they need quickly enough?

Start with the work.

Let the work inform the strategy, governance, and platform.

Find a process where the value matters.

Keep people responsible for the decisions that require human judgment.

Assign someone to own the outcome.

Prove that the information can be trusted.

Then expand.

The goal isn't to become the A&E firm using the most AI.

It's to become a better A&E firm because of how you use it.