Why AI implementation stalls at architecture and engineering firms
Most firms already pay for Copilot, ChatGPT or Claude, and a few people use them every day. This guide covers the six points where implementation stops between the first pilot and firm-wide use, and what to do about each one.
Where implementation stalls
The first stage of AI in an architecture or engineering firm usually goes well. Leadership approves a tool, a few people try it on proposals, meeting notes or renderings, and someone shows a good result at a staff meeting. Six months later the same few people use it, the rest of the firm works the way it did before, and nobody can say what the pilot saved.
BCG splits the effort of scaling AI into 10% for algorithms, 20% for technology and data, and 70% for people and processes, meaning training, workflow redesign, roles and governance. The six stall points below all sit in that 70%, and each has a fix a firm can make without buying anything new.
Six stall points
No defined workflow
A workflow written down before the pilot gives the pilot something to measure and someone to own it. For an RFI response, the trigger is an RFI logged by the contractor. The steps are finding the relevant spec section and drawings and drafting an answer. The reviewer is the project architect, the output is the response in the RFI log, and the record is that log entry with its sources. The prompt is one step inside the workflow, and the tool choice comes after it.
Licenses without a rollout
In the 2024 Work Trend Index, Microsoft and LinkedIn found that 78% of people who use AI at work bring their own tools, and LinkedIn found that 39% of users had received AI training from their employer. Handing out licenses adds one more tool to that mix. A rollout tied to one workflow, with training on that workflow and a check on who uses it after the first month, tells you whether the licenses are earning their cost. The tools staff bring themselves show where the approved ones fall short, which our article on shadow AI at AEC firms covers.
Scattered context
An AI answers from what it can read. When the current fee schedule, the latest spec template or the approved project description sits in someone's inbox, the AI reads an older copy or nothing, and the reviewer has to check every line. A verified folder for each workflow, with a named owner who replaces outdated files, gives the AI one place to read from and the reviewer one place to check against.
Nobody owns the result
One person can build a custom GPT, Copilot agent or Claude project in a few hours, and it often stops working without anyone noticing when that person moves on. A short register of shared workflows, each with an owner and a backup, keeps them maintained. The governance guide covers how that register fits into a firm AI policy.
No baseline
Without a before number, nobody can show what the pilot saved, and the decision to keep it rests on opinion. Five recent examples of the task, timed and with their rework noted, are enough for a baseline. The same measures taken during the pilot let the firm decide to expand, fix or stop on a set date.
Review takes back the savings
When the reviewer does not know what to check, they check everything, and the time saved on drafting goes into checking. A short review checklist for each workflow fixes that, such as code sections cited, quantities matching the drawings, and client names correct. Testing on past cases with known answers before going live shows the reviewer where the AI tends to be wrong.
Change management
Prosci's Best Practices in Change Management research found that 88% of projects with excellent change management met or exceeded their objectives, against 13% of projects with poor change management, which is about seven times more likely. In an AI rollout, change management means the following roles and steps.
- One point person owns the rollout and reports progress to leadership each month.
- One champion in each department or studio owns that group's workflow and answers questions where the work happens.
- A short survey of 5 to 10 power users records the tools and prompts they already use, before the firm picks anything.
- Each pilot covers one workflow, ends on a set date, and closes with a decision to expand, fix or stop.
- Training shows each role how the workflow helps their own work, with examples from colleagues in the firm.
For the data and permissions side of implementation, see AI implementation for architecture and engineering firms. To place your firm on the path from first tools to connected workflows, see the AI maturity stages for AEC firms.
Sources: BCG, AI at Scale and the 10-20-70 approach. Microsoft and LinkedIn, 2024 Work Trend Index, May 2024. Prosci, Best Practices in Change Management, as reported in Prosci's article on the correlation between change management and project success.
Frequently asked questions
Why do AI pilots fail in architecture and engineering firms?
Most pilots stop for operational reasons rather than technical ones. The pilot starts from a tool demo instead of a written workflow, licenses go out without training, the source documents are scattered, nobody owns the result, nobody measured the hours before, or senior review takes back the time the AI saved.
How do we measure whether an AI pilot worked?
Before the pilot, record the hours and the rework on five recent examples of the same task. During the pilot, record the same two numbers plus whether a senior person would send the output with light edits. Compare the two sets at a fixed decision date and decide to expand, fix or stop.
Who should own AI implementation in a firm?
One named person owns the rollout and reports to leadership, often the design technology or IT lead. Each department or studio also names a champion who owns that group's workflow, and every shared agent or custom GPT has an owner and a backup.
Is buying Copilot or ChatGPT licenses enough to implement AI?
Licenses give staff access. Implementation is the written workflow, the source documents the AI reads, the training on that workflow, the reviewer and the measurement. Without those, use stays with the few people who would have used AI anyway.
Get the next AEC tech analysis in your inbox
Original research on AEC tools, AI adoption, and industry data. Free, irregular cadence, easy unsubscribe.
Need help applying this to your firm?
We advise AEC firms on technology and AI strategy. Start with a free 30-minute AI strategy session: your workflows, the right tools, and quick wins you can act on.
Book a free AI strategy session