AI strategy for AEC firms: how to get past the pilot
Most construction and engineering practices are past the question of whether to try AI. The pilot happened. Somebody showed a specification review that took twenty minutes instead of two days, the room was impressed, and eleven months later the same three people are still the only ones doing it.
That gap is the actual subject of an AI strategy, and it is an organisational design problem rather than a technology one. A pilot proves a task is possible. Capability means the firm can do it without the person who proved it.
Why an AI strategy built on pilots stalls in AEC firms
Pilots are selected for the wrong property. They get chosen because they demonstrate well, which usually means a self-contained task with a clean input and a visible output. Specification summarisation demos beautifully. It also touches almost nothing else in the business, which is precisely why nothing else changes when it succeeds.
Three structural reasons the stall repeats.
The pilot ran on the enthusiast's tacit knowledge. The person who built it knew which prompt patterns worked, which documents to exclude and what wrong output looked like. None of that was written down, because it lived in their head and the pilot did not need it to be anywhere else.
Nothing downstream was changed to receive it. If the review is now faster but the QA gate, the fee structure and the programme all still assume the old duration, the time saved has nowhere to go. It gets absorbed, and absorbed time is invisible.
No second person was ever obliged to do it. Adoption that depends on volunteering plateaus at the number of volunteers, which in a busy Irish practice carrying broad individual remits is a small number.
What an AI strategy for an AEC firm has to contain
Two pages, and none of it is about models.
A capability statement per discipline. What a competent structural engineer, QS or bid coordinator in this firm should be able to do unaided by the end of the year. Written as observable behaviour rather than awareness.
The workflows being redesigned, named. Not 'improve document handling'. The specification review gate, the monthly valuation reconciliation, the RFI response cycle. Three is plenty. Ten is a wish list.
What each redesigned workflow releases, and where it goes. The most commonly skipped element and the one that determines whether anything sticks. Released time either gets reinvested somewhere named or it silently refills.
The verification standard. Which outputs carry a named human signature. Anything touching calculation still has to stand against Irish building regulations and the Eurocodes regardless of what produced the first draft.
The sequencing. Which discipline goes first and why, which depends on where the willing senior practitioners sit rather than where the biggest theoretical saving is.
AI strategy: the capability map for an AEC practice
Capability moves in a recognisable order, and skipping a rung produces the stall described above.
At the first level individuals produce useful output inconsistently, and quality depends entirely on who is at the keyboard. At the second, a team produces consistent output because the prompt patterns, the exclusions and the checks have been written down and shared. At the third, the workflow itself has been redesigned around the new duration, so the old version is no longer available to fall back on. At the fourth, the firm can onboard a new starter into that workflow without the original enthusiast being involved.
Most Irish AEC practices sit somewhere between the first and second level and describe themselves as being at the third. The test is simple and uncomfortable: if the person who built it left in the morning, what survives?
Measuring an AI strategy by capability rather than usage
Licence and usage dashboards are the most misleading instrument available.
Seat counts say nothing. Monthly active users says slightly more and still rewards opening the application. A firm can show ninety per cent activation and have changed no workflow at all.
What is worth measuring instead, and all of it is observable.
Task completion without the expert. Can a second person run the workflow end to end using only the written instructions?
Rework rate on AI-assisted output. If it has gone up, the verification step is wrong rather than the tool.
Cycle time on the named workflow. Not on everything. On the three you chose.
Where the released hours went. Named destination, or the saving is notional.
AI Institute's position on the commitment worth holding any provider to is four hours saved per person per week, measured after the fact rather than promised in the proposal. That measurement is only possible where the workflows were named in advance, which is another reason the strategy comes before the training rather than after it.
Who owns the AI strategy in a construction or engineering firm
The pattern that fails is a steering group. Committees generate principles, and principles do not redesign a valuation cycle.
The arrangement that works in Irish practices, where directors already carry broad remits, is an accountable owner at director level, one practitioner per discipline who has genuinely done the work, and a quarterly review with a short written record. The practitioners matter more than the director, because colleagues copy people rather than policies, and a respected senior engineer using the new method in front of others does more than any internal launch.
Sectoral context is worth keeping an eye on. The Society of Chartered Surveyors Ireland and the national enterprise policy position published through the Department of Enterprise, Tourism and Employment both shape what Irish clients will start expecting to see in submissions.
What an AI strategy will not fix
It will not make a firm with no written method suddenly consistent. If the specification review process was undocumented before, adding AI produces faster inconsistency.
It will not survive a fee model that rewards hours. Where the commercial structure prices time rather than outcome, a genuine time saving reads as lost revenue, and no strategy document overrides that arithmetic. That conversation belongs at board level before the training budget is spent.
It will not substitute for the policy a client will ask to see, which is a separate artefact covered in AI governance in Ireland.
And it will not work as a document alone. The Sisk programme is a useful reference for what scaling looks like when it is treated as an operational change rather than a technology rollout, set out in how Sisk is scaling AI across construction operations.
Where to take an AI strategy next
Pick the three workflows. Write the capability statement for one discipline. Name where the released hours go. That is a morning's work and it is more strategy than most practices currently hold.
For scale, the Ethos Engineering programme ran across 54 senior staff over 18 months at €185,000, which is what a firm treating capability as a multi-year commitment looks like. Most practices start with one discipline and one cohort, which is the right way in.
The foundation layer, the one that makes the second capability level reachable, is covered in core AI skills for AEC teams, and the wider built environment programmes are built around construction and engineering workflows rather than adapted to them.
