Why this study exists
The number everyone cites does not exist.
Write anything about productivity in this industry and you hit the same wall. You want to say what share of a licensed professional's week goes to design versus administration. You go looking for the study. There isn't one.
What exists instead:
- Utilization rate — the share of paid hours that are billable. The industry median is 58.9% across roughly 900 firms (Deltek Clarity, 47th edition). It is a real and useful number, but it tells you only that 41.1% of hours were not billed. It does not tell you what those hours were spent doing, and it says nothing at all about how the billable 58.9% divides between design, coordination, documentation and rework.
- The $177 billion figure — that construction professionals lose 35% of their time to non-productive activity. Widely quoted at design professionals. The surveyed population was 49% general contractors and 36% specialty trades. It is not a study of designers.
- Vendor case studies — company-reported, unaudited, and selected for the result.
The gap, stated plainly
There is no published, methodologically transparent measurement of how architects and engineers allocate their working hours. The profession has been arguing about AI's impact on productivity without a baseline to measure impact against.
This study is designed to produce that baseline — and to make it citable by anyone, including people who disagree with us.
How we are running it, and why that matters
This page is published before data collection, deliberately. The instrument, the sampling plan, the analysis plan and the disclosure of who is paying for it are all fixed in advance and in public. That makes it much harder for us to quietly discard inconvenient findings later — which is the point, given that we have an obvious commercial interest in the answer.
Declared conflict of interest
Oxvelo is building an AI-augmented engineering practice. We benefit commercially if this study finds a large administrative burden, and we are disadvantaged if it finds a small one. That is precisely why the instrument is pre-published, why the raw anonymised dataset will be released alongside the findings, and why we invite anyone to re-analyse it. If the study finds that design professionals spend most of their week on design, we will publish that.
The instrument
Estimated completion time: 8–11 minutes. All questions optional except screening. No respondent is identified in any published output.
Section A — Screening and context
A1 · Single choice
Which best describes your role?
Licensed engineer (PE) · Licensed architect (RA) · Engineer-in-training / unlicensed technical staff · Designer / drafter · Project manager (non-licensed) · Principal / owner · Other technical · Non-technical — screen out
Screens to design-professional staff. Non-technical respondents are excluded rather than diluting the sample.
A2 · Single choice
What is your primary discipline?
Civil · Structural · Mechanical · Electrical · Plumbing / fire protection · Geotechnical · Environmental · Transportation · Architecture · Interior design · Landscape · Multi-discipline · Other
A3 · Single choice
How many people work at your firm?
1 · 2–4 · 5–9 · 10–19 · 20–49 · 50–99 · 100–499 · 500+
Matched to AIA and Census firm-size bands so results can be compared against published distributions.
A4 · Single choice
Roughly what share of your firm's revenue is fixed-fee or lump-sum, as opposed to hourly, time-and-materials or cost-plus?
Don't know · 0–20% · 21–40% · 41–60% · 61–80% · 81–100%
Directly tests whether fee structure correlates with reported administrative burden and with AI adoption. We expect it to. If it does not, a central claim of our own published work is wrong.
A5 · Single choice
Years of professional experience
Under 3 · 3–7 · 8–15 · 16–25 · Over 25
Section B — The core measurement
B1 · Constant-sum, must total 100
Think about your last typical full working week. Allocate 100 points across these activities in proportion to the time you spent on each.
Technical design, analysis and calculation
Drawing and model production
Checking, QA/QC and reviewing others' work
Code and standards research
Coordination with other disciplines or consultants
Client communication and meetings
Internal meetings and management
Proposals, fee estimates and business development
Contract administration — RFIs, submittals, change orders
Document control, filing and searching for information
Timesheets, invoicing and administrative tasks
Training, mentoring and professional development
Rework caused by errors, omissions or changed information
Other
Drawing and model production
Checking, QA/QC and reviewing others' work
Code and standards research
Coordination with other disciplines or consultants
Client communication and meetings
Internal meetings and management
Proposals, fee estimates and business development
Contract administration — RFIs, submittals, change orders
Document control, filing and searching for information
Timesheets, invoicing and administrative tasks
Training, mentoring and professional development
Rework caused by errors, omissions or changed information
Other
Constant-sum rather than hour estimates, because people estimate proportions far more reliably than absolute hours. "Last typical week" rather than "an average week" reduces idealisation. This single question is the study.
B2 · Single choice
How typical was that week?
Very typical · Somewhat typical · Unusual — heavier than normal · Unusual — lighter than normal
Lets us weight or exclude atypical weeks, and detect whether the sample skews toward crunch periods.
B3 · Numeric
Approximately how many hours did you work that week?
Free numeric entry
Converts the proportions in B1 into hours, and independently measures unpaid overtime — a cost invisible in utilization statistics.
B4 · Single choice
Of the time you spent on rework, what was the most common cause?
Client changed their mind · Incomplete or late information from others · Our own error · Coordination clash discovered late · Code or reviewer comment · Scope creep · Not applicable
Section C — Where the friction is
C1 · Rank top three
Which activities feel most disproportionate to their value — that is, they consume more of your time than the value they add to the project?
Same list as B1
Separates volume of time from perceived waste. An activity can be large and valuable, or small and infuriating. Both matter, and conflating them is how vendors produce misleading pain-point charts.
C2 · Single choice per row
For each activity, how repeatable is it? Could a well-trained new hire follow a documented procedure to do it?
Almost entirely repeatable · Mostly repeatable with judgment calls · Roughly half and half · Mostly judgment · Almost entirely judgment
This is the automation-potential axis of Figure 4 in our published report, measured rather than asserted. It is also the question that most directly tests whether our editorial judgment was right.
C3 · Free text, optional
If you could delete one recurring task from your week without consequence, what would it be?
Open response
Produces quotable verbatim material and catches categories our fixed list missed.
Section D — Tools and AI, asked last and deliberately
D1 · Single choice
Do you personally use AI tools in your professional work?
Daily · Weekly · Occasionally · Tried and stopped · Never · Not permitted at my firm
D2 · Multiple choice, shown only if D1 ≠ never
For which activities?
Same list as B1
D3 · Single choice
Has a client ever asked you about, restricted, or required disclosure of AI use?
Yes — required disclosure · Yes — restricted or prohibited it · Yes — asked but no requirement · No · Don't know
No published survey has ever asked this. Every AEC AI survey we could find — RICS, RIBA, Chaos, Arup, Deltek, AIA — surveys practitioners about their own adoption and asks nothing about the client relationship. This question alone may be the most novel data in the study.
D4 · Single choice
In your experience, have clients begun to expect lower fees or faster delivery because they believe AI reduces the effort involved?
Yes, frequently · Yes, occasionally · No · Don't know
Tests a real commercial risk. Two-thirds of professional-services providers report clients becoming "more demanding while less willing to pay" on AI grounds, but no A/E-specific evidence exists.
D5 · Single choice
If a licensed professional reviewed and sealed AI-produced work, would you consider that acceptable practice?
Yes, with adequate review · Yes, but only for certain work types · Unsure · No · No, and I think it should be prohibited
The profession's own view of the central legal question. Sealing work prepared by others is already lawful and routine; whether practitioners extend that to AI-produced work is unmeasured.
Section E — Optional identification
E1–E3 · Optional
Email for the results · Willingness to be interviewed · Willingness to be quoted by name
All optional and stored separately from responses
Responses are analysed and published anonymously regardless. Contact details are held apart from the response record and never joined in any published output.
Sampling and methodology
| Parameter | Commitment |
|---|---|
| Target sample | Minimum 150 completed responses for publication; 400+ for discipline-level breakdowns |
| Population | Practising design professionals at US and Canadian architecture and engineering firms |
| Recruitment | Professional networks, discipline associations, LinkedIn, and direct outreach to firms. This is a convenience sample, not a probability sample, and every published output will say so |
| Known bias | Self-selection toward the digitally engaged, and toward people with strong views about AI. We will report the D1 adoption distribution against Deltek's firm-level figures so readers can judge the skew for themselves |
| Field period | Opens on release of this instrument; closes at 400 responses or 90 days, whichever is first |
| Question order | Time allocation (B) is asked before any AI question (D), so that thinking about AI cannot prime the core measurement. This is why Section D is last |
| Analysis plan | Medians and interquartile ranges, not means — the distribution is expected to be skewed. Cuts by discipline, firm size, seniority and fee mix. Pre-specified test: administrative burden correlates with firm size |
| Data release | Anonymised row-level dataset published as CSV alongside the findings, under a permissive licence |
| Corrections | Any error found post-publication is corrected in place with a dated changelog, matching the verification-log practice in our other reports |
What would make us wrong
Stated in advance, so it cannot be rationalised afterwards:
- If the constant-sum allocation shows design professionals spending more than 60% of their week on technical design and production, the "administrative burden" thesis is substantially overstated and we will say so.
- If C2 shows practitioners rating most of their non-design work as mostly judgment rather than repeatable, the automation-potential framing in Figure 4 of our report is wrong.
- If D4 shows widespread client fee pressure already, the commercial model in our report needs rebuilding, not defending.
Take part
The study is open to any practising architect, engineer, or technical staff member at a design firm in the US or Canada. Responses are anonymous, the instrument takes under twelve minutes, and every participant receives the full dataset and findings before publication.
To participate, or to circulate the study within your firm or association, contact research@oxvelo.com.
This instrument accompanies The AI-Augmented Engineering Firm, which sets out why the missing baseline matters and what currently stands in for it.