Site progress × Research note
Where site productivity leaks: the fifty-year puzzle and the weekly signals
Construction is the one major industry whose measured productivity has gone backwards. Is that real or a measurement artefact? The research is clearer than the headlines, and it points at process and regulation as much as technology.
- 1%
- a year: construction labour-productivity growth over two decades, against 2.8% for the world economy[1]
- $1.6tn
- of extra value a year if construction productivity caught up with the wider economy[1]
- 3.6%
- annual productivity growth in manufacturing over the same two decades, against about 1% in construction[1]
Key takeaways
- McKinsey Global Institute found construction labour productivity grew about 1% a year for two decades, against 2.8% for the world economy and 3.6% for manufacturing.
- In one of the most detailed national studies, value added per construction worker was roughly 40% lower in 2020 than in 1970, and physical measures of housebuilding productivity are flat or falling.
- It is not just a measurement error: Federal Reserve economists found the likely bias is too small to change the conclusion.
- Productivity fell most where building is hardest: dense urban cores, tight supply constraints and long permit times.
- The gap shows up on projects as waiting, rework and re-keyed information, which is why process and data fixes pay before new technology does.
01The headline numbers
In 2017 the McKinsey Global Institute published Reinventing construction: a route to higher productivity. Its headline finding was that global construction labour-productivity growth had averaged about 1% a year over the previous two decades, against 2.8% for the total world economy and 3.6% for manufacturing. If construction caught up with the wider economy, it estimated, the sector’s value added could rise by about $1.6 trillion a year[1].
Figure
Annual labour-productivity growth, about 1995–2015
Detailed national data can be starker still. Austan Goolsbee and Chad Syverson, in a 2023 NBER paper titled The Strange and Awful Path of Productivity in the U.S. Construction Sector, found that value added per full-time construction employee was about 40% lower in 2020 than in 1970. Had it instead grown at a modest 1% a year, labour productivity across the whole economy would have been about 10% higher[2].
02Is it real, or a measurement problem?
This is the right question to ask. Productivity is output divided by input, and construction output is hard to measure: every project is different, and turning spending into “real” output needs a price index that separates inflation from better buildings. If the price index rises too fast, measured productivity falls even when crews are working as well as ever.
The Bureau of Labor Statistics took this seriously. BLS economists led by Leo Sveikauskas developed new, better quality-adjusted productivity measures for four construction industries (single-family and multifamily housing, highways and bridges, and industrial construction), noting that reliable output deflators are the core difficulty[4]. BLS now publishes these measures[5].
The most direct test came from Federal Reserve economists Daniel Garcia and Raven Molloy. In their data, construction is the only major industry to have recorded negative average productivity growth since 1987, and they asked how much of that could be explained by unmeasured improvements in building quality. Their answer: even under generous assumptions, the bias is not large enough to overturn the conclusion that construction productivity growth has been weak[3].
Even under generous assumptions, measurement bias is not large enough to overturn the conclusion.
03What the data point to
If the decline is real, the next question is why. The research offers several pieces of evidence:
- Physical output per worker is flat. Goolsbee and Syverson look past price indices to physical measures in housing, such as homes or square feet built per worker, and find productivity falling or at best stagnant over decades[2].
- Materials are used less efficiently. The same paper finds a decline in how efficiently firms turn materials into output[2].
- Productive firms don’t grow. In most industries, more productive producers win market share. Across regions, construction shows no sign of this: regions with more productive construction sectors do not gain share of national activity[2].
- Constraints matter. Garcia and Molloy find productivity fell most in areas with more building in the urban core and tighter supply constraints, especially where permit times are long[3].
The short-run numbers show how sensitive the measure is to demand. BLS reports that single-family housebuilding productivity rose 12.4% a year from 2019 to 2021 as output grew much faster than hours worked, then declined in 2022 and 2023 as output fell and hours held steady[5][6]. Productivity in construction moves with how smoothly work flows, not only with how hard people work.
04Where the gap shows up on a project
National statistics are abstract; site productivity is not. The same McKinsey report grouped its recommendations into seven areas: regulation, contracts, design and engineering, procurement and supply chain, on-site execution, digital technology and materials, and workforce skills[1]. Three years later, McKinsey described construction as one of the least digitised industries of all[7].
In ERP implementation work, the losses that show up most clearly in the data are rarely in the craft work itself. They are in the gaps between steps:
- waiting for an approval, a drawing revision, a material delivery or a permit;
- re-keying the same information into site reports, spreadsheets and the ERP;
- rework caused by building from the wrong information;
- decisions made late because the cost or progress report arrives after the fact.
Each of these is a process and data problem before it is a technology problem, which is why mapping the process comes before choosing a tool.
05Case study: Site Progress & Schedule Control
From my portfolio · project 05 of 15
Synthetic data- Problem
- Slippage hides in weekly progress updates, and submittals sit unapproved while site work continues.
- Decision supported
- Which activities and approvals to chase this week.
- Data
- 50 activities, 750 weekly progress records and 300 submittals.
- Method
- Explainable rules: actual ÷ planned progress below 0.85 for two consecutive weeks; submittals open over 21 days.
- Result
- 31 activities flagged (100% recall, 54.8% precision against labels); 25 stuck submittals.
- Limits
- Historical flags on synthetic data; the ratio is not an earned-value SPI.
06The signals a project can watch
National productivity is the sum of thousands of projects losing time in the same few places. My Site Progress & Schedule Control project turns two of them into weekly worklists[8]:
- Sustained slippage. An activity is flagged when actual progress is below 85% of planned progress for two or more consecutive weeks. One bad week is noise; two in a row is a pattern.
- Stuck approvals. A submittal is flagged when it has no approval date after more than 21 days open. Waiting for approvals is exactly the kind of lost time the productivity research points to.
| Check | Result | What it means |
|---|---|---|
| Activities with sustained slippage | 31 of 50 | Historical flags, anywhere in the activity’s record |
| Agreement with labelled delays | 100% recall, 54.8% precision | Catches every labelled delay, with 14 extra flags |
| Submittals stuck over 21 days | 25 of 300 | Approvals holding up work |
The 54.8% precision is worth dwelling on. A simple rule that never misses a real delay will also raise some false alarms, and the team has to decide whether that trade is worth it. For weekly site meetings it usually is: a false alarm costs a five-minute conversation, while a missed slip costs a milestone.
07How strong is the evidence?
Not every finding in this note rests on the same kind of evidence. This is how I would weigh each one before acting on it.
| Finding | Evidence | Strength | Main caveat |
|---|---|---|---|
| Construction productivity has grown far slower than the economy | MGI global analysis; national accounts | Strong | Output is hard to measure in construction |
| The decline is not just a measurement artefact | Federal Reserve test of deflator bias | Strong | Relies on assumptions about quality change |
| Physical output per worker in housing is flat or falling | NBER study of physical measures | Moderate | Housing only |
| Long permit times go with the biggest declines | Federal Reserve metro-area estimates | Moderate | Association, not proof of cause |
08What a project team can act on
- Measure flow, not just effort. Track waiting time for approvals, RFIs, submittals and deliveries alongside labour hours.
- Enter information once. Map where the same data is typed in more than once and connect those steps.
- Attack rework at design stage, where most of it is decided.
- Plan approvals as a risk, with probabilities and float, not a fixed date.
- Standardise and repeat. Repeatable details, packages and processes let a team learn from one project to the next.
NotesSources
- McKinsey Global Institute (2017). Reinventing construction: a route to higher productivity.
- Goolsbee, A. and Syverson, C. (2023). The strange and awful path of productivity in the U.S. construction sector. NBER Working Paper 30845.
- Garcia, D. and Molloy, R. (2023, revised 2025). Reexamining lackluster productivity growth in construction. Finance and Economics Discussion Series 2023-052, Federal Reserve Board.
- Sveikauskas, L., Rowe, S., Mildenberger, J., Price, J. and Young, A. (2016). Productivity growth in construction. Journal of Construction Engineering and Management, 142(10).
- U.S. Bureau of Labor Statistics. Construction labor productivity (productivity highlights).
- U.S. Bureau of Labor Statistics (2022). Labor productivity rose in single- and multi-family residential construction during pandemic. The Economics Daily.
- McKinsey & Company (2020). The next normal in construction.
- Iwale, A. (2026). Site Progress & Schedule Control. GitHub.
Figures are quoted from the sources above as published; where a source reports a range or a survey estimate, it is described that way. Results from my own projects say whether they use real public data or synthetic data.