Interview, Podcast
a16z Podcast | Construction Under Tech -- The Build
- Construction productivity has remained flat since 1995 while manufacturing productivity has doubled, largely due to construction's inability to utilize real-time feedback loops.
- Currently, 98% of construction projects exceeding $1 billion run an average of 80% over budget, with rework costs accounting for approximately 20% of total project expenses.
- A primary source of inefficiency is the "fracture" between design plans and field execution, where rich pre-construction data sets (like 3D models) are often "dumbed down" into thousands of static pages of documents for field workers.
- Traditional contracting models (design-bid-build) foster low trust between engineers and subcontractors, leading to excessive detail in plans that are often unbuildable and frequent Requests for Information (RFIs).
- Integrated project delivery and design-build contracts increase trust by allowing subcontractors to contribute constructible details earlier in the design phase.
- The current industry lacks formal feedback loops; a study of 19 global projects found no systematic process for closing the loop between planned and actual performance regarding schedule, cost, or quality.
- Doxel utilizes autonomous robots equipped with deep learning computer vision and LiDAR (accurate to 2mm) to scan construction sites daily, extracting granular progress and quality data.
- This technology replaces manual, episodic reporting with automated, continuous data collection that updates Gantt charts in real time, allowing managers to see "actual vs. plan" daily rather than weekly.
- DPR Construction, a partner in this technology, has accelerated its pace to the point of requiring daily reports, driven by a strategy of off-site prefabrication (Lego-like kits) and virtual design and construction.
- Errors as small as two inches in MEPF (mechanical, electrical, plumbing, fire) installation can create massive ripple effects, causing weeks of delays or requiring the demolition of set concrete.
- Historical data on component installation times allows companies to refine future estimates and schedules, moving from "belief-based" predictions to data-driven projections.
- Doxel's AI uses deep learning to generalize across different construction sites and sensor types, overcoming the limitations of older, custom-programmed support vector machines (SVM) that required months of re-engineering for new objects.
- The technology facilitates a cultural shift from blame to collaboration, allowing teams to identify and fix issues (like missing boxes) before they impede subsequent trades, reducing the need for rework.
- Objective progress data enables fairer payment terms for subcontractors, mitigating cash flow issues caused by owners withholding payment due to fear of hidden defects or schedule delays.
- In healthcare projects, reducing "time to market" is critical to prevent obsolescence of high-cost equipment (e.g., MRI machines) before facility turnover.
- Regulatory inspectors currently require site access and exposed work for verification; future adoption of this data could allow for remote auditing and trust in a "system of record," reducing inspection-related downtime.
- The integration of construction performance data back into the design phase could enable "design by constraint," where architects see real-time cost and lead-time data (e.g., 6-week lead time for specific doors) while modeling.
- A pilot test applying this data loop to steel structures reduced the design-estimate feedback cycle from 8 weeks to hours and decreased structural costs by 13% while cutting build time by 20%.
- If the industry achieves a 20% reduction in rework and efficiency gains across the board, construction productivity could improve by 40%, theoretically freeing up 4% of GDP in major economies.