Predicting construction delays before they happen
The construction industry loses an estimated 15–20% of project value to delays and cost overruns every year. Most of that loss is not from unpredictable events — it is from signals that were visible weeks before the slip but went unnoticed or unactioned.
Why delays compound faster than anyone expects
Construction schedules are dense dependency networks. A concrete pour delayed by three days does not push the project by three days. It cascades: the rebar team waits, the formwork gets reallocated, the subcontractor misses their mobilisation window, and suddenly you are looking at a three-week setback from a three-day event.
The signals that predict delays
AI delay risk models are trained on a combination of structured and unstructured signals that PMOs traditionally track manually or not at all:
- Schedule variance on critical path activities over the last 7 and 14 days
- Material delivery confirmation rates versus planned delivery dates
- Subcontractor attendance patterns at the site
- RFI response time from architects and engineers
- Weather forecast alignment with weather-sensitive activities
- Photo-based progress deviation from planned drawings
How a risk score changes PMO behaviour
When a risk score runs continuously — updated every few hours from field inputs — PMO conversations change. Instead of reviewing last week's progress report, the Monday morning meeting opens with: 'Three activities have elevated risk scores. Here are the contributing factors and recommended actions.'
“We used to spend the first 20 minutes of every site meeting figuring out what had slipped. Now we spend those 20 minutes deciding what to do about the risks we already know about.”
— Project Director, Infrastructure Group
What 14 days of early warning is worth
On a 200-crore project, catching a 3-week slip 14 days earlier typically saves 4–8% of project cost in acceleration, rework, and penalty avoidance. That is 80–160 lakh on a single project. Across a portfolio of 10 concurrent projects, the AI scoring layer pays for itself in the first month.
- 01Identify the 15–20 activities on your critical path that carry the most dependency risk.
- 02Configure daily field updates for those activities — even a 2-minute photo and progress percentage is enough.
- 03Set risk thresholds that trigger escalation, not just alerts.
- 04Review the score in your weekly PMO meeting before any other agenda item.
