Data, dashboards and analytics

Every digital tool on a project produces data: model versions, RFIs, submittals, daily logs, inspections, timesheets, deliveries, costs, sensor readings and photos. Most of it is never analyzed. Used well, these data can show a project team early that a job is drifting, help executives compare projects fairly and help a firm learn what actually drives cost, schedule and safety performance. Used badly, they produce cluttered dashboards that nobody trusts. This lesson covers the foundations: data sources and quality, key performance indicators, earned value and safety measures, dashboard design and the basics of analysis, with an honest look at what data can and cannot tell a construction manager.

Where construction data come from

Construction data come from many systems, each designed for its own purpose. Accounting holds commitments, costs and payments. Scheduling holds activities, dates and progress. The field platform holds RFIs, submittals, logs, inspections and safety records (Lesson 7). The CDE holds models and documents (Lesson 4). Time and attendance systems hold hours by worker and cost code. Equipment telematics report engine hours, fuel and location (CON-250). Sensors report concrete temperature, humidity, vibration or noise. Reality capture provides progress measurements (Lesson 8).

The difficulty is that these systems rarely share identifiers. A cost code in accounting, an activity ID in the schedule, a location in the field platform and a zone in the model may all describe the same work in different ways. Connecting them requires a common data structure: standard cost codes, a location breakdown that is the same in the model, schedule and field platform, and consistent project and vendor identifiers. Without it, analysis is mostly manual reconciliation.

Data quality

Analysis is only as good as the data. Common data-quality problems in construction include missing values (daily logs not completed), inconsistent coding (the same work charged to different cost codes on different jobs), late entry (costs posted weeks after they are incurred), duplicates, and free-text fields that cannot be analyzed. Improving data quality is mostly about process: making the right data easy to enter at the source, validating entries automatically, and showing people that the data are used.

Key performance indicators

A key performance indicator (KPI) is a measure chosen because it reflects progress toward an important goal. Good KPIs are few, clearly defined, measured the same way across projects, linked to decisions and balanced between leading indicators (which predict outcomes, such as RFI turnaround or near-miss reporting) and lagging indicators (which record outcomes, such as cost overrun or recordable injuries).

Percent plan complete (PPC), from the Last Planner System, is a useful leading indicator: the share of weekly work commitments that were completed as promised. Low PPC means the plan is unreliable, which tends to precede schedule slippage.

Earned value

Earned value management (EVM) combines cost and schedule into a consistent set of measures. It compares three quantities at a status date: planned value (PV), the budgeted cost of work scheduled to date; earned value (EV), the budgeted cost of work actually completed; and actual cost (AC), what the completed work actually cost. From these come variances and indices.

An SPI below 1 means less work has been done than planned; a CPI below 1 means the work done has cost more than budgeted. The estimate at completion (EAC) shown here assumes that the cost performance to date will continue; other formulas exist for different assumptions. Because a 5D model linked to the schedule (Lesson 6) produces planned value automatically, and reality capture or field progress produces percent complete, EVM is easier to run in a digital workflow than it used to be.

# Earned value for four work packages at a status date (all figures in $)
packages = {
    #  name          budget (BAC)  planned %  earned %  actual cost
    "Structure":     (3_600_000,     1.00,      0.95,    3_610_000),
    "Envelope":      (2_400_000,     0.60,      0.45,    1_150_000),
    "MEP rough-in":  (3_000_000,     0.50,      0.52,    1_480_000),
    "Interiors":     (2_800_000,     0.10,      0.05,      160_000),
}
PV = EV = AC = BAC = 0
for name, (bac, pl, ea, ac) in packages.items():
    pv, ev = bac * pl, bac * ea
    print(f"{name:13s} SPI {ev / pv:4.2f}  CPI {ev / ac:4.2f}")
    PV += pv; EV += ev; AC += ac; BAC += bac
spi, cpi = EV / PV, EV / AC
print(f"Project: PV ${PV:,.0f}  EV ${EV:,.0f}  AC ${AC:,.0f}")
print(f"SPI {spi:.3f}  CPI {cpi:.3f}  EAC ${BAC / cpi:,.0f}  (BAC ${BAC:,.0f})")

Structure     SPI 0.95  CPI 0.95
Envelope      SPI 0.75  CPI 0.94
MEP rough-in  SPI 1.04  CPI 1.05
Interiors     SPI 0.50  CPI 0.88
Project: PV $6,820,000  EV $6,200,000  AC $6,400,000
SPI 0.909  CPI 0.969  EAC $12,180,645  (BAC $11,800,000)

The output shows the project slightly behind schedule and slightly over cost overall, with the envelope as the main problem on both counts: it is well behind plan, and the work done has cost more than budgeted. The MEP rough-in is ahead and roughly on budget. A dashboard would highlight the envelope for the project manager's attention; the conversation would then turn to why (crew size, material delays, rework, weather) and what to do.

Safety measures

Safety performance is usually reported as rates per hours worked so projects of different sizes can be compared. OSHA's total recordable incident rate (TRIR) multiplies the number of recordable injuries and illnesses by 200,000 (roughly 100 full-time workers for a year) and divides by hours worked; the DART rate does the same for cases involving days away, restricted work or job transfer. Owners also look at the experience modification rate (EMR) used by workers' compensation insurers. A rate is only comparable when the definition is the same.

Designing dashboards people use

A dashboard is a visual summary of the measures a particular audience needs to make decisions. The most common failure is trying to show everything to everyone. A superintendent needs this week's constraints, inspections and safety observations for the site; a project manager needs cost, schedule, RFIs, submittals and change; an executive needs a portfolio view of margin, risk and safety across projects. Each needs a different dashboard.

• Start from decisions: what will the viewer do differently based on each measure?
• Show few measures, each with a target or threshold and a trend, not a single number without context.
• Use simple chart types: bars for comparisons, lines for trends, tables for lists of exceptions.
• Highlight exceptions (projects or items outside thresholds) rather than making viewers hunt.
• State the data source and refresh time so viewers know how current the information is.
• Do not rely on color alone; use labels and symbols so the dashboard works for color-blind viewers and in print.

A dashboard should be reviewed in a regular meeting, or it will be ignored. Many contractors build a monthly project review around a standard dashboard, with the project manager explaining any measure outside its threshold and the actions planned.

Analysis beyond dashboards

Dashboards describe what is happening. Analysis tries to explain why and what might happen next. Simple techniques go a long way: comparing projects on normalized measures (cost per square meter, hours per unit), looking at trends over time, breaking a measure down by trade, location or cause, and checking correlations between leading and lagging indicators across many projects.

Correlation is not causation. A firm might find that projects with more RFIs have larger cost overruns; that does not mean RFIs cause overruns. Both may result from incomplete design. Analysis should prompt questions and investigation, not automatic conclusions. Lesson 10 returns to these cautions in the context of AI predictions.

Data governance

As data become more valuable, firms need rules for who owns which data, who can see them, how they are defined and how long they are kept. A data dictionary defines each measure (for example, exactly what counts as an RFI answered on time). A data owner is responsible for the quality of each data source. Access rules protect sensitive information such as wages, health and safety records and commercial terms. Retention rules follow legal requirements for records (often several years after completion, longer for some documents) and the firm's own needs. Lesson 11 covers privacy and security in more detail.