We’re putting $5.5M behind research and development of smaller, specialized AI models for real-world deployment — Read our manifesto →
    All posts

    [Case Study] Construction Progress Without Guesswork: NCC's AI Progress-Tracking Deployment

    How daily visual capture and model-based schedule comparison reduced manual reporting by 70% and improved on-time task performance by 2.3 times.

    Conscious Engines

    Industry: Construction
    Organization: NCC
    Use case: Automated site progress capture and schedule comparison
    Evidence basis: Buildots customer case using NCC-reported results
    Disclosure: This is an independent analysis by Conscious Engines. The source is published by the product vendor, and the results have not been independently audited in the public account.

    1. Outcome at a Glance

    NCC piloted AI progress tracking on an 8,000-square-meter project in Helsinki, then expanded use across 68,500 square meters. The company reported 70% less manual reporting work and a 2.3-times improvement in tasks completed on time.

    Key Outcomes

    8,000 square meters

    Initial pilot

    Helsinki project reported in the cited case.

    68,500 square meters

    Expanded portfolio

    Reported deployment scope.

    70% reduction

    Manual reporting effort

    Company-reported estimate.

    2.3 times improvement

    Tasks completed on time

    Relative operational result.

    MeasureReported resultEvidence note
    Initial pilot8,000 square metersHelsinki project
    Expanded portfolio68,500 square metersReported deployment scope
    Manual reporting effort70% reductionCompany-reported estimate
    Tasks completed on time2.3 times improvementRelative operational result
    Dispute impactTens of thousands avoidedCompany description, currency and method not fully published

    The project team also reported fewer arguments over status because visual records and model-based comparisons created a shared source of evidence. One NCC leader said teams spent less time in meetings and disputes, and more time fixing issues.

    The figures are useful directional benchmarks. “2.3 times more tasks on time” needs a baseline rate and project-mix context before it can be compared across contractors. The dispute-saving claim is less precise and should not be converted into a formal ROI without supporting records.

    2. The Operational Problem

    Construction schedules are detailed, but actual progress is often recorded through weekly walks, photos, spreadsheets, and subjective status updates. A delay can remain invisible until it blocks another trade. The resulting meeting then focuses on establishing what happened instead of deciding what to do.

    The data problem is spatial and temporal. The enterprise needs to know what was installed, where, when, against which design and schedule activity. Ordinary site photos are useful evidence but difficult to organize consistently. Manual percentage-complete estimates vary by observer.

    The downstream costs include schedule slippage, rework, idle trades, late material decisions, change-order disputes, and management time. The model does not need to replace the project manager. It needs to make site state observable enough for earlier intervention.

    3. What Was Built

    Buildots uses regular 360-degree site capture, computer vision, and model comparison to map observed work against the project plan.

    System at a Glance

    Site capture

    Record current conditions during routine walks.

    Spatial mapping

    Align imagery with locations and model coordinates.

    Computer vision

    Identify installed elements and progress states.

    Schedule mapping

    Connect observed work to planned activities.

    Analytics

    Flag variance, incomplete work, and emerging delay.

    Collaboration

    Give project teams and trades shared visual evidence.

    LayerFunction
    Site captureRecord current conditions during routine walks
    Spatial mappingAlign imagery with locations and model coordinates
    Computer visionIdentify installed elements and progress states
    Schedule mappingConnect observed work to planned activities
    AnalyticsFlag variance, incomplete work, and emerging delay
    CollaborationGive project teams and trades shared visual evidence

    The model's value depends on the project information layer. Design model structure, schedule activity mapping, location breakdown, trade definitions, and change control all affect output quality. Poor project data cannot be repaired by computer vision alone.

    Language models can extend the system by summarizing daily risk, drafting trade-specific actions, retrieving requirements, and turning voice observations into structured issues. Those outputs should remain linked to images, drawings, schedule activities, and responsible owners.

    4. How It Reached Production

    NCC's movement from an 8,000-square-meter pilot to 68,500 square meters reflects an evidence-led expansion.

    Capture frequently and consistently. Daily or near-daily coverage makes change visible and reduces memory-based reporting. Walk routes and capture quality need operating standards.

    Map the schedule carefully. Progress only becomes actionable when visual observations correspond to the correct work package, location, and date.

    Use the data in planning meetings. Dashboards alone do not fix delay. Teams need a routine for reviewing variances, assigning action, and confirming resolution.

    Preserve evidence. Time-stamped imagery can reduce factual disputes, but retention, privacy, subcontractor terms, and access must be governed.

    Measure intervention value. Track reporting hours, schedule variance detected early, tasks completed on time, rework, disputes, and delay days avoided.

    For workforce privacy, projects should define where cameras may capture, how faces are handled, who can access imagery, and how long evidence is kept.

    5. What Construction Leaders Should Take Away

    NCC's case shows that construction AI creates value by making progress observable. The 70% reporting reduction matters, but the 2.3-times improvement in on-time tasks suggests greater value from earlier action.

    A production-ready construction system connects visual progress models with project RAG, drawing and specification extraction, daily voice capture, schedule-risk models, and a constrained project agent. The assistant should answer questions with site images, model locations, schedule activities, and cited documents.

    The primary metric is cost per schedule variance detected and resolved before downstream impact. This connects model performance to project economics rather than image-classification accuracy alone.

    Sources

    • Buildots, NCC partners with Buildots
    • Metrics are vendor and customer reported. Buyers should request baseline definitions, project comparability, and calculation details for dispute savings.