How a Network Infrastructure & Solution provider can use Azure AI to improve image quality, automate validation, and accelerate construction daily processing.
Customer Details
Customer:
$ 130M New Jersey HQ - Network infrastructure & Solutions provider for Telecom
Scale:
Estate:
Summary
One of our customer’s construction client project processes approximately 15,000 images per month as part of construction dailies. Personnel currently perform image validation through manual review, creating a recurring operational burden and a dependency on skilled personnel to identify duplicate, non – compliant, or incomplete imagery. The validation workflow also contributes to delays in moving dailies from submission to validation.
An analysis of 3,445 dailies shows that 1,408 (40.8%) contains issues such as missing metadata or duplicate images. All these factors lead to delayed revenue recognition.
The proposed solution uses Azure’s Vision AI capability and cloud – native Azure services to automate duplicate – image detection, metadata extraction, and metadata validation via Optical Character Recognition (OCR). The proposed workflow moves routine checks into an automated layer while retaining human review for exceptions.
Engagement snapshot
- The use case focuses on construction projects and the client – mandated Quality Control
process for construction dailies. - Site foremen submit multiple images as part of dailies documentation, including duplicate
and non – compliant images. - The current operating baseline is approximately 400 images per day and 15,000 images per month.
- Validation is currently 100% manual, with an average review effort of three minutes per
image and an approximate labour rate of $50 per hour (approximate benchmark used for estimation purposes)
Landscape
The current process begins with field teams uploading construction images through the existing
workflow. Images form part of the daily submission and reviewers subsequently assess them against the client’s quality and metadata requirements.
Google Cloud stores a reference image dataset containing previously validated project photos. The relational database (RDB) records validation results and serves as the centralized source for duplicate detection, metadata validation, and reporting.
The proposed architecture integrates with the existing environment rather than replacing the
submission process. The proposed technology stack uses Azure Functions for serverless, event –
driven processing, Azure AI Vision for OCR and image analysis, and Azure Storage for image ingress, interim queues, and raw image storage.
Operating Pressures
- Image redundancy: An estimated 20–30% of the image volume consists of duplicates and non – compliant image.
- Manual validation : Personnel currently review100% of submitted images manually.
- High issue rate : 1,408 of 3,445 dailies in the analysed sample (approximately 40.8%) were flagged with issues.
- Validation delay: Client dailies remain in the Submitted – to – Validated queue for an average of 5.05 days.
- Revenue timing: Approximately 41% of billable dailies were stalled due to validation issues.
- Linear operating effort: At the baseline volume, manual validation represents approximately 750 human hours per month.
(an approximate benchmark used for estimation purposes)
Approach and Design Direction
The business case focuses on automating the repeatable parts of image quality validation while keeping the existing field submission process intact. The workflow comprises four capabilities:
- Duplicate image detection: Automatically identify exact duplicates by comparing embedded image metadata and flags duplicate records in the database.
- Textual metadata extraction and validation: Use Azure AI Vision OCR to extract timestamp, Global Positioning System (GPS) coordinates, address, city, county, and state, then validate the presence of mandatory fields.
- Exception visibility: Categorizes images as Complete, Partial, or Absent based on metadata validation and surfaces discrepancies for review through the dashboard Interface.
- Compliance reporting: Maintains centralized results in a database, including NTP ID, Daily ID, image count, duplicate count, and metadata issues, with the option to integrate the reporting layer with Power BI.
Technical Workflow
The Vision AI architecture uses an event – driven, queue – based processing model to process construction daily images asynchronously. The workflow uses timer – triggered Azure Functions, queues, control tables, Google Cloud Storage, and Azure Vision AI to manage image ingestion, OCR processing, validation, result persistence, and synchronization back to the Relational DB.
1. Identifying and Scheduling New Dailies
- A scheduled job runs at regular intervals to identify construction dailies that are ready for image validation.
- A log of the previous run ensures only new submissions are picked up, so the same work is never processed twice.
- Each daily is registered against its NTP ID and Daily ID and queued for validation, allowing identification and image processing to run independently of each other.
2. Retrieving and Tracking Images
- Images are retrieved directly from the existing cloud storage location, so they do not need to pass through every component of the architecture.
- A tracking record is created for every individual image, giving each one its own status from the moment it enters the workflow.
- Any image that cannot be queued for validation is flagged as failed, so that no submission is lost without a record.
3. Extracting Metadata with Azure Vision A
- Azure AI Vision reads the text embedded in each photograph and returns the metadata the business case depends on timestamp, GPS coordinates, address, city, county, and state.
- The extracted values are matched back to the image and the daily they belong to, ready for validation against the client’s mandatory – field requirements.
4. Recording Results and Job Status
- Each image is marked as completed or in error depending on the outcome of extraction, and the extracted metadata is stored alongside it.
- Daily – level figures – images submitted, images processed, technical errors, and overall status – are updated so that progress is visible at both image and daily level.
- Results are written to the results store in periodic batches rather than one record at a time, which keeps the process efficient at production volumes.
5. Synchronizing Results and Maintaining Auditability
- A separate process identifies dailies whose validation is complete, but whose results have not yet reached the relational database, marks them as in progress, and confirms them as synchronized once the transfer succeeds.
- The relational database holds the validated image – level information and its processing status as the single source of truth for reporting.
- Because control records and run logs are maintained throughout, every image can be traced from daily identification through retrieval, extraction, validation, and result storage to final synchronization.
AI techniques & tools
- Azure AI Vision – Extracts embedded textual metadata from submitted images.
- Image and metadata – based duplicate detection – Identifies duplicate images using the
metadata and GPS matching logic described in the business case - Azure Functions – Provides event – driven processing for image validation workflows.
- Azure Storage – Supports image ingress, interim processing queues, and raw image storage.
- Relational DB – Serves as the centralized reporting and compliance layer for validation results.
Solution Impact & Coverage
The figures below establish the scale of the current manual process.
Measure | Baseline |
Images processed per month | 15,000 |
Images processed per day | ~400 |
Redundancy rate | Estimated 20–30% |
Current validation method | 100% manual review |
Average review time per image | 3 minutes |
Monthly human effort | 750 hours |
Dailies analysed | 3,445 |
Dailies flagged with issues | 1,408 |
Dailies with issues | ~40.8% |
Average time to validate | 5.05 days |
What this means for the business
- The current validation process consumes approximately 750 hours of human effort each month at the stated baseline volume.
- Data – quality issues affect a substantial share of submitted dailies, with 40.8% of the analysed sample flagged with issues.
- Validation delays are connected to the invoicing pipeline, with an average 5.05 – day validation cycle and estimated delayed revenue recognition in the analysed period.
- A centralized validation record in the system can provide a single source of truth for duplicate and metadata issues, supporting operational visibility and compliance monitoring.
- The proposed architecture is cloud – native and can be integrated with the existing submission process, allowing automation to be introduced without changing how site foremen upload images.
Way forward
The business case recommends using the initial Vision AI capability as a foundation for broader
quality and compliance improvements. The following directions are explicitly identified:
- Expand AI validation beyond one client – Following successful stabilization, the capability can be scaled across additional clients and programs to standardize quality controls.
- Shift from detection to prevention – Introduce real – time guidance and guardrails in the system so errors can be prevented at the point of daily submission.
- Institutionalize data – driven compliance – Use centralized results to build Power BI dashboards that support data – driven leadership decisions.
- Leverage validation insights for workforce enablement – Use validation trends to identify training gaps by crew, subcontractor, or region, and support targeted coaching and performance – based incentives.
The longer – term direction is to establish Vision AI as a scalable quality and compliance capability. As construction image volume grows, the proposed approach provides a foundation for improving validation consistency, operational visibility, and revenue recognition without relying solely on manual review.
