Client Profile.
The client operates in the utilities and telecommunications infrastructure sector and keeps georeferenced records of its field assets, including utility poles. It uses those records for network planning, maintenance scheduling, compliance reporting, and capital investment decisions. The annotated dataset also feeds machine learning models, GIS platforms, and asset management systems.
Business Need.
The client had a large inventory of utility pole imagery to annotate. Each record needed the pole type, condition indicators, attached infrastructure, and geolocation attributes, so the client could join raw field data with its asset management and predictive maintenance platforms. Phase 2 covered 9,976 pole records across 94 node locations, assigned in batches.
The work required a primary annotation pass and a secondary review. The client needed a delivery partner that could stand up a trained, supervised workforce quickly, with QC checkpoints and reporting at the node and annotator level.
Phase 2 also tested repeatability. The client wanted proof that HitechDigital could hold quality at this volume and show the process maturity to scale as the workflow grows.
Challenges.
- Handling 9,976 records across 94 nodes, with single contributor loads reaching 4,338 records, so completion risk needed tracking from day one.
- Keeping nine distributed contributor groups on the same definitions and calibration, because inconsistent labels would force rework across the remaining volume.
- Getting every group productive fast: several groups started slowly, so onboarding, access, and task-complexity blockers had to be found and cleared quickly.
- Getting verified records into the client’s GIS and asset systems before the full batch finished.
- Meeting data governance standards suited to a regulated utilities environment.
Solution.
HitechDigital ran a role-based annotation model across nine contributor groups, with one training manual, two-stage QA, and folder-level progress tracking. The first 1,160 verified records reached the client on a rolling basis, so its GIS and asset teams could begin using them before the full batch finished.
Approach.
- Split delivery into phases that matched the client’s data ingestion schedule, and released the 9,976 records in controlled batches starting April 19, 2025. That set the release pace and tied QA cycles to batch milestones instead of single records.
- Assigned work through a role-based model with tiered workloads. Nine contributor groups ran in parallel, and each group’s load matched its demonstrated capacity and validated domain expertise.
- Trained every group on one manual (V2-Final) covering pole classification definitions, acceptance criteria, and workflow procedures.
- Classified each pole and labeled its attributes: type, condition indicators, attached infrastructure, and geolocation.
- Ran an independent verification review on each record and logged it separately in the status tracking system. Only records that cleared both stages counted as Completed.
- Tracked progress by folder for every group, spotted throughput gaps early, and moved capacity to where it was needed.
- Released verified records on a rolling basis, so the client could load them into downstream systems without waiting for the full batch.
Technology Used:.
Centralized status tracking system with folder-level, node-level, and annotator-level reporting.
