Service Stream verifies field work with computer vision

Service Stream is using computer vision to verify completion and safety of field work, Simon Fisher, the company's head of data and AI, told the AWS Summit Sydney, according to ITNews. Fisher said the organisation receives "over 1 million images" per month and that its models have reached "94-95 percent accuracy on average" while delivering an estimated "50 percent time saving" compared with human review. The article reports that verification traditionally involves analysts comparing before-and-after images to maps, diagrams and written work orders.

Simon Fisher described the task as object- and scene-recognition at scale: locating "pipes, conduits, ladders and concrete" across many images and reasoning about their arrangement. Editorial analysis: companies applying computer vision to field operations typically combine object detection, instance segmentation and temporal or multi-image reasoning to move beyond single before/after comparisons. Achieving high precision on safety- and payment-linked checks often requires iterative dataset curation and workflow redesign to reduce false positives and surface edge cases for human review.

verifying field work at the scale Fisher cites creates a clear operational bottleneck where automation can materially affect cashflow for contractors and subcontractors. For practitioners, the reported accuracy and time-savings numbers are notable because they imply usable models in production settings, but they also imply ongoing labeling and review effort to reach those metrics.

observers should watch for public details on error modes, how the system handles ambiguous cases, integration with invoicing and approval workflows, and whether third-party audits or pilot results are published. ITNews did not publish additional technical artefacts or external validation; Fisher framed the gains as internal performance figures at the summit.