Field Operations · Client-facing

Field Service Vision QA

Technicians photograph completed installs; the model checks against the install standard and flags defects before the crew leaves site.

PilotCritical
JS
Back to portfolio

Business owner

Helen Okafor

Executive sponsor

Dana Kirsch, VP Field Operations

Start date

Apr 13, 2026

Target rollout

Feb 26, 2027

Completion

28%

ROI

Finance-reviewed, annualized where noted

Hours saved / mo

120

Hours saved / yr

1,440

Cost avoided

$21K

Margin impact

$67K

Confidence level: Low — based on baseline cycle-time study and realized utilization to date.

Health & Risk

Critical — accuracy below the field-safety bar; pilot limited to 2 depots.

Blockers

  • Low-light performance
  • Field device camera fragmentation

Risk flags

  • Worker imagery captured incidentally in background of photos

Mitigation notes

On-device blur for background people shipped Aug 8. Low-light training set collection runs through October; decision gate Nov 15.

Activity feed

Synced from connected delivery systems

  1. S

    #field-ai: depot 12 reported false defect calls at dusk

    3 days ago

  2. J

    FLD-58 open — low-light dataset collection

    5 days ago

PISMO Output QA

Last validation run Aug 12, 2026 · 15:35 UTC · 88 samples

84.2%

Reviewer notes · Arun Patel

Low-light and rain conditions drive most failures. Defect recall falls to 71% on outdoor evening jobs.

Governance posture

NIST AI RMF-aligned control checks

2.1 / 5
  • Risk AssessmentFlagged

    Safety-adjacent use case, elevated tier.

  • Bias MitigationNeeds Review

    Lighting/skin-tone robustness study incomplete.

  • Data PrivacyFlagged

    Incidental worker imagery — DPIA in progress.

  • TransparencyNeeds Review

    Technician-facing disclosure being rewritten.

  • SecurityPassed

    Images encrypted in transit and at rest.

  • Human OversightPassed

    Supervisor confirms every defect call.