Skip to content
English
Contact us
  • There are no suggestions because the search field is empty.

PRD: Ops Vision

1. Executive Summary

One-sentence positioning

An automated image-auditing pipeline that uses computer vision to identify delivery risks, such as packages left in the rain or unsafe locations, and flags them for dispatcher review.

Problem We’re Solving

  • The Core Problem: Shippers and dispatchers lack the resources to manually audit thousands of proof-of-delivery (POD) photos, leading to undetected compliance issues, weather-related damage, and theft risks.

  • Who experiences it: Dispatchers, safety/compliance officers, and fleet owners.

  • Why it matters now: Rising insurance claims and customer expectations for delivery quality require a proactive, rather than reactive, approach to delivery de-risking.

Proposed Solution (High Level)

  • What we are building: An asynchronous computer vision pipeline that analyzes driver-uploaded photos for specific "risk indicators" (e.g., wet pavement, curb-side placement) using Azure AI Vision 4.0.

  • What fundamentally changes: Auditing moves from a manual, needle-in-a-haystack search to an exception-based workflow where only high-risk deliveries are reviewed.

Primary Value Delivered

  • Risk reduction: Prevent claims by detecting packages left in high-risk zones (rain, public streets).

  • Operational efficiency: Automate the auditing process for 90%+ of standard deliveries.

  • Strategic differentiation: Positions our platform as a "quality-first" logistics provider that leverages the latest technologies in AI/ML.

Who This Is For

  • Buyer / Decision Maker: Fleet Owners, Operations VPs.

  • Primary Users: Dispatchers and Audit/Compliance Teams.

  • Secondary / Indirect Users: Drivers (via quality feedback) and Shippers (end-to-end transparency).

2. Strategic Context & Alignment

2.1 Why Now

Advancements in Large Multimodal Models (LMMs) and Azure’s Image Analysis 4.0 allow for "dense captions" and "read" features that can now understand complex scene contexts (e.g., "package on a wet porch") with high confidence. This allows us to advance our AI-first, autonomous TMS vision, via APIs and without significant AI/ML resource investment.

2.2 Strategic Alignment

  • Marketplace Strategy: This serves as a premium "upsell" add-on for high-value cargo shippers (i.e. Medical and Pharma).

  • Platform Vision: Moves our application from a simple tracking tool to an intelligent enforcement engine.

2.3 Non-Goals (Explicit)

What this project is intentionally not doing

  • Real-time in-app coaching for drivers during the photo-taking process (this is handled by our “Edge Computer Vision” with Captur.ai).

  • Automated claim filing or driver pay/settlement adjustments.

3. Problem Statement & Opportunity

3.1 Current State

Today, photos are simply stored as static blobs. Unless a customer reports a missing or damaged package, these images are rarely viewed. If a package is left in the rain, it may be hours before the dispatcher or customer realizes the error.

3.2 Impact

  • Business Impact: High claim costs and potential churn of premium shippers.

  • User Impact: Dispatchers are overwhelmed by data they cannot process.

3.3 Opportunity

By using the "Dense Captions" and "Tags" features of Azure AI Vision, we can programmatically "see" risks. For example, detecting "wet" or "puddle" keywords in a scene description can trigger an immediate alert to a dispatcher to contact the customer or driver.

4. User Experience: End-to-End Journey

4.1 Feature Discovery

Where users encounter this

  • Where: Visual Dispatch Board: Map (driver and stop cards) and Grids (shipments table, POD column). Shipments Form (Attachments section). In-app Marketplace under "AI & Compliance Tools."

  • Contextual Prompt / Upsell Motions: Suggested to users in the "Enforcements and Actions" section via “Computer Vision” Enforcement/Label (disabled). Suggested to users on the Shipments form in Attachments section (info text)

What users see

  • Value proposition

  • Who it’s for

  • Any prerequisites or limitations

4.2 Activation & Access Control

  • Permissions: Dispatchers and Drivers can view. Only Admins can activate the feature in the Marketplace.

  • How: Enabled via Marketplace card. Once active, users can apply the “AI Vision Audit" label to specific Customers, Drivers, Orders, etc. via “Labels” (same as other ‘Actions’ today).

  • Key Decisions: Users define the "Risk Threshold" (e.g., "Only flag photos with <70% confidence of residential context") and parameters (investigating prompt-based strategy).

4.3 Post-Activation Experience (Steady State)

  • Dispatcher View: The Dispatch Board shows flagged deliveries visually on the Map and in the Shipments and/or Driver panels.

  • Notifications Center: The notifications center (top nav) will increment a badge counter when an image is flagged for manual review. Clicking on this notification open the relevant Order and open the Attachments section.

  • Comparison UI: Side-by-side view showing the original photo and an "Analyzed View" with bounding boxes identifying objects (e.g., package, curb, rain/wetness) and a plain-language explanation of the risk.

  • User Confidence Signals: AI-generated "Scene Context" descriptions (e.g., "a package near a public road") help the dispatcher make 5-second decisions.

5. Scope & Phasing

Phase 1 — MVP / Initial Release

  • Weather Detection: Flagging "wet," "rain," or "snow" context.

  • Placement Detection: Flagging packages left near "roads," "curbs," or "sidewalks".

  • Label OCR: Validating that the address on the box matches the order destination using OCR.

  • Safety Confirmation: Automating passes for packages detected on "porches," "mats," or "garages".

  • Architecture: The initial version will be deployed as a stand-alone microservice, accessible by both the CXT and eCourier platforms.

Phase 2 — Future Enhancements

  • Damage Detection: Training custom models to detect crushed corners or torn cardboard.

  • Prompt-Based Parameters: Users can specify what they want analyzed in images by providing custom instructions in the Enforcement Action configuration.

5.1 Historical POD Processing (Scope Clarification)

(update 25 January 2026)

Default Behavior

  • By default, Vision Ops analyzes Proof-of-Delivery (POD) images uploaded after the feature is activated and audit scoping is applied.

Optional Historical Processing

Vision Ops may optionally support processing of historical POD images that were uploaded prior to feature activation.

Processing of historical PODs:

  • is disabled by default,

  • requires explicit enablement as part of the Vision Ops module configuration at customer activation time,

  • applies only to PODs that fall within the defined audit scope.

Rationale

Historical POD processing may incur additional compute cost and operational load. As such, it is treated as an explicit configuration decision rather than a runtime user option.

Non-Goals

This phase does not require:

  • real-time progress visibility for historical processing,

  • user-initiated reprocessing of individual PODs,

  • retroactive changes to shipment lifecycle state.

6. Functional Requirements (High Level)

  • Asynchronous Processing: Images uploaded to Azure Storage must trigger a background function for analysis.

  • Rating Engine: Must assign a "Risk Score" based on Azure AI Vision findings.

  • Gating Logic: The system must only run analysis if the specific "Audit Label" is present on the entity (Customer/Driver/Order).

7. System Architecture & Technical Considerations

7.1 Systems Involved

  • Driver App: Image source.

  • Azure Blob Storage: Trigger for the pipeline.

  • Azure AI Vision API: Analysis engine (Dense Captions, Tags, OCR).

  • Operations: User Interface and interaction.

7.2 Data Ownership & Flow

Driver uploads image --> Azure Blob Trigger --> AI Vision Analysis --> Results stored in Internal DB

Flagged items appear in Dispatcher UI.

7.3 Key Technical Decisions

7.4 Known Technical Risks

  • TBD

8. Data Model & Schema Strategy

Standardized Fields

  • Core attributes required

Extensibility

  • What remains configurable

  • How customers retain flexibility

Explicit Exclusions

  • Sensitive or regulated data we do not store

9. Monetization, Packaging & Enablement

9.1 Pricing & Packaging

  • Add-on vs tier-gated

  • Trial behavior

  • Usage considerations

9.2 Sales & Support Enablement

  • How Sales positions this

  • What Support can / cannot override

  • Documentation requirements

10. Success Metrics & Guardrails

Primary Success Metrics

  • Adoption

  • Revenue or risk reduction

  • Time-to-value

Operational Guardrails

  • Failure rates

  • Support impact

  • User friction indicators

11. Risks, Tradeoffs & Open Questions

Open Product Decisions

  • Behavior toggles

  • UX tradeoffs

  • Enforcement vs flexibility

Business Risks

  • Customer pushback

  • Sales friction

  • Over-constraint of workflows

Technical Risks

  • Migration complexity

  • Scale or latency issues

12. Delivery Plan & Epics

Primary Epic(s)

  • Jira links

Dependencies

  • Other teams

  • External partners

  • Platform work

Target Milestones

  • MVP: (Final Mile Forum)

  • Beta: Q3 2026

  • GA: T+30 from Beta

13. Appendix & Links

PRD Quality Bar (Checklist)

Before marking this PRD “Ready”:

  • Clear user journey from discovery → mature state
  • Explicit non-goals documented
  • System boundaries defined
  • Activation and permissions clearly stated
  • Success metrics agreed upon