Receiving Shipments

Overview #

Shipments are the centerpiece of Trackberry. A shipment tracks a batch of fresh produce from packing house to destination, including all associated documents, pallet data, and quality check results.

Creating a Shipment #

Via Email #

The simplest way to receive a shipment is through email:

1. Your approved senders email packing lists and quality documents to your organization's inbox
2. Trackberry automatically creates a shipment and attaches the documents
3. Data extraction begins immediately
4. You receive a notification when the data is ready for review

Trackberry also stores the email body itself — not just the attachments. If the email contains status updates, arrival notices, or delay information, Trackberry uses AI to classify and summarize it. See the Inbound Email Processing guide for details.

Email threading: When someone replies to an existing shipment email thread, the new documents and messages are automatically attached to the correct shipment. If a forwarder sends an out-of-thread email with the transport reference (B/L or AWB number) in the subject, Trackberry matches it to the right shipment too.

Via Direct Upload #

To manually create a shipment:

1. Navigate to Shipments in your organization menu
2. Click New Shipment
3. Fill in the shipment details (reference, vessel, origin, etc.)
4. Drag and drop your documents into the upload area
5. Click Create Shipment

Supported file formats:

Tip: When possible, ask your suppliers to send packing lists as Excel (.xlsx) files rather than PDFs. Excel files produce the most reliable extraction results because Trackberry reads cell values directly — there's no need for table detection, OCR, or guessing column boundaries. PDFs work well too, but they require an extra step of reconstructing the table structure from the page layout, which can occasionally lead to misaligned data.

Trackberry also recognizes these document types:

• Packing List — Pallet details, produce, calibre, box counts, weights
• Quality Report — Quality inspection results and observations
• Transport Document — Bill of Lading (B/L) or Air Waybill (AWB)
• Phytosanitary Certificate — Plant health inspection certificates
• Warehouse Receipt — Receipt confirming goods received at warehouse

How Document Processing Works #

When a document is uploaded, Trackberry processes it through a multi-stage pipeline designed to be as fast and self-reliant as possible. The system prioritizes deterministic, local parsing that runs entirely on Trackberry's own infrastructure — only reaching out to external AI services when the document format genuinely requires it.

Stage 1: Deterministic Parser #

For PDFs and Excel files, Trackberry first extracts data using its own parser — no external services involved:

1. Table extraction — For digital PDFs, tables and cell boundaries are detected automatically. For Excel files, cells are read directly. The result is a clean, structured table.
2. Column mapping — The parser analyzes column headers to identify which columns contain pallet numbers, produce names, calibre, box counts, weights, etc.
3. Template matching — Trackberry fingerprints the document structure and checks if it matches a previously seen format. If so, it reuses the learned column mapping for instant extraction.
4. Data extraction — Rows are parsed into structured pallet records. Supports both standard layouts (one row per item) and pivot layouts (calibres as column headers).

If all required fields are found, extraction is complete — no external calls, near-instant processing.

Stage 2: Targeted Assistance #

If the deterministic parser extracts most of the data but a few fields remain incomplete, Trackberry makes small, targeted requests to fill in the gaps:

• Individual field extraction — Instead of processing the whole document externally, only a small snippet of context is sent with a focused question for the specific missing field (e.g., "What is the container number?")
• Column disambiguation — When headers are ambiguous, a quick call resolves which column maps to which field
• Uniform field fill — Fields like produce type or box weight that are the same for every row are resolved with a single request

This approach minimizes external dependency while still handling edge cases.

Stage 3: Full Extraction (Last Resort) #

Only when local parsing and targeted assistance both fall short does Trackberry send the full document content for external processing:

• Text-based documents — The entire extracted text is sent with a comprehensive extraction prompt
• Images and scanned PDFs — The file is processed using a vision model that can read directly from the image
• Multiple provider fallback — If one provider is unavailable, the system automatically tries the next

This handles virtually any document format, including ones never seen before.

Learning & Templates #

Every successful extraction teaches the system. Trackberry fingerprints the document's structure (column headers, layout, format) and stores the successful extraction strategy as a parsing template. The next time a document with a matching fingerprint arrives, it uses the deterministic parser with the learned column mapping — no external services needed.

This means:

• First time seeing a format — External processing may be needed to understand the document structure
• Second time onward — The system recognizes the format and parses it deterministically using the learned template

In practice, most organizations work with a limited set of suppliers, each with their own packing list formats. After a few shipments, virtually all incoming documents match a known template and are processed entirely on Trackberry's own infrastructure — faster, more reliable, and independent of external service availability.

Think of it like training: Trackberry learns how to read each supplier's documents, then applies that knowledge automatically from then on. The more documents you process, the more self-sufficient the system becomes.

Reviewing Shipments #

Once processing is complete, you'll see:

• Pallet table — All extracted pallets with their attributes
• Check results — Pass/warning/fail status for each quality check
• Document viewer — Original documents for reference

Approving a Shipment #

If all checks pass and the data looks correct:

1. Review the pallet data
2. Click Approve on the shipment page
3. The shipment status changes to approved
4. Subscribers are notified

Rejecting a Shipment #

If issues are found:

1. Click Reject on the shipment page
2. Add a reason for rejection
3. The sender is notified to correct and resend

Superseding Documents #

If a sender sends an updated version of a document:

1. Navigate to the shipment
2. Find the outdated document
3. Click Supersede and upload the replacement
4. The old document is marked as superseded
5. Data is re-extracted from the new document
6. Checks re-run automatically