Supply Chain Visibility Software: Architecture & Build Guide

Supply chain visibility software is the engineering layer that ingests asynchronous carrier status messages, vessel and vehicle telemetry, IoT sensor payloads, and warehouse scan events, resolves them against commercial purchase orders, and computes actionable arrival predictions before a delay cascades into a stockout or penalty. On a vendor slide deck, visibility looks like a world map covered in green and red pins. In production, a GPS coordinate is the easiest 5% of the problem. The hard 95% is stitching together a coherent shipment record when identifiers fracture across suppliers, ocean containers, air waybills, cross-docks, and regional carriers that report hours late in incompatible formats.

Most articles ranking for this topic summarize marketing brochures for Real-Time Transportation Visibility Platforms (RTTVPs) without explaining how an ingestion pipeline normalizes a batched EDI 214, reconciles a split ocean booking, or calculates an ETA accounting for driver Hours of Service (HOS) clocks. Having shipped production logistics platforms, including our HaulBreeze supply chain management system and the DispatchMVP auto-transport TMS, we build visibility from the canonical event graph upward. Below is the 2026 architectural blueprint: why map-dot tools fail across modes, how to model a multimodal shipment graph with GS1 EPCIS 2.0, how to sanitize dirty telemetry, how probabilistic ETA and exception loops work, and when to build versus rent an aggregator network.

The short version

• "Dots on a map" fail because identifiers fracture at every handoff: An order starts as an ERP purchase order line, packs onto a pallet with a GS1 SSCC-18 barcode on an ASC X12 856 ASN, loads into an ocean container tracked via DCSA Track and Trace 2.2, and deconsolidates onto domestic trucks tracked by FMCSA ELD GPS pings.
• Canonical shipment graphs require GS1 EPCIS 2.0: Modeling physical movements with GS1 EPCIS 2.0 (ISO/IEC 19987) (ObjectEvent, AggregationEvent, TransactionEvent, and TransformationEvent) preserves parent-child lineage across palletization, container stuffing, transshipment, and cross-dock deconsolidation.
• Data hygiene dominates ingestion engineering: Raw X12 214s and UN/EDIFACT IFTSTA messages arrive out of order, lack UTC offsets, and abuse reason codes; a trustworthy pipeline requires event-time watermarks, spatial geofence verification, and idempotent deduplication.
• Probabilistic ETAs beat static schedules: Combining live position with receiver dwell distributions, port berth queues, and driver ELD duty-clock math produces P50 and P90 arrival windows that prevent ocean demurrage, equipment detention, and OTIF chargebacks.
• Hybrid architecture beats pure build or pure buy in 2026: Rather than rebuilding 100,000 carrier ELD connections from scratch or locking order-to-inventory logic inside a rigid RTTVP, leading shippers and 3PLs consume commodity carrier feeds via API while owning the canonical EPCIS graph, ERP sync, and exception control tower.

Why "dots on a map" fail when shipment identifiers fracture across modes

Supply chain directors rarely complain that a truck's coordinates are off by fifty meters. They complain that a planner cannot answer a basic commercial question: "Purchase Order #48210 has 400 units of SKU-99 due in Chicago on Thursday; where are those exact units right now, and will they clear the receiving dock before Friday's production shift?"

Conventional RTTVP dashboards fail this test because they treat a vehicle movement as the primary database entity. In any supply chain spanning more than a single point-to-point full-truckload (FTL) move, the identifier attached to physical goods mutates at every modal boundary:

1. Commercial Layer (ERP / OMS): Demand originates in your ERP or order management system as PO #48210 containing line items identified by SKU and Global Trade Item Number (GTIN-14), delivery dates, and Incoterms. No carrier, container, or truck ID exists yet.
2. Supplier Fulfillment & Serialization (X12 856 / DESADV): The supplier picks 250 of the 400 units (partial fill), packs 25 cartons across two pallets labeled with 18-digit GS1 Serial Shipping Container Codes (SSCC-18), and sends an ASC X12 856 Advance Ship Notice (or UN/EDIFACT DESADV).
3. Ocean Consolidation (Booking / BOL / Container ID): A forwarder consolidates those two pallets at a Container Freight Station alongside other shippers' cargo into a 40-foot container (MSCU1234567) under a House and Master Bill of Lading (HBL/MBL). The carrier's DCSA Track and Trace API and satellite AIS transponders only know the vessel IMO, voyage, and container ID (MSCU1234567), with zero awareness of PO #48210 or SSCC-18 pallets.
4. Air Cargo Expedites (AWB / IATA ONE Record): The remaining 150 backordered units are expedited by air under a Master and House Air Waybill (MAWB/HAWB) inside an aircraft Unit Load Device (ULD) tracked via IATA ONE Record linked-data APIs.
5. Port Deconsolidation & Domestic Legs (X12 204 / 214 / ELD): After MSCU1234567 discharges at Los Angeles and clears customs, a drayage carrier hauls it to a transload warehouse. The container is stripped: one pallet is tendered via an X12 204 Load Tender to an LTL carrier under PRO #99812 and tracked via X12 214 status messages and FMCSA ELD GPS coordinates, while two urgent cartons go to UPS/FedEx under 1Z numbers for final last-mile delivery.

If your database models a "shipment" as a flat table with one tracking_number column, the chain shatters the moment the supplier splits the purchase order or the transload warehouse strips the container.

Architecting a canonical multimodal shipment graph with GS1 EPCIS 2.0

To maintain unbroken visibility from purchase order line to receiving dock scan, a modern logistics management software platform separates the commercial order hierarchy, the physical handling-unit hierarchy, and the transport movement hierarchy, joining them through an immutable event graph grounded in the GS1 EPCIS 2.0 standard (ISO/IEC 19987):

• What: Objects identified by GS1 keys (GTIN for products, SSCC for pallets/parcels, GRAI for returnable containers, GINC/GSIN for consignments).
• When: Event timestamp (eventTime) paired with eventTimeZoneOffset and separated from ingestion time (recordTime).
• Where: Physical read point (readPoint, e.g., Dock Door 4 or GPS fix) and business location (bizLocation, via GLN or UN/LOCODE).
• Why: Business step (shipping, receiving, loading, customs_clearance, transporting), disposition (in_transit, active, quarantined), and commercial references (bizTransactionList pointing to po, bol, inv, or desadv).

How the four EPCIS 2.0 event types resolve multimodal handoffs

Storing every pack, load, unload, and split as one of four EPCIS 2.0 event types in PostgreSQL ensures the engine always knows which vehicle carries which purchase order units:

1. ObjectEvent (Observation): Records a status milestone or sensor reading against handling units without changing containment, such as an X12 214 AT7*AF pickup scan, a customs release at Rotterdam (NLRTM), or a logger reporting +3.8°C.
2. AggregationEvent (Parent-child containment): Solves multimodal handoffs. When 25 cartons stack onto Pallet SSCC-A, an AggregationEvent (ADD) binds those cartons to parentID: SSCC-A. When SSCC-A loads into Container MSCU1234567, a second ADD event binds SSCC-A to MSCU1234567. At sea, every AIS position or DCSA webhook for MSCU1234567 propagates down to SSCC-A, its 25 cartons, and PO #48210. When the transload warehouse strips the container in Los Angeles, a DELETE event unbinds SSCC-A from the container, and a new ADD event binds SSCC-A to the LTL trailer (PRO #99812).
3. TransactionEvent (Commercial document binding): Associates handling units (SSCC or Container_ID) with commercial documents (PO #48210, ASN #8812, BOL #7741). When a carrier sends an X12 214 referencing only BOL #7741, the graph traverses the TransactionEvent edge to resolve the underlying pallets and purchase order lines.
4. TransformationEvent (Kitting or repacking): Captures irreversible conversions where input lots/SKUs produce new output SKUs (such as a 3PL software kitting station combining component SKUs into a display pack), preserving lot genealogy for FDA FSMA 204 or DSCSA compliance.

Handling the "missing link" problem: Fuzzy entity resolution

When suppliers or regional carriers omit the PO reference from an X12 214 or mistype a BOL number, our two-stage entity resolver prevents orphan shipments:

• Stage 1 (Deterministic key cascade): Attempts exact normalized matches in order: (Carrier_SCAC + PRO_Number), (Master_BOL / Booking_ID), (Container_BIC_ISO6346), (Shipment_Tender_ID from X12 204), and (PO_Number + Origin_GLN).
• Stage 2 (Probabilistic candidate scoring): Evaluates open shipments within a +/- 48-hour pickup window using Jaro-Winkler similarity on facility names, H3 spatial distance between postal codes, and weight/pallet tolerances. Matches above 0.92 auto-link with an audit flag; matches between 0.70 and 0.92 route to an operator review queue.

Multimodal telemetry ingestion and data hygiene

If raw external feeds write directly to your database without quarantine, deduplication, and timestamp normalization, your control tower floods operators with false-positive alerts. Below are the four primary telemetry streams and their required hygiene rules.

1. Motor freight EDI X12 204/214/856 and UN/EDIFACT (IFTMIN / IFTSTA / DESADV)

In North American trucking and rail, the ASC X12 214 Shipment Status Message is the core feed, paired with UN/EDIFACT IFTSTA in Europe. Reliable ingestion solves three flaws:

• Event-time (eventTime) vs Ingestion-time (recordTime) skew: Carriers often batch EDI transmissions every few hours. An X12 214 arriving at 15:30 UTC might report an X6 (En Route) event from 10:15 Local Time, before an ELD geofence arrival processed at 14:00 UTC. Overwriting current_status with the latest message regresses the shipment from Arrived to In Transit. State machines must enforce monotonic progression by normalized UTC eventTime while recording out-of-order messages in the append-only log.
• Ambiguous timezone offsets: X12 AT7 timestamps often omit explicit offsets (LT local time). Our normalizer resolves local timestamps against the IANA timezone polygon of the stop's geocoded facility (America/Chicago, Europe/Berlin), handling DST transitions.
• Carrier status code drift: Because carriers map AT7 status and delay reason codes inconsistently, we maintain a per-SCAC dictionary mapping partner codes into our canonical EPCIS bizStep and disposition taxonomy.

2. Ocean container tracking: DCSA Track & Trace 2.2 and satellite AIS

Major container lines (Maersk, MSC, CMA CGM, Hapag-Lloyd, ONE) expose REST webhooks conforming to the DCSA Track and Trace API 2.2 specification, categorizing updates into TransportEvent (vessel ARRI/DEPA), EquipmentEvent (container GTIN, LOAD, DISC, GTOT, STUF, STRP), and ShipmentEvent (document CONF, ISSU, RELS).

Because carrier webhooks suffer from 6-to-24-hour lags at congested transshipment ports and rolled-booking blind spots, we fuse DCSA webhooks with terrestrial and satellite AIS vessel streams (polled every 5 to 15 minutes). Intersecting live vessel draft, speed, and heading against port-anchorage and berth polygons in PostGIS detects port queue anchoring and berth arrival hours before the carrier emits a DISC webhook.

3. Over-the-road telematics and FMCSA ELD streams

To eliminate EDI latency, visibility platforms ingest live truck telemetry from ELD gateways (Samsara, Motive, Geotab, Platform Science). As covered in our fleet management software guide, US devices comply with the FMCSA ELD technical specification under 49 CFR Part 395. Consuming ELD feeds across contracted carriers requires two guardrails:

1. Load-locked privacy windows: Owner-operators reject 24/7 tracking across loads for other shippers. Our broker accepts GPS pings for a tractor VIN starting 4 hours before scheduled pickup (or upon X12 204 acceptance) and terminates collection the moment the truck exits the receiver's geofence with a POD event.
2. Dead-zone replay and map-matching: Exiting a cellular dead zone, the gateway flushes hundreds of buffered fixes in a burst. The stream processor deduplicates bursts by (device_id, fix_timestamp_utc), filters GPS multipath bounce via Hidden Markov Model (HMM) map-matching (OSRM / Valhalla), and prevents historical replay points from triggering false live geofence alerts.

4. Air cargo IATA ONE Record, EU eFTI, and BLE/cellular cold-chain IoT

For pharmaceuticals, perishables, and electronics, visibility extends to regulatory payloads and environmental telemetry:

• Air Cargo ONE Record: Replacing legacy teletype Cargo-IMP (FSU/RCS, FSU/DEP, FSU/RCF), modern air visibility integrates the IATA ONE Record standard, exposing Piece, Shipment, TransportMovement, and IoTDevice objects over JSON-LD REST APIs.
• European eFTI Regulation: Across EU corridors, the EU eFTI Regulation (Regulation (EU) 2020/1056) mandates standardized machine-to-machine sharing of electronic freight regulatory data (eCMR, ADR dangerous goods) with authorities.
• Cold-Chain IoT Loggers: Cellular trackers (Tive, Sensitech, Controlant) and BLE beacons stream temperature, humidity, light/door-open tamper events, and shock g-forces. Our pipeline computes rolling Mean Kinetic Temperature (MKT) so QA teams know immediately if a tarmac spike breached a biologic batch's stability budget.

Probabilistic ETA prediction and automated exception management

Showing an operator that a shipment is in Memphis is trivia. Telling the operator that the load has an 84% probability of missing its Thursday 14:00 Chicago receiving window by 4.5 hours because the driver has only 75 minutes left on their mandatory FMCSA 11-hour driving clock is actionable engineering. Static ETA formulas (Remaining_Miles / 55_mph) fail because freight transit times are governed by facility dwell distributions and regulatory rest breaks, not highway speed.

How we architect a production multi-modal ETA engine

In our AI software development engagements, we replace naive speed heuristics with a quantile-regression ETA pipeline (LightGBM / CatBoost) outputting P10 (optimistic), P50 (median), and P90 (conservative buffer) arrival windows decomposed into four additive terms:

T_arrival = t_now + Σ[k in segments] T_drive(k) + Σ[s in stops] T_dwell(s) + T_HOS_rest + T_customs_port

1. Truck route-segment travel time (T_drive): Computed over commercial truck-routing graphs accounting for Class 8 speed limiters (62 to 68 mph), mountain grades, urban congestion, and active weather polygons.
2. Facility dwell distributions (T_dwell): Learned from historical ELD geofence entry-to-exit durations at each facility GLN (Facility A drops trailers in 35 minutes median, whereas Facility B averages 3.8 hours of live-unload lumper dwell on Mondays with P90 = 6.2 hours).
3. FMCSA Hours-of-Service simulation (T_HOS_rest): Under 49 CFR Part 395, drivers require a 30-minute break after 8 hours driving, max 11 hours driving in a 14-hour window, and a 10-hour off-duty reset (or EU Regulation 561/2006 tachograph rules). If a truck is 6.5 driving hours from delivery at 13:00 with 3.0 hours left on the driver's ELD clock, our simulator injects the 10-hour rest reset and predicts arrival at 09:30 the next morning instead of a false 19:30 same-day ETA.
4. Port berth, rail ramp, and customs dwell (T_customs_port): Incorporates terminal discharge-to-gate-out velocity, rail grounding times, and customs hold probabilities by HS code and port.

Four high-ROI operational loops that pay back the build

A map with ten thousand green dots is wall art for the lobby. Real supply chain visibility is an exception filter that keeps 96% of shipments silent so your operators can intervene on the 4% that are about to cost you demurrage, spoiled cargo, or an OTIF chargeback.

Reference architecture and 2026 Build vs Buy economics

A mid-market shipper or 3PL tracking 15,000 active shipments ingests 5 million to 25 million telemetry events daily. We engineer the stack across five layers:

1. Multi-protocol ingestion gateway: Adapters for AS2/SFTP EDI drops (X12 204/214/856, EDIFACT IFTSTA/DESADV), DCSA 2.2 ocean webhooks, and MQTT/WebSocket ELD and IoT streams, persisted to immutable S3 storage for replay.
2. Event streaming & normalization (Kafka / Redpanda): Payloads are validated, enriched with IANA timezones and H3 indexes, and published as canonical GS1 EPCIS 2.0 JSON-LD events.
3. Dual-storage persistence (PostgreSQL + TimescaleDB / ClickHouse): PostgreSQL + PostGIS holds commercial entities and EPCIS containment edges; TimescaleDB or ClickHouse stores compressed high-frequency GPS, AIS, and sensor time series.
4. Probabilistic ETA & rules workers (Python / Node.js): Stateless workers evaluate geofence transitions, recompute P50/P90 arrival windows, and trigger exception workflows.
5. Control tower UI & ERP sync: A React console organized around prioritized exception queues, customer portals, and Transactional Outbox connectors syncing SAP, NetSuite, or Dynamics. We proved this unified architecture in HaulBreeze and DispatchMVP.

Commercial RTTVP aggregators vs composable custom control tower

Commercial RTTVPs (Project44, FourKites) spent hundreds of millions building carrier ELD and ocean networks; rebuilding 150,000 carrier ELD connections from scratch makes no sense. Conversely, forcing a commercial RTTVP to act as your custom order-to-inventory graph, 3PL billing engine, and ERP orchestrator is why large IT deployments fail: McKinsey and Oxford found that large IT projects run 45% over budget on average and deliver 56% less value than expected.

What custom visibility software costs to build in 2026

As detailed in our supply chain software cost guide, build budgets depend on team geography and integration scope. Grounded in Eurostat's ICT hourly labor cost index (lc_lci_lev), employer ICT labor costs in Central and Eastern Europe sit at €24.00/hr in Bulgaria, €26.00/hr in Romania, €29.30/hr in Poland, and €38.40/hr in Estonia, versus €48.20/hr across the EU-27 and €58.50/hr in Germany. Against US developer compensation from the U.S. Bureau of Labor Statistics (where US agencies bill $150 to $220/hr), senior CEE teams (see our European IT staff augmentation rates 2026 report) deliver across three brackets:

• Bracket 1: Customer visibility portal & order tracker ($60,000 to $105,000 | 10 to 14 weeks): Unifies existing TMS EDI 214s, parcel webhooks, and ERP order status into a customer self-service timeline.
• Bracket 2: Multimodal EPCIS 2.0 visibility platform ($130,000 to $220,000 | 18 to 26 weeks): Full ingestion for X12 (204, 214, 856), DCSA 2.2 ocean webhooks, AIS, FMCSA ELD streams, and cold-chain loggers, with an EPCIS 2.0 graph, HOS-aware probabilistic ETA engine, and SAP/NetSuite sync.
• Bracket 3: Enterprise supply chain control tower ($230,000 to $380,000+ | 28 to 40 weeks): Adds closed-loop exception orchestration (demurrage prevention, dynamic dock re-slotting, 3-way freight audit, and MILP re-routing) across transportation management, warehouse management, and supply chain planning.

Through our logistics and supply chain, custom software development, and product development practices, we start with a 2-to-4-week Technical Discovery phase that audits your carrier EDI payloads, maps identifier fractures, and delivers a fixed-scope sprint budget.