An Inbound Transportation Management System in Rel

When shipping freight from their vendors to their warehouses, retailers face many challenges related to the timely arrival of the product at its destination: the ability to track freight at a high level of granularity, planning manpower and docks at distribution centers, managing exceptions, and minimizing shipping and logistics costs.

The effective transportation of goods depends on many factors, which are interrelated in many complex ways:

Vendor locations
Available forms of transportation
Various types of costs (for example, shipping and inventory costs)
Transition times
Transportation capacity constraints
Freight availability requirements (at the retailer)

In addition to these factors, retailers need to have visibility into the overall transportation process so that they can identify hotspots and bottlenecks in the network, compute (aggregate) costs per product / time period / network partition, and pin down non-compliant vendors (for example, vendors that tend to send the goods too late or too early) in order to better manage their supply chains.

A solution to a problem of such complexity and size requires three things:

Accurately modeling the transportation process
Computing optimal transportation modes and times for the necessary freight
Providing the ability to summarize and export the results in meaningful ways

This post describes how RelationalAI’s inbound transportation management system, built entirely within the AI coprocessor, tackles these problems in their entirety.

The Problem

Transportation management systems (TMSs) play an important role in supply chains: they ensure the timely delivery of freight and goods, provide visibility into day-to-day transportation operations, and support trade compliance. In more detail, a TMS is responsible for:

Shipping freight between different locations (for example, vendors to warehouses)
Ensuring the timely arrival of products at their destination
Offering visibility into transported tracking freight at a high level of granularity
Enabling planning for staffing and docks at distribution centers
Managing exceptions
Minimizing shipping and logistics costs

We identify two key limitations in current TMSs and explain how we overcome these in our approach:

Most current TMSs employ simple heuristics or partial optimization to model the entire transportation infrastructure and process. Thus, they are bound to miss important costs and other resource-saving opportunities (time, personnel etc.), while having difficulty fully satisfying business rules. We propose a system that is supported by a holistic mathematical model that provides many savings opportunities while ensuring conformance to business rules (for example, freight arrival constraints).
Current TMS solutions are usually provided as standalone software systems, which run separately from the rest of the data infrastructure. A system like this needs cumbersome and error-prone ETL processes to collect, extract, and integrate the necessary data. We provide an ‘embedded’ software solution, which runs within the organization’s data infrastructure and can greatly simplify and facilitate the entire process.

Key to the viability of such a system is the use of a knowledge graph management system (KGMS) that is able to integrate and share data between the customer and several external businesses (i.e., the carriers, cost providers and the vendors). Each of those external businesses maintains different information and can represent it in vastly different ways. Being able to mine the information and represent it in a uniform fashion is best accomplished with the help of a KGMS. Our solution, as part of RelationalAI’s AI coprocessor, a state of the art KGMS with advanced analytics and optimization capabilities, natively satisfies this requirement.

RelationalAI’s Solution

The objective of an inbound TMS is to optimize the shipping of products from vendors to company warehouses (usually called distribution centers or DCs).

Our TMS uses an advanced mathematical optimization engine to solve the problem in its entirety, including selection of shipping days, consolidation of freight, choosing shipping routes and modes of transportation, and deciding on the least expensive approved carriers. All those while meeting user-defined service requirements.

In more detail, our TMS is able to:

Maximize cost savings
Provably support all business rules (we model them in the form of mathematical constraints)
Minimize computation costs so that the model can support large transportation infrastructures
Facilitate data and software integration

We have designed a flexible and extensible software solution that can easily be adapted to proprietary business objectives (for example, fastest deliveries versus cost savings).

The proposed solution is seamlessly integrated with your data as part of the AI coprocessor, providing a unique and novel solution to the TMS problem.

Next, we dive into the details of the proposed TMS, including stakeholders, functionality, and operating procedures.

TMS Stakeholders

Different entities, each with different requirements that need to be satisfied in a unified fashion, will use the TMS on a daily basis. They include:

At the (retail) company site: merchants and planners issuing purchase orders, sales staff managing and tracking products, transportation and logistics staff managing transportation costs and expedites, and company executives reviewing trends and transportation KPIs.
At the company DC sites: staff responsible for managing product arrivals and put-away functions, daily manpower planning, and daily management of loading and unloading docks.
At the vendor sites: staff following TMS-provided instructions for building freight groups (for example, groups of products with the same vendor location, delivery DC and date requirements), confirming freight group availability, confirming carrier arrivals and departures, and handling exceptions.
At the company or third-party consolidation center (CC) sites: staff confirming product arrivals, entering truckload departures to DCs, manpower and dock planning.
At the carrier sites: staff receiving load tender electronically, having the ability to accept/reject loads and updating route and shipment status. The TMS also needs to maintain the necessary information for staff performance evaluation.

The stakeholders and their requirements are summarized in the image below.

High-Level Functionality

A schematic representation of the high level functionality of the TMS is provided below. The (retail) company sends daily purchase orders (POs) to vendors and to the TMS at the line-item level with requested quantities and delivery dates. Vendors interact with the TMS and the company, so that the POs are adjusted if necessary (quantities, delivery window, and shipping location). The TMS computes a delivery window at the company DCs and a shipping window at the vendor shipping location for each freight group. We assume that there is a small permissible lax, say of +/-1 day from the specified delivery dates.

The TMS optimization model is run on business days and determines optimal routes using different modes of transportation (for example, track load - TL, less than track load - LTL, UPS package, etc.) that encompass direct shipping from vendor shipping locations to company DCs and shipping to intermediate consolidation centers (CCs) where freight is aggregated first and then shipped to company DCs at a lower cost.

The TMS selects and tenders freight to the most economical carriers from approved lists of carriers for each lane (a lane is a source/destination on a given date part of the transportation infrastructure). It tracks the freight, computes service levels, and determines compliance by various stakeholders.

Next, we describe the main procedures above that take place in and/or are supported by the TMS software.

Operating Procedures

Receive Purchase Order Line-Items

Purchase orders (POs) and their line-items are sent daily from the company to vendors and to the TMS. Vendors may adjust quantities and delivery schedules and receive company approval. Based on desired company service requirements, for each purchase order line-item, the TMS computes a shipping window from the vendor shipping location and an arrival window at the destination distribution center.

Construct Freight Groups

Purchase order line-items may be very small in number and space and may occupy only a fraction of a pallet. As such, they do not represent the right shipping unit. To overcome this issue, purchase order line-items are consolidated into freight groups that simultaneously satisfy the following four properties:

Must be supplied by the same vendor
Must be shipped from the same vendor location
Must have the same distribution center as their destination
Must have the same required arrival window at the distribution center

The TMS optimizes the transportation of the freight groups and tracks their status. Since each purchase order line-item has a unique parent freight group, our TMS provides unprecedented tracking for purchase order line-items.

Execute Optimization Software

A key foundation of our TMS is our novel multi-commodity flow-based optimization model, which computes consolidation of freight groups, shipping days, mode of transportation, and carrier, minimizing transportation and inventory costs while meeting service requirements.

First, the TMS prepares all the data required to execute the optimization software, including:

Transportation network data: facilities, lanes, lane mileage, transit times, carriers of various modes of transportation
Cost data: unit cost data for truck loads (TLs) and multi-stop TL routes, less than truck-loads (LTLs) Czarlite tables, specialized tariffs (for example, UPS Ground)
Service requirements including shipping and arrival windows
Freight data: freight group weight and pallets, shipping locations, destination distribution center

Second, the optimization software is executed. Only the routes that depart on the next business day are taken into consideration and implemented. Routes leaving on future dates may be revised on the next optimization run as additional freight groups become available at vendor shipping locations.

Tender Routes to Carriers

Selected routes by the optimization are electronically tendered to the most economical carrier on each lane and for the chosen mode of transportation. If a positive response is not received from the carrier in one hour, the optimized route is automatically tendered to the next carrier on the list. Routes are then executed on the next business day.

UIs, APIs, and Reports

The TMS is an operational system that extensively uses UIs and software APIs to communicate with various stakeholders. These include viewing and managing purchase order line-items and freight groups, viewing and approving routes by company transportation staff, tendering and accepting routes, confirming route pickups and deliveries, and entering exceptions.

In addition, the TMS generates management reports that address monthly volumes, transportation costs stratified by modes of transportation and carriers, and the performance of all stakeholders including company staff, vendors, carriers, and consolidation center operators.

Summary

The TMS presented here is an embedded software solution, supported by RelationalAI’s AI coprocessor — a high-performance KGMS with advanced analytics. The AI coprocessor ensures smooth integration with a retail company’s data clouds and simplifies the process of sharing information with essential external systems.

RelationalAI to Host “The Promise of AI and Relational Knowledge Graphs” Webinar

Former Snowflake CEO Bob Muglia to join RelationalAI for a virtual event on September 28, 2023 dedicated to exploring the future of AI and relational knowledge graphs.

RelationalAI Recognized as a One to Watch in Snowflake's Modern Marketing Data Stack Report

RelationalAI enables joint customers to build intelligent solutions on Snowflake's Data Cloud and make better business decisions.