Lift the hood on most supply chain applications, and you tend to find the same linear programming-based engines. ILOG, the France-based leader in linear programming, is most often the brand of choice.

As a mature technology, it is worth asking what alternatives are around, and if these present real opportunities for distinction in the market. While still in an early phase in terms of application development, agent technology offers one such alternative, with often faster, more reactive, more realistic planning, decision support, and simulation.

Agent technology is a very granular approach to IT, allowing individual objects (agents) within an application to intelligently seek interaction with other objects within a wider system for their own and the wider system’s benefit. Each object, which has its own needs and constraints, attributes, and preferences, constantly communicates, negotiates, and trades with other agents seeking to maximize its own and the totality’s benefit. Agents have the ability to sense events, reason, plan, and act. They even know when to put their hands up to ask for help.

This may sound laborious, but collaboration between thousands of agents can occur in seconds, making them very alert to change. For example, in a road transport application, an agent might be defined as representing a truck looking for a load and a driver. Another agent could be a load looking for a truck, and yet another can be a driver looking for a truck. The limitations of the truck, size of the load, its delivery timing requirement, and restrictions on and preferences of the driver are all programmed into their respective agents. This allows them to collaboratively see if they can find an ideal outcome.

The proponents of agent programming point to multiple benefits, but most important is the ability to constantly react to a changing environment and adapt accordingly without outside intervention. Consider the following:

Unlike traditional applications, this agent-based system does not need to be stopped and batch processed. The agents constantly interact, and proximate solutions can be found almost immediately while the agents continually look for better ones. New events and new agents are respectively acted upon and added in as they arise.

Knowledge, constraints, and exceptions are easier to build into each of the agents. Traditional linear programs struggle with the increasing array of real-world constraints and local knowledge (think, for example, about the variety of working time laws across Europe for truck drivers). Agents can be programmed individually, and knowledge developed previously can be constantly fed into their reasoning in contrast to more traditional rule-based approaches.

The technology has found success in transport, such as at Air Liquide, and in manufacturing, such as at SCA Packaging, which have used planning and simulation. Or look at Tankers International, a pool of leading oil tanker owners, which is using an ocean transport application based on an agent technology system from Magenta Technology, an Anglo-Russian supply chain software vendor. Tankers is using the application as a decision support tool to schedule and simulate tanker movements. It identified a key benefit in being able to model and mine the wealth and variety of knowledge locked up in the heads of the very large crude carrier (VLCC) scheduling experts. That is passed into the system to not only improve decision making, but also reduce knowledge risk. The way the software is used is equally different from the one big batch run and the plan set in stone. The application acts as a constantly updating marketplace between loads, tankers, and terminals. Schedulers are constantly feeding and refining the system, and are able to make changes up to the last minute.

Every technology goes through a maturity curve, and the engines that drive today’s planning software are no exception. Companies wishing to keep ahead need to scan alternatives, such as those presented by agent technologies, but they also need to consider the negatives:

New tricks mean more complexity-An agent-based approach presents a completely new set of requirements to manage the set of distributed agents across a supply network. Managing the policy, performance, and availability of these agents is something that can add cost and complexity to the environment.

Lack of end-user input-Are customers really ready to toss the decision making on supply chain issues to a set of intelligent agents? We see a difference between the promise of agents to automatically identify and correct issues in real time and how far customers actually take the technology. The things being monitored by the agents still have too much variability, plus agents have a finite set of tasks they can identify and take automated action upon. The amount of automation is directly dependent on how well users can create the policies on automated response.

The unknown-Okay, so this approach might be a better way, but do we know how much better? Is it worth the cost of ripping out how you currently perform optimization to put this in? It’s still early to be able to collate a representative sample of customer scenarios to build consistent and defensible data around true total cost of ownership.

While this could be an interesting alternative approach to doing supply chain planning, interested companies will also need to spend the time to examine the cost and impact of the approach.

Source: AMR Research/Guy Dunkerly, Dennis Gaughan, Nigel Montgomery

The Glossary of Terms in Logistics & Shipping is the most comprehensive paper-based dictionary and therefore the standard for defining terms used in the area of Logistics and Shipping. Use this powerful tool to expand your professional vocabulary and ensure that everyone on your team is speaking the same language. www.theKnowledgeTransfer.com

paperback student version $ 19,99 hardcover executive version $ 29,99

| Permanent Link | Print This eLog |

(Visited over 19274 times)