Supply chain automation with software agents

The lean thinking versus IT debate is fast becoming old hat – with the caveat that users shouldn't think of IT either as an alternative to lean and better business processes, or as a panacea. What's starting to happen now, however – beyond every ERP system provider and his dog marketing lean (or demand-driven) suites of add-ons – is that niche IT players, seeing the potential contradiction of lean's preference for simplicity and local decisions versus mostly centralised systems, are harnessing newer technologies and launching interesting ideas. Among the latest is multi-agent computing – the idea of autonomous, software 'agents' providing a virtual organisation, co-operating and negotiating with each other to solve changing, complex and multi-dimensional business problems. Michael Wooldridge, head of computer science at Liverpool University, is one championing that subject, making the point that today's computing isn't up to the job of global businesses, lean or not, because of the sheer proliferation of information and event nodes in semi-connected supply chains, business networks, departments and the range of intelligent systems from PDAs up. RFID triggers There are already a few examples of multi-agent technology in action, notably from developers like Magenta Technology and CA. But, what's particularly interesting, and has the potential to move the 'lean and IT' debate along, results from the happy coincidence of supply chain event management (SCEM) systems – particularly those based on hosted web portals, such as Wesupply's – and automatic track and trace technologies, notably RFID (radio frequency identification). The former provide low cost, pragmatic mechanisms for getting supply chains more collaborative, with, for example, changing demand signals available to all who need them, regardless of IT sophistication (as long as suppliers can at least receive and send emails), although the more automated the better. Then RFID enables detailed visibility of physical goods, materials, components and assemblies moving through supply chains – including internal supply chains – that can generate more instant and far more accurate feedback. The point is that the vision of supply chains – which are inevitably complex, involving a lot of companies, departments, people, material movements, shopfloors, customers and so on – as well-oiled, demand-driven machines, could become a reality. Kurt Kammerer, president and CEO of supply chain intelligence software firm VI Agents, is one that reckons these technologies can deliver – and particularly with lean's goals of simplification and waste removal in mind – by helping supply chains to become responsive to events locally, as they occur. Sounds arcane, but the idea is to use systems to support distributed local management rather than escalating everything to a central decision-making engine that's too far removed from the action and that few understand. At a base level, he prefers to limit the use of MRP, for example, so that production decisions are made more on the shopfloor, supported by whatever is needed – schedulers, simulations, kanbans... Great, but that idea only works if the combination of lean processes, systems, roles and business rules has already been set up, so that local decisions don't run counter to the greater good, doesn't it? Unless, of course, local degrees of freedom only come into play when something isn't going to plan. But that requires visibility of what's changed, what's now required and where everything is – and on today's plants, that's not usually available till too late. Kammerer points out that in the automotive sector, for example, engineering groups happily co-define and co-develop components, with visibility of models and designs provided through 3D CAD visualisers at one end of the spectrum, and PLM (product lifecycle management) systems at the other. But when it comes to production, although supply chain and ERP systems co-ordinate call-offs, line-side delivery and the rest, the monitoring of events and material movements is not so slick – primarily because of the cost and/or manual effort that would be involved in getting near real time. "Delivery of [supply chain] demand-pull is being hampered by lack of communications between entities," says Kammerer. "We're using RFID and other auto ID technologies for this. RFID is a catalyst: it allows you to monitor the flow of mobile assets and goods non-intrusively. You don't need manned reading or scanning points in the warehouse or the production area. Flow of goods and assets is identified by the tags so it's a digital mirror of the flow." OK, but that means a lot of data chasing a lot of events, and additional transactions into the business systems. Except that VI Agents' solution is, first, exception management, second, riding on the back of existing business transactions and electronic documents, and third, using software agents – logically distributed but potentially running on a single server. "If there are no changes against the plan, so no exceptions, then there is no need for an alert," says Kammerer. So nothing changes. "And we can use an ASN or a bill of lading that companies' systems already produce." So, again, nothing changes – as long as everything goes to plan. Then when it doesn't, as determined by RFID data, local intelligent software agents effectively negotiate with one another to determine the best remedial options, given all the real-time 'sensor' information available. "The value then is that the systems present only the information that's needed so that informed decisions, not gut feel, can be taken locally. Multiple work groups can figure out what they have to do, rather than some central system." And he adds that the only prerequisites for this to work are defined responsibilities ("who does what and in which case") and access to the document flow. Use existing systems Kammerer agrees that what his company is doing is essentially building on the ideas of web-based SCEM systems, but insists that it's a significant addition, and that his system doesn't need SCEM to be in place – it works with whatever the shopfloor or supplier are using – Excel spreadsheets, whatever. "The low penetration of SCEM is because of most [software vendors'] overly sophisticated ways of doing things. IT people try to translate supply chain management into a data structure, which adds complexity, loses people and creates work. We take that out. We use agent technology to make sure that the plan is being executed. If material arrives in full when it should, that's a tick in the box. But if it's late, or the wrong quantity, our approach is to have the system notify the relevant groups so that they can determine what action to take." Quite how far anyone will go with agent technology is open to conjecture. VI Agents claims US Department of Defense customers in manufacturing and distribution, but can't name them. Kammerer believes fast moving, volatile industries, like high tech electronics and consumer packaged goods, are most likely to see the benefits. Incidentally, he insists that cost and timescales won't be hurdles: "We market our solutions based on a usage – we don't charge upfront license fees. Typically, customers go live within weeks." Interesting isn't it.