Information Technology Reference
In-Depth Information
to take into account such as the quantity of components to be ordered, the date by
which they are required, the periodicity in which they are requested, the supplier who
provides them, the prices offered, the storage cost to be paid for them once they have
been received as well as the possible delivery delays that may occur among other things.
In essence, in dealing with the suppliers the agent acts as a reverse auctioneer in multi-
attribute auctions. Issues like sole sourcing and multiple sourcing are important in this
respect as well as the ability of an agent to switch from one supplier to another if needed.
The customer problem.
The success of an agent does not only depend on the supplier
strategy adopted, but also on dealing with the customers and selling manufactured prod-
ucts. An agent competes with all other agents in the game to secure customer orders.
The components acquired will have been bought at different prices throughout the game
so a particular PC may cost a different price say on day 80, than on day 145. On top of
that, the agents have to take into consideration the cost that they have to pay while the
raw components and the finished products lie in storage. This side of the SCM game
resembles competing against other bidders in an auction. Although some information
on aggregate price statistics is revealed during the game, this is rather limited. For a
more detailed description see [3].
3
Related Work
The TAC SCM game has been running since 2003 with minor modifications every year
to improve its efficacy. One of the major problems in the game is obtaining components
to start production. In particular, the way that the suppliers worked in 2003, i.e. the
big discount rates given for orders early in the game, gives all agents the incentive to
request large quantities of components in the beginning. This problem has been coined
the “0-day effect”. Most of the agents in the 2003 competition had a day-0 procure-
ment strategy, i.e. they ordered large quantities in the beginning of the game. This was
also helped by the fact that there was no storage cost for keeping massive numbers of
components in the inventory.
RedAgent [11], the winner of TAC SCM'03, based its functionality on a multi-agent
design, in which it used simple heuristic agents for procuring components from the
suppliers. The main idea was to use internal markets to provide price estimates for the
extra components that it needed to purchase. TacTex [12] used the day-0 strategy and
sent extra RFQs during the game based on a prediction of the future inventory according
to the current usage of components. HarTAC [7] procured components with the day-0
strategy and tried to maintain a reasonable quantity of all components in stock at all
the times by ordering small quantities of components through the game; Botticelli [4]
and PSUTAC [13] only relied on the day-0 strategy without sending extra RFQs during
the game. DeepMaize [8] used a preemptive strategy to block agents that used day-0
strategy. The preemptive strategy worked by submitting a big RFQ to each supplier
for each type of component. This RFQ had the effect of preempting subsequent RFQs
because the supplier would have committed its production capacity for the rest of the
game. There were few other agents using conservative strategies, such as PackaTac [6],
who played a low-risk strategy maintaining a low level inventory of components. It
decided to use this strategy to counteract the day-0 strategies that emerged during the
