By Dr. Mustafa Çagri Gürbüz, ZLC Professor.
Truthful information sharing between supply chain entities should lead to improved performance for the whole supply chain as a result of better capacity planning decisions, less changes to orders (e.g., order cancellations), and more efficient allocation of total inventory across downstream supply chain members. However, this is easier said than done.
A company may hold centralised inventory to meet demand from downstream supply chain members (e.g., retailers). If demand uncertainty is high, lead times are long, profit margins are low, products become obsolete quickly, in industries such as fashion, consumer electronics, semiconductor manufacturing or automotive, this approach should be beneficial. But it only works if inventory can be efficiently allocated across downstream partners, especially when there is a mismatch between supply and demand (surplus or shortage). There are several possible allocation strategies that work well if real demand is known to all supply chain partners. But if the downstream partners are not reporting demand truthfully, the choice of allocation mechanism becomes critical.
There may be technical barriers to gathering and sharing demand information, and often critical but local information on customers, accounts, orders, order cancellations, and trends is not readily available to central planners. But misreporting demand (through orders placed on the Central Distribution Center, aka CDC) may as well be a conscious strategy. Policies to improve overall supply chain performance may not look like a win for individual supply chain partners (even when they are under the same ownership and theoretically on the same side). Cost/benefit allocation rules may not be well-aligned with individual interests.
Even when incentives are aligned, supply chain members may not understand allocation mechanisms, be able to predict the strategic behaviour of other members, or be able to work out which behaviours are optimal as they attempt to ‘game’ the system – deliberately ordering in excess of the true demand in the expectation that their order will be cut back in the case of a systemwide inventory shortage.
Which is just what happens. But the CDC can’t know by how much an order has been inflated, and therefore whether a partial delivery will meet the real demand, or create a genuine shortage. Central planners have investigated if schemes such as turn-and-earn, bonus payments, punishment mechanisms, and forecast-accuracy based inventory allocation rules can be used to align incentives, encourage true demand signaling and thus securer more efficient inventory allocation. (There is also the converse situation where inventory is in surplus. It may be necessary to follow a “push” strategy allocating all the inventory downward to the retail locations, perhaps because stock can’t be retained at a central location, (for example some online platforms with no storage capabilities), or because downstream players can handle excess inventory better locally through markdowns or local disposition options).
Together with colleagues from Minneapolis and Tilburg, I have been exploring these issues, motivated by the inventory allocation challenges that a large Western European dairy company faces. The firm supplies own-brand and retailer-specific fresh dairy products, and currently uses the ‘proportional allocation’ rule (see below).
Specifically, we have tried to understand whether some allocation mechanisms encourage retail managers to be truthful about local demand, or alternatively whether they are encouraged to misreport demand, with effects on profit both overall and at individual retailer level. We have looked at both shortage and surplus scenarios (inflated and deflated order quantities).
We considered three common allocation rules: (1) proportional rule where each retailer receives a percentage of his/her order, and this percentage is that particular retailer’s order size divided by the total orders from all retailers, (2) linear rule where each retailer order is reduced/increased by the same amount when total orders exceed (fall below) available inventory, and (3) uniform rule where no retailer receives more (less) than ordered when there is a shortage (surplus) and the difference between the allocation quantities is reduced as much as possible. The table below shows how these rules work for a simple situation of two retailers ordering against a total inventory of 100 units. Note that in lines one and three there is an apparent shortage of stock; there is a surplus in instance two.
| Allocation quantities | |||||
| Retail Orders | Proportional | Linear | Uniform | ||
| (80,60) | (57,43) | (60,40) | (50,50) | ||
| (40,30) | (57,43) | (55,45) | (50,50) | ||
| (80,40) | (67,33) | (70,30) | (60,40) | ||
Our research (see “Ordering Behavior and the Impact of Allocation Mechanisms in an Integrated Distribution System”, POM, Vol. 31 (2), 2022, by Spiliotopoulou, Donohue, and Gurbuz) suggests that the uniform allocation mechanism incentivises the retailers to order their real known demand. The proportional and linear allocation rules, on the other hand, encourage retailers to ‘strategically’ inflate or deflate their orders. Under both the proportional and linear rule, each retailer can hope to beneficially influence his/her allocation, and profit, by ordering above (below) his/her true demand when there is systemwide shortage (surplus). But that must necessarily have a negative impact somewhere else.
But under the uniform rule, over-ordering is likely to result in the retailer receiving goods they cannot sell even though there is an overall shortage, while under-ordering risks losing sales through stock-out despite a surplus in the system. Thus the best strategy should be to report true demand and not try to game the system.
We tested the theory in laboratory experiments at the Network-Institute Tech Labs, VU University Amsterdam, and CentERlab, Tilburg University, putting subjects through a full range of scenarios across all three allocation rules. Our experiments suggest that choice of allocation mechanism does indeed have a major effect on both the likelihood and magnitude of retailers’ order deviations from true demand. Under the uniform rule retailers ordered their true demand 45% of the time – which doesn’t sound great until it is realised that under both proportional and linear rules, less than 10% of orders were ‘truthful’. Deviations from the truth were also significantly smaller under the uniform rule. Together, this means that using the uniform rule yields a significantly better allocation efficiency, and thus realised profits (as a percentage of the best that was theoretically available).
We were also able to make some interesting observations on what drives retailer (mis)behavior:
- Retailers have to inflate/deflate orders the most under the linear programming rule to get the same desired response from the central distribution centre.
- Misreporting (deviating from true demand) hurts overall supply chain profits as expected. Although retailers believe they are gaining strategic advantage by misreporting under proportional and linear allocation rules, our experimental results suggest that they do not actually benefit from this behaviour and their own profits are actually reduced.
- Retailers do not either always or never order their true demand as theory would suggest. They manipulate their orders to some extent in all allocation mechanisms, even when truthful information sharing would be optimal, if they think there is a risk of inventory shortage or surplus.
- The magnitude of order deviation increases with time (retailers learn that other retailers are behaving strategically and start inflating or deflating more). However, once ‘everyone is doing it’, this does not seem to have a major negative effect on allocation efficiency.
- Not all retailers appear able to incorporate observed demand into their beliefs about systemwide inventory surplus and shortage and so find the ‘right’ level of order manipulation. Retailers may be reducing their own profits by trying to be strategic, but not being able to do it properly.
- There seems to be a strong recency effect, in the sense that retailers adapt their ordering strategy based on past (recent) order rationing allocation experiences, anchoring more on the difference between quantity ordered and quantity received in the most recent period instead of the quantity received versus observed demand.
- Most participants seem to either tell the truth or follow a “centered” strategy manipulating their orders similarly in both directions. The percentage of retailers who ‘systematically’ inflate or deflate is rather low.
Evidently, if firms such as our dairy company move to the uniform allocation rule they may be able to solicit true demand information from retailers/customers more frequently and thus improve allocation. Good news is that this may be possible without additional investments (e.g., in acquiring sell-out data) and the uniform rule can readily be implemented.
Exploring how retailers behave in a real-life setting as opposed to a lab environment in multi-period interactions could provide additional insights to understand the behavioral causes of order manipulation as well as the effect of learning. Also, the choice of the right allocation mechanism may not only depend on allocation efficiency or profits. Other concerns such as service level guarantees, contractual agreements, foregone profits, fairness ideals might also lead to one allocation rule being preferred over another even though it leads to reduced profits on average. Investigating which mechanism will be the most preferred under such conditions/concerns would help understand when each allocation rule will be more practical and likely to be deployed by companies.
For more information, contact Dr. Mustafa Çagri Gürbüz at [email protected]
