COOP’s logistic system
The grocery company COOP transports about 3.5 million cubic metresFootnote 6 of grocery items, refrigerated and frozen food and non-food products per year in Sweden.Footnote 7 The company’s warehouses are located close to Stockholm: in Bro for non-refrigerated foods and non-food products, in Västerås for refrigerated food and in Enköping for frozen food. For incoming products from outside Sweden different predominantly intermodal transport chains are used. The port of Helsingborg in the south of Sweden is one of the main import ports. A high proportion of the incoming products are transported to the warehouses via Helsingborg. The products are assembled in the warehouses and go from there to COOP’s over 700 Swedish stores. The distance between Helsingborg and the warehouses in the Stockholm area is about 600 km (see Fig. 1).
Traditionally, COOP bought in all its products and services, including transport. However, in 2009 the company made an effort to ensure control over their incoming and outgoing transports and developed among other changes a combined road-rail system. We explore this in section 3.3.
Since September 2009 two shuttle trains carrying 36 trailers move 5 days a week between Helsingborg and Bro. With the northbound train, suppliers send goods and the warehouse in Bro is the receiver. With the southbound train, the warehouse sends assembled goods and the shops in the south of Sweden are the receivers. The southbound train has stops in Alvesta (about 150 km north of Helsingborg) and Helsingborg. According to information from COOP based on the rail operator’s coding for the period 1 June 2011 to 29 February 2012, the infrastructure holder, the Swedish Transport Administration, was responsible for more than half of the delays (55 %). The share was at 88 % highest in January 2012 and was higher for the northbound train (61 %) than for the southbound train (41 %). The delays were caused by a wide range of different factors – problems related to the electricity system, the signal system, the switches, derailments, and construction works. The rail operator was responsible for 16 % of the delays. COOP, as sender of the goods, caused 8 % of the delays. Twenty-one percent of the late arrivals were due to other reasons, like weather conditions, elk on the track, and copper theft.
The warehouse in Bro has access to the rail network via an industrial track. The southbound train is part of an intermodal transport chain: the trailers are transferred to trucks and transported to the shops in Småland (from the railway station in Alvesta) and Skåne (from the railway station in Helsingborg).
COOP planned to use the shuttle train to also carry highly time-sensitive fresh fruits and vegetables. A trial was started but discontinued after two major breakdowns. Today these transports are performed by truck. Bread and milk products, which represent a high volume percentage of the products sold in COOP’s shops, are transported in separate systems.
Transport time variability data
COOP has kindly provided us with data that describe the performance of the two shuttle trains between Helsingborg and Bro for the period 1 June 2011 to 30 September 2012. Assuming an average of 29 loaded trailers per trainFootnote 8 and around 10 tonnes per trailer, the two trains have transported around 200,000 tonnes or 74,000 cubic metres during the 16-month period.
Between June 2011 and September 2012 both the northbound and southbound trains were cancelled twice due to a heavy storm in the end of November 2011. The average delays were moderate but there were several delays of over and below one hour: the northbound trains arrived in Bro more than 1 h too late on 53 occasions (15 %). The southbound train was more than 1 h too late in Alvesta on 27 occasions (8 %) and more than 1 h late in Helsingborg on 48 occasions (14 %).Footnote 9 Arrival delays of more than an hour lead to major problems for the warehouse in Bro and – even more seriously – the shops in the south of Sweden. Major delays in Helsingborg or Bro also cause problems for the rail operator as the shuttle train cannot turn around as planned. COOP does not experience problems related to too-early trains. Figures 2 and 3 show that many trains arrived up to nearly 2 h before schedule.
Table 1 shows the means and the standard deviations (excluding cancellations) for the stations that the shuttle train serves.
The mean, exclusive of early arrivals, is the conditional expectation given that we have a delay. The standard deviation, excluding early arrivals, is the so-called lower partial standard deviation (LPSD) or the square root of the semi-variance. It uses only negative deviations from the time-table, squares those deviations to obtain a semi-variance and takes the square root to get a standard deviation for the left-tail. The LPSD is therefore the square root of the average squared deviation, conditional on a negative outcome (late arrival). The LPSD does not capture, however, the frequency of delays.
The figures in Table 1 also show that the average delays, including too-early arrivals, are relatively small (9 to 24 min) and the average delays (excluding too-early arrivals) are more substantial (38 to 53 min). Hence, it is necessary to study the real distribution of the delays (see Fig. 4) and the risks for large delays that firms that are senders and receivers of goods, like COOP in our case study, face.
The standard deviation (including early arrivals) for the transport time of COOP’s shuttle train is lower (on average 63.3 min) than the standard deviation for freight trains in Sweden (of 76.5 min, see ). The skewness is −4.65 going south and −4.94 going north (the normal distribution has a skewness of zero) and the kurtosis is 39.55 going south and 39.46 going north (the normal distribution has a kurtosis of 3).
Table 2 shows that the average delay in the 90th percentile was around 2.5 h and that the 10 % worst delays contributed to more than half of the total delays going south and almost two-thirds of total delays going north. In a sense, Table 2 quantifies the risk for COOP of using rail instead of road if road is showing no, or a very low degree of, variability in transportation time.
If we take southbound transports to Helsingborg as an example, in 5 % of all transports the delay is almost 2 h (118.9 min) or more according to the percentile value. That is, once per 20 transports a delay of this size or larger is expected to happen southwards, and since there are two trains per day (one northbound and one southbound), it means that once every 10 days or three times a month a delay of this magnitude occurs. The average delay in each percentile is the conditional tail expectation (CTE). For example, going north to Bro there is a 10 % probability of a delay of at least 1 h. The average within the percentile is close to 2.5 h, so that with a 10 % probability (once every 10 transports or once every 5 days) COOP has to expect an average delay of 2.5 h. Hence, the CTE gives a better estimate of how big the risk is in the tail of the delay distribution.Footnote 10 Whereas the percentile values mark the best outcome of any given worst-case scenario (e.g. 10 % worst cases), the CTE gives the average for a given worst-case scenario.
Costs to manage transport time variability
In 2009 COOP shifted transportations from road to a combination of road and rail.Footnote 11 Ideally such a shift should not increase the overall costs, although rail transport services are in general less reliable than road services in Sweden. Below we try to estimate COOP’s precautionary costs that are measures taken in advance in order to mitigate the consequences of major delays and the operational costs that were paid as consequence of the major train delays and cancelled trains during the period June 2011–September 2012. According to our reasoning in section 2, to reach an optimum it will not generally be necessary to insure against all delay costs. The calculations below are based on information from COOP and own assumptions.
When it comes to precautionary costs, COOP did not have extra safety stock costs since the shelves in the shops, which serve as safety stocks, were assessed to be sufficiently deep and therefore did not need to be extended.Footnote 12 However, COOP had costs associated with controlling the intermodal transportation concept in form of 1a) extra personnel to manage the concept and 1b) investments in trailers that can easily be transferred from rail to road in case of major delays:
Three out of COOP’s four transport planners work more or less full-time on the transfers from rail to road, the organization of the road transports, and the handling of the consequent effects for COOP on the turnaround of the shuttle train in case of major train delays. We calculate costs of about €235,000 for the 16-month period we study, assuming three additional planners, monthly wages (including social costs) of €4412 and €24,000 costs for office space for the three planners (€4412×16×3+€24,000).Footnote 13
In total 72 trailers (36 per direction)Footnote 14 are used that can easily be transferred from the shuttle train to a truck in case of a major train delays. The investment costs for these transferable trailers are around €23,529 higher than for standard trailers. Based on an assumed life-span of 10 years we estimate the additional costs for the more expensive trailers to be around (72 × €23,529/10 = €169,000 per year and about €223,000 for the 16-month period.Footnote 15
In addition to the precautionary costs COOP paid operational costs during the 16-month period, additional costs in case of major delays made up of 2a) extra road transport costs, 2b) extra costs in the receiving shops, and 2c) extra costs in the receiving warehouse:
If we assume transfers from rail to road on the 24 occasions when the train delays were more than 2 h, an average 29 trailers per train and road transport costs of €529 per trailer (which corresponds to half the costs for a trailer transport by truck from Bro to Helsingborg), we get extra road transport costs of €368,000 (24 × 29 × €529). According to COOP, no costs fall away.
The southbound train that serves the stores is more time sensitive than the northbound train that goes to the warehouse. If the train arrives late at the railway stations in Alvesta or Helsingborg, the trucks probably reach the stores too late as well. According to COOP, this means that the employees who pick up the products in the shops have to wait and as a consequence working hours are lost.Footnote 16 COOP assumes staff costs of about €29 per hour and that it takes about 1 h to pick up one rolling pallet. This implies extra staff costs of €403,680 if 20 rolling pallets per trailer, 29 trailer per train and 24 major delay occasions are anticipated (24 × 29 × 20 × 29).Footnote 17
The additional staff costs in the warehouse due to late arrivals of the northbound shuttle train on 75 occasions are assumed to be much lower than for the southbound train. We assume €12,000.
For the two cancelled trains during the 16-month period COOP estimates direct costs in the form of increased transport costs to be € 200,000.
The estimates for COOP’s precautionary and operational costs are listed in Table 3. We have also calculated the precautionary and operational costs per tonne-hour (expressed in € per tonne-hour) based on the information that COOP experienced about 400 delay hours between 1 June 2011 and 30 September 2012 and assuming 29 trailers per train and ten tonnes per trailer. The precautionary costs of about €4 per tonne-hour are the VTTV for the transports we study. It is obvious that there is a trade-off between the precautionary and operational costs. For instance, the additional road transport costs in the case of major delays would be higher without the transferable trailers. Alternatively we can arrive at a VTTV estimate based on revealed preference data for the specific COOP case by dividing the precautionary costs by the standard deviation of the transport time. If we assume costs for precautionary measures of around €458,000 (see Table 3) and standard deviation of 63.3 min (see section 3.2), we compute a VTTV related to the cargo component of € 2.2 per hour standard deviation.
The figures in Table 3 are based on assumptions and have to be regarded with caution. However, they show COOP’s willingness to take precautionary measures such as the employment of extra staff to manage the intermodal road-rail concept and the investment in more expensive transferable trailers that can be used to reduce the costs in case of major train delays. They also show that COOP paid additional operational costs when the trains were heavily delayed or cancelled, largely explaining the difference between precautionary costs and operative costs. As mentioned above, there is clearly a trade-off between the precautionary and operational costs; it also has to be taken into account that the combined road-rail concept is quite new and there are probably learning curves for all stake holders. As an example, in October 2011 COOP engaged a rail operator for the shuttle train that uses more powerful locomotives than the operator that ran the shuttle train 2009–2011. This is expected to reduce the delays caused by leaves on the track.Footnote 18
For COOP’s incoming products that go to the warehouse, transportation costs are in general the most important factor for mode choice. However, for outgoing transports to the shops transport time and frequency are also important mode choice criteria. Transport time variability is seen as closely related to the frequency. Low reliability is manageable if frequency is high (see for example Vierth ).
Rail transports costs per tonne-km are generally estimated to be lower than road transport costs. However, it is regarded as a disadvantage that deliveries by train (containing a maximum of 36 trailers) are concentrated at one point of time and not – like deliveries by truck – spread over time. Typically, damage in intermodal transport chains is more costly than in pure truck transports, since it is often unclear where in the transport chain the damage occurred.
There is a trade-off between the exploitation of economies of scale that requires high volumes (and low frequencies) and the number of departures. Due to the relatively low frequency of the train (compared to trucks), major delays and cancellations typically lead to major consequences. According to COOP, one remedy would be to use shuttle trains jointly with other grocery companies. Many food producers are located in the south of Sweden, while several grocery companies have their warehouses close to Stockholm. This would reduce the transportation costs per unit (economies of scale can be exploited), the frequency could be increased (if several firms started shuttle trains), and the whole (larger) system would probably be less vulnerable.
As mentioned above, COOP has given up using the intermodal road-rail system for the transportations of time-sensitive fresh fruits and vegetables. This means that the reliability problems in the rail system prevent the grocery company from using the cheapest mode for a share of their transports. However, rail transport costs do probably increase when the transport time variability for rail transports is decreased.