Grocery distribution is often the leader in logistics practice, and attention understandably focuses on the major national and global players, with their sophisticated and automated Distribution Centres and e-commerce/home delivery networks driving down costs and broadening the offer. But many, perhaps most, inhabitants of urban areas in developing, and even advanced, economies, struggle to take advantage of these developments – they may lack mobility, or credit, or just the ready cash to afford a full retail pack of a product. For their daily staples they rely on the large numbers of very small, hyper-local ‘nanostores’.
But the nanostores face an increasing challenge from supermarkets, chains of convenience stores (which may offer other services as well) and, to a lesser but growing extent, e-commerce and while many of their customers can’t make this move, it only takes a small loss of trade to render nanostores unviable, with serious economic and social implications. They, and the distributors that supply them, need to find ways of reducing costs and improving service, but with little or no access to working capital, many of the techniques adopted by ‘modern’ grocers are simply unavailable. Are there low-investment approaches that this sector can adopt to remain competitive?
That is the question we have been studying with a distributor in Guadalajara, Mexico. This fairly new company buys goods in (relative) bulk from suppliers, wholesalers and central markets and distributes them, in quantities down to single items, to a current roster of 121 nanostores, or ‘tienditas’ as they are known in Mexico
The operation could be a definition of ‘lean’ – assets essentially comprise one warehouse with operator, and one delivery van with driver. However, it also defines ‘stochastic’ – customer orders are typically for one or two boxes of mixed goods, quite unpredictably – when stores are about to run out, or when they have the funds for replenishment (this is a Cash on Delivery operation). Even if they could fund purchase in larger volumes, few tienditas have any ‘back-office’ storage space. Demand data, and thus the opportunity for forward planning, is effectively non-existent.
In an attempt to introduce a little more visibility, the company’s offer to storeholders is a mobile and online ordering app linked to an easily accessible online catalogue, for order delivery within two days. They have enhanced this by making storeholders an offer of free delivery on a set weekday. This has been popular and has helped the company smooth out to some extent the peaks and troughs in delivery rounds. It has also helped the company to begin to collect demand data which is of potential value both to their own operations and to suppliers and wholesalers. These are also typically small firms – this form of distribution has few attractions for the larger food, grocery and CPG (consumer packaged goods) companies. A further opportunity lies in partnering with credit providers who are prepared to offer storeholders small, recurring lines of credit based on this newly available data. It also helps create better ordering patterns and thus improve in-stock rates.
So far so good. But the company aims to scale up its operations (the ‘universe’ of tienditas in Guadalajara is around 19,000 so there is plenty of upside growth potential!). They asked us to research possible further improvements at current scale, especially in more efficient route planning, whether and how their model would work across more clients and locations, and what the impacts on costs and resources would be.
Mapping value (and waste)
Our starting point was a Value Stream Mapping Analysis. Results were fed into supply chain network design which enabled us both to validate our baseline model and to explore the effects of future strategic, tactical and operational decisions on resources, capabilities and costs – in total 14 different future scenarios were modelled. We addressed the location routing problem (LRP) by using a machine-learning clustering method (K-means) to segment the customer base, and estimated the routing costs using a Continuum-Approximation approach.
It could be argued that the approach is a little simplistic and approximate, but this was both appropriate and necessary. For example, using an exact method like Mixed Integer Linear Programming for the LRP would have been practical at current scale, but the intermittent nature of demand and other uncertainties would render this fiendishly complex at larger scale for doubtful benefit, or even a quite spurious illusion of accuracy. Likewise, deliveries were represented by a single ‘product’ – a box of unspecified mixed goods. Breaking this down to individual SKU level would again greatly increase the complexity, and anyway adequate demand data at this level is just not available.
Nonetheless this approach has yielded many actionable insights, some of them unexpected or counter-intuitive.
We looked for the classic ‘seven steams of waste’ (inventory, waiting, movements, defects, overproduction, transport, overprocessing) in the current state Value Stream Map, following activity from receipt of order to delivery. We identified scope for improvement including batch picking, reorganisation of the purchasing process, centralisation of order processing information, and the introduction of a warehouse allocation system for more efficient order picking and maintenance of a First in First out regime (the assortment includes perishable goods). The company was anyway contemplating a Warehouse Management System.
Unexpectedly, the major bottleneck in the current operation actually lies outside the scope of the VSM – in physical procurement, collecting goods from suppliers and wholesalers. This is a very variable but time-consuming activity, often involving extra hours, and requiring use of the delivery van which, of course, is only available once daily order deliveries have been fulfilled. Buying or leasing a second, smaller, vehicle and hiring another driver, is a potential solution. The procurement process generally needs to be separated from warehouse operations.
Overall, we identified ways of increasing the warehouse staff full-time equivalent from a half to one, which with other improvements would raise capacity from an often inadequate 19 orders per working day to around 78.
Looking at vehicle use, as hoped the clustering approach successfully reduces the variability of daily operations, and thus the need for an additional delivery vehicle. At present there are times when delivery requirement exceeds capacity although on average the van has only a 25% utilisation factor. With effective clustering and fixed-day deliveries, additional powered vehicles shouldn’t be needed until the customer base rises to around 500. There may be scope for addressing demand peaks by bicycle delivery (we also looked at the possibility of using smaller vans of the ‘Kangoo’ type). The model successfully drew out the contrast between the high fixed cost of van leasing and driver salary, as opposed to the low fixed but high variable (wages) costs of bicycles.
The high fixed-cost element means that there is not a big difference in delivery costs between a highly concentrated and a more dispersed customer base: linehaul distances are not very relevant in a city environment, and what matters is how many deliveries can be made on a round, not the distance between them. In practice the greater the number of deliveries, the shorter the distance travelled on each round.
The company also asked us to look at any benefits from moving to a warehouse closer to the central market (the major supply location) but we were unable to identify efficiencies that would offset the increased rent.
In summary, our approach, although simplified and approximate, enabled us to offer the company a practical decision-making tool to properly cluster the customer base into segments and evaluate the impact on costs and service quality, incorporating new data on future demand and location of potential customers. We successfully identified a number of areas where improvements can be made for relatively small investment, the scaling points at which significant new investment would likely be required, and also areas where an apparent need for investment can in fact be avoided. We think this sort of approach could have wider applicability to start-ups and small firms looking to scale up the physical capabilities of an e-commerce driven operation and to extend the benefits to the smallest retail outlets and their patrons.
The research was carried out by Teresa Barros, Christopher Czech, Rodrigo Pérez , for their ZLC Master of Engineering in Logistics and Supply Chain Management thesis titled “Lean operations design and Machine Learning-enhanced scenario planning approach for an e-commerce platform”. The thesis won the ZLOGb Outstanding Thesis Award.