Uncertainties that plague logistics operations come in many forms from many different causes. Operational uncertainty arises when customers do not provide clear or correct data on the quantity, dimension or characteristics of the cargo to be picked up or when it is to be picked up, or may send the order too late and may not be ready for the goods to be picked up at the appointed time. There may be waiting time incurred because of other trucks waiting to load or unload. There may be errors or lack of clarity in the handling instructions or delivery address or destination dock constraints. Strategic uncertainties may arise when demand shifts or when ports become congested and resources such as trucks, containers or trucks may not be available due to disasters, strikes, lockdowns, etc. All these uncertainties and the lack of coordination, and variable demand/poor information between customers, shippers and carriers result in significantly reduced efficiency and performance. But how are uncertainties currently handled and how can they be handled better?
Most logistics operators try to deal with operational uncertainties by seeking better information on the whole operation – where the shipments are, availability of docks, etc – and build command and control centers in which the latest and the most detailed data are available for managers to visualize what is happening. While better data never hurts, the problem with this approach is the mistaken belief that the humans will be able to discern what is happening, work out what is likely to occur and know what to do once they have right and timely information. Unfortunately, in complex evolving situations (and logistics operations are full of these), humans cannot detect problematic anomalies and predict their consequences reliably. They often develop experience-based heuristics involving creating sufficient buffers to cater for uncertainties which may not only be inefficient but also may not adequately deal with the problems which arise. Uncertainties are too difficult and too dynamic for humans to anticipate well and they resort to reacting to situations as best as they can.
I came across two examples of such behaviors which I will share to illustrate how logisticians tend to deal with uncertainties. The first is the deployment of airtugs at Changi Airport to push aircraft out for departure or tow them to maintenance hangars. The crews of these airtugs used to know where to go to service departing flights as gate allocations used to be quite static. However, as the airport became more congested, the gate allocations became unpredictable and crews started to return to base as they did not know what their assigned flights will be and which gate they will depart from. Productivity plunged as the crews took a longer time to get to their assigned flights. The problem cannot be solved by getting more information faster as the underlying problem is the uncertainties caused by congestion and delays or early arrival of inbound flights. Another example of uncertainties affecting productivity is that of a typical airfreight logistic company that receives a stream of delivery and pickup jobs from customers sometimes just before drivers are due to leave or have left the depot and sometimes amend these jobs. Despatchers then scramble to amend the work assignments to fit the jobs into existing routes as best as they can – and they have learnt to keep spare capacity in each truck to cope with uncertainties. As a result, it is challenging to achieve high productivity in terms of minimum distance travelled and maximized capacity utilization.
So how can one do better in the face of such uncertainties? Are not the management in these examples already doing the best they can to tackle the problem? Well, to answer these questions, let us see what happens when we anticipate rather than react. In the case of the airport, each flight had departed from a handful of gates with different probability for each of these gates. Human planners will struggle to discern anything more than the obvious patterns from historical data and are thus unable to place their bets well. In the case of airfreight, the likelihood of a delivery or pickup order and the likely timing, size and nature of the job can also be discerned from past data to enable these jobs to be anticipated. Travel times between locations and dwell times at each location also can be predicted as they evolve on different days of the week or time of the year. Using AI to discern patterns from historical data helps in anticipating jobs in such a way that resources are deployed more efficiently than one in which despatchers and drivers are reacting to the jobs as they arrive. Strategic uncertainties are best dealt with anticipation as well but they involve explicit considerations of the likelihood and consequences of each scenario so that a good set of contingency plans and their triggers can be developed.
Of course anticipation is never perfect and thus routes and plans must still be developed with some buffer for the uncertainties but such buffers would be optimized for the residual uncertainties and not allocated based on imprecise estimations of humans based on subjective judgment. The execution of uncertainty-robust plans also require careful consolidation of actual jobs with anticipated ones to not only avoid double accounting but also to continuously learn as the future unfolds. The experience and ingenuity of despatchers will still be important in scenarios in which anticipated pickup jobs are not confirmed even though the truck is already about to arrive at the location, or actual jobs exceeded the size of the anticipated ones, but the anticipated jobs would reduce the impact of uncertainties considerably so that despatchers do not have to scramble so often. Developing a data-driven decision culture will help the logistics industry cope with the challenges of uncertainties especially in such difficult times.
Some people may think that futuristic ideas, e.g. drones, flying warehouses or self-driving cars, will eliminate the problems brought by uncertainties. Such dreams of a seamless delivery experience are hardly new but couriers such United States Postal Service, United Parcel Service (UPS) and FedEx have still stuck to their static terrority-based system which suffer from large imbalances in demand and supply in each driver territory. The digital age is putting last-mile delivery to the forefront of retailers’ minds as e-commerce and the Amazon effect require them to offer fast and free delivery, or become uncompetitive. A missed delivery, meanwhile, could mean lost sales or an less-than-happy customer. Consequently, the needs to industry seriously consider how to harness AI to move from their static processes to dynamically optimized ones. This article outlines how IDSC have partnered its customers to begin to embrace uncertainties to develop anticipatively optimal fleet routing and load planning solutions which addresses some of the challenges involved in logistics.
Bio: Mr. Cheng Hwee Sim is the managing director of INTEGRATED DECISION SYSTEMS CONSULTANCY PTE LTD (IDSC) located in Singapore.