5 Steps of Predictive Field Operations
For many years, we have been working on optimization of various field operation issues in may different industries, utilizing Operations Research and Cloud Computing technologies. Recently, we have analyzed 10 different industries and designed a joint model that can address the common problems we came across in these industries. Trying to represent our studies in 3 words, we came up with a new (?) term Predictive Field Operations (PFO). (Well at least Google says its new...)
Most of the businesses that we come across on a daily basis, may be subject to PFO as long as there is supply & demand in the field with respect to certain goods and services at various locations.
There are 5 common problems in PFO, that are usually addressed in the following order:
- Inventory Management
- Order Planning
- Distribution Planning
- Disruption Management
The reason we refer the process as PFO is that the first step is prediction. This could be the prediction of transactions in an ATM, number of newspapers sold each day, total gas sold at a gas station, or the beer consumption at a pub and many more.
The prediction can be based on different techniques including time-series analysis, regression and neural-networks. In particular, importance of recurring events are usually analyzed with pattern recognition methods. Depending on the problem, choosing the best method might be based on a different metric such as MSE, NAE, etc.
Once there is a forecast for a reasonable horizon, the second question is to identify the critical inventory limits at each location. This is not a straight-forward calculation since any prediction has a margin of error.
Therefore it's important to know that error margin before hand, and arrange the reserve stocks such that regardless of the prediction error, the ATM will have cash, the gas station will have gas, the stand will not run out of today's paper, ...and most importantly of course, the pub will always have enough pints of Guinness.
Now that we have the predictions and critical inventory targets levels, the next problem is to decide optimally when to visit each location as well as the amount of replenishment.
In Banking, the output could be; for 99.7% availability, load the ATM with $50.000 on Wednesday and send additional $40.000 on Friday. For the newspaper stand, it could be; if you want 95% paper availability, send 8 papers tomorrow, if you want 98% availability, send 12. For the gas station, if you want to keep minimum 20%, send 5 tons, if you want to keep 10% send 4.6 tons. The general concept is again similar across different industries.
Once the orders are planned, the next step is to allocate those orders to the available fleet for distribution. Essentially a Vehicle Routing Problem (VRP), usually with a time-window, many different constraints as well as pick-up and delivery combined.
The aim at this stage is to minimize the cost by using minimum vehicles, and/or minimizing distance covered by them, while satisfying the delivery constraints. These delivery constraints can be reaching to the Central Bank before 4 p.m., or delivering the newspapers before the first subway line arrives near the news stand.
Last but not least, since this is a very practical and day-to-day operation, anything can happen. Road blocks, emergency demands, plan changes, customer requests are just some of the incidents that can occur any time. Therefore the fifth step is to re-optimize the system near real-time with ever-changing circumstances. A bank branch requesting additional cash, or a package delivery company being requested a pick-up, and re-routing the delivery operation are two typical examples.
In this stage, much faster, usually heuristic algorithms are required. But also, it's important that the routing stage is planned flexible enough so that, the re-optimized system will not be adding formidable additional costs.
In some of the PFO problems, these 5 issues can be solved separately, making life easier. For example, in newspaper distribution, every day, each newsstand is delivered newspapers. So the problem of deciding how many newspapers will be delivered can be solved independently from the routing and delivery optimization. (Even in this one, we have some assumptions such that the trucks are not capacity limited with respect to newspapers.)
However in cases such as Gas distribution, even the last step cannot be solved independently from the first step. The Gas Stations are usually filled to 100% capacity when replenished. Therefore, the amount that can be delivered changes depending on which hour of the day the truck arrives. So, when the gas station stops of the replenishment truck are re-ordered, the delivery amount for each gas station needs to be updated as well.
For cash optimization, replenishing an ATM could be much more cost effective if a nearby ATM is already being replenished on that day. Hence route optimization and order planning are solved together. But, if the CIT operation is outsourced, each ATM stop is charged a fix cost to the Bank, so the order planning can be optimized independently from the route planning.
Regardless of complexity, PFO concepts will likely be getting more and more important as concepts such as same-day delivery, even drone based delivery becomes a part of our everyday lives. We will be needing end-to-end approaches similar to PFO, as the number of constraints keep increasing at the same time with the cost pressure on operations.
But one thing is for sure: it will be fun to solve them :)