( This article is excerpted from the complimentary report
Preemptive Freshness Management Empowering Workers to Improve Delivered Freshness,
available for download here )
In part one of this series, we examined how workers can be empowered by a system that provides them proactive alerts, guidance on goals and tradeoffs, prescriptive corrective actions at each critical step of each process in the cold chain. Here in part two of this series, we take a look at the process mapping and modeling, and predictive analytics required to make it all work.
Process Mapping and Modeling
In addition to the elements of situational awareness and context knowledge listed in part one of this series (i.e. conditions, requirements, preferences, and constraints), intelligent recommendations require a detailed and accurate understanding of the various process steps throughout the cold chain. A set of process maps is required, modeling the various processes across the operations and how they interconnect. For example, the process model used by Zest Fresh factors in the temperature the night before, current temperature, expected temperature, current backlog and expected wait time for pre-cool (influenced by the relative priority of the cases waiting for precool). All of these factors, and more, impact the temperature of the produce at the time it is finally placed into the pre-cool unit. The internal temperature of the produce, along with the type of produce, moisture, and other factors, in turn impacts the amount of time needed to pre-cool the produce to the optimal temperature. To optimize resource planning and allocation, the solution needs a process map that incorporates all of these interrelationships and has the needed mathematical models to factor them into its recommendations.
Modeling Throughput Rates
Continuing on the example above, there is a standard throughput rate assumed for the precool unit: for produce type A,1 the unit can precool X cases per hour. But if the produce is coming in at an elevated temperature, then the throughput of the precool unit will be reduced to something less than the standard X units per hour.2 In that case, the system might make a recommendation to reduce the number of crews harvesting, in order to match the rate of harvest with the capacity of the precooler. Or it might give the operations manager an alert the day before (based on the weather forecast) and recommend starting the harvest at 4AM under lights, so that there is time to precool the committed number of orders. Similarly, the throughput of the operation will be impacted by the distance of the fields from the packhouse (less or more travel time per truckload, impacting both time and temperature), whether the produce is being picked before, at, or after peak maturity (impacts days of freshness and produce size), and many other factors.
Developing throughput models and process models that consider all of these various factors takes many years of in-the-field experience. A solution that has built-in mature process maps and models, based on depth of experience, will give much more reliable and intelligent recommendations than one that uses an overly simplistic set of assumptions.
Modeling Deliverables Targets and Tradeoffs
Each operation will have a set of ‘deliverables targets’ — i.e. customer orders, with a specific number of cases of produce in each order for each shift and for the whole day. This is fairly straightforward when the operations are on a clear path to achieving their production goals for the day. It’s when you get congestion or added constraints in the operations and you need to start making tradeoffs that things are not so simple. When this happens (and it is quite common) the process model needs to take into account all the different factors associated with those production targets, such as the number of days drive time to each customer’s destination, management preferences, and differences in precool time for different types of produce, in order to optimize the allocation of available pallets of produce.
The process maps, combined with a near real-time view of the current situation and requirements, enables the system to do predictive analytics. First and foremost, it means that the system can predict expected shelf life, based on the handling and temperature exposure history of each pallet of produce. Zest Fresh also does ‘what if’ analysis, predicting the impact on shelf life of various scenarios. It is able to answer questions such as, ‘if pallets 1 through 10 sit out an extra hour, so that pallets 11 to 20 can be prioritized in the precool, can we still meet the shelf-life commitments and requirements for those first 10 pallets?’ It can evaluate thousands of potential options — much more than any human could — and come up with the best combination of actions, guided by the goals, preferences, and constraints of the operation. A grower can adjust these business rules to meet the needs and specific priorities of their own operations. When workers are making the tradeoff decisions, they are often suboptimal, resulting in reduced shelf life and freshness, loss of customer satisfaction, rejection of shipments, and reduction in revenue. By letting the system find the optimal solution, an operation can dramatically reduce waste, increase quality (freshness) consistency, and meet a much larger percentage of customer commitments, all with the same (or reduced) amount of labor and equipment.
Figure 2 – Example Process Maps and Model Parameters for Each Process Step Across the End-to-End Cold Chain
Solutions Needed for Each Process Step
An unbroken chain of intelligent, preemptive-freshness processes is required, all the way from the field to the retail shelf. If even one step is done poorly, then freshness and quality is compromised. Each process step requires its own unique process map and model, yet they are also interconnected and interlocking, as every step is dependent on the results of all prior process steps. An end-to-end process map might include models for the following processes:
The process steps are interlinked, so the system needs to be modelled holistically, end-to-end. For example, if there are unusually high temperatures one day, the system might reasonably be expected to tell workers to precool each pallet longer, to compensate for the heat and allow the produce to reach the ideal temperature. However, if the model does not take into consideration the rate the produce is being harvested and brought in from the fields, then pallets could back up in the yard waiting for their slot in the precool (or worse yet, product goes unharvested and spoils in the field). This could result in congestion, longer wait times, longer exposure to high temperatures (therefore diminished shelf life) and so forth. Since the workers in the field are paid based on how many cases they harvest, and anyway they don’t even see what is happening in the yard, they are not going to slow down just because the yard is backing up. Thus, each process step is interconnected and cannot be modeled in isolation.
As another example, consider the interconnected processes in a retailer’s receiving/putaway and replenishment operations. Each day a pallet of produce sits in the retailer’s warehouse is a day of shelf life lost. So, ideally the retailer strives to ‘pick to zero,’ sending out to its stores all pallets of produce it receives every day. In practice, it doesn’t always work that way and some retailers average up to three days of storage time in their DC for produce, often due to poor visibility into actual store demand and replenishment needs, which may vary from plan.3 In these constrained situations, the system needs to help the receiving personnel make decisions about which pallets to cross-dock and which to put away. A delivery of berries arriving with only four days of shelf life left should be prioritized over a delivery of lettuce with ten days left. Other factors that may be considered in prioritizing the work are current levels of inventory in each store, the velocity of consumption in the stores, and perhaps other management preferences.4 The system needs to be configurable to take all of these interlocking process steps and priorities into consideration.
Making it Easy for Front Line Workers to do the Right Thing
Many decisions in the produce cold chain are made every day by the workers and supervisors on the front lines: in the field, packhouse, DC, and stores. Most of them really want to do the right thing. But workers can only be as smart as the tools and goals (metrics) they are given. Often the primary metric of success is simple throughput, with little or no consideration of the impact on delivered freshness (i.e. days of shelf life remaining by the time the produce is delivered to the end customer), and other strategic goals, such as ensuring that large, long-term, key accounts are well served first. Furthermore, it is impossible for a worker, especially in the heat of battle when things are backing up, to consider all of the different tradeoffs.
What is needed is a system that provides preemptive freshness management capabilities. The best we have seen is Zest Labs’ Zest Fresh solution. No other system we’ve encountered offers the same level of preemptive freshness management capabilities for fresh produce, providing prescriptive guidance to workers at each step of each process, end-to-end, taking into account the impact across the end-to-end process model, in order to maximize the strategic goals and success of the grower and retailer. This makes it easy for the front-line workers to do the right thing, and thereby make the overall operation more competitive, successful, and a more satisfying place to work.
1 Produce with less surface area and more core, like broccoli or cauliflower, takes longer to precool than produce with more surface area and less core, like lettuce. These standard throughput rates are generally known for each type of produce for each different model of pre-cool unit and should be incorporated in the model. — Return to article text above
2 Conversely, cooler than average produce will increase the throughput above the standard X cases per hour. — Return to article text above
3 Due to weather or other events, a store may sell out quickly or have slower than expected sales. If they neglect to inform the DC of that variance in demand, the DC continues to replenish the store based on the regular rate of demand, and the store ends up being over- or under-supplied. Thus, DC fulfillment processes need to be intimately interconnected with store replenishment processes and needs. — Return to article text above
4 For example, one type of produce might be higher priority because it is a key reason shoppers come to the store. Or a seasonal or holiday-related item might take priority, like strawberries and blueberries that must arrive in time for Fourth of July sales. — Return to article text above
To view other articles from this issue of the brief, click here.