Optimising High-Speed Packing with Predictive Analytics

5/1/2025

In my last post, I highlighted the importance of pairing data platforms with the right expertise to solve business challenges. Today, I’d like to share a practical example of how I helped a business optimise its high-speed packing operations by tackling a key bottleneck: packing lane reconfigurations during run changes. Using predictive models and custom analytics, I empowered operators to make faster, data-driven decisions in this area. This reduced downtime and improved their ability to adapt to shifting priorities in real time.

The Challenge: Data Without Direction

The company had robust data collection systems and real-time dashboards, but these tools only provided lagging indicators—alerting operators to problems after they occurred. While this helped identify issues, it didn’t help prevent them. Decision-making remained largely reactive, and preventable delays continued to impact production.

A particularly difficult bottleneck was managing packing changes between runs. Three key challenges stood out:

Inaccurate Grower Data
Operators relied on grower-provided averages for fruit size, which were often inaccurate and lacked information regarding variability. Since fruit size and grade weren’t determined until after bins were processed, operators had little time to adjust packing lanes. Although the company had plenty of data, it lacked the right expertise to turn it into the information they needed.
Repeated Setup Changes
Operators frequently struggled to optimise lane configurations, leading to repeated adjustments and inefficiencies. Packing lanes were often reconfigured multiple times for the same product, disrupting workflows and wasting time. Operators knew there had to be a better way to optimise these decisions but lacked the right tools.
Limited Ability to Adapt
Production priorities changed frequently, but operators found it challenging to implement updates quickly. Priority changes were shared through emailed PDFs or printed sheets, while run schedules were stored in SharePoint Excel files. These manual processes caused delays and reduced the team’s ability to adapt in real time.

These inefficiencies often forced operators to spend up to 30 minutes adjusting setups after a run had already started, limiting their ability to focus on other tasks.

A Smarter Tool for Proactive Decisions

I developed a web-based tool that integrated seamlessly with the company’s existing systems. Built in Python and hosted on a central server, the app brought together data from SQL databases, PDFs, and Excel files, making production schedules and packing priorities easily accessible.

Operators could update run attributes and packing priorities directly within the app, which then recalculated production forecasts and generated optimal lane assignments. This provided the real-time flexibility needed to adapt swiftly, aligning production with the latest priorities.

How it Worked:

Predictive Analytics with Bayesian Models
I built Bayesian regression models to forecast the distributions of fruit sizes and quality based on historical and real-time data. These forecasts allowed operators to proactively plan lane configurations well in advance, reducing the need for last-minute adjustments.
Optimisation Engine
The optimisation engine combined predictive insights from the Bayesian models with key business rules and production constraints to recommend the most efficient lane setups. By factoring in forecasted fruit size distributions, packing priorities from marketing contracts, and throughput capacities of packing lanes, the engine prioritised high-value packaging while minimising unnecessary reconfigurations.
Real-Time Adaptability
As new data arrived, the tool recalculated forecasts and lane recommendations, ensuring operators could dynamically adapt to changes with minimal disruption. Operators could tweak forecast outputs and re-optimise in response to changes to maintain an efficient and responsive production workflow.
Empowering Operators
The tool was designed to augment operator expertise rather than replace it. By providing data-driven recommendations while allowing manual overrides, it enabled operators to remain in control of the process. This balanced approach not only improved decision-making but also increased operator confidence and engagement.

Here are some key ways in which solutions could be customised:

Modify Production Plan: Operators could adjust bin attributes to account for things they may know about the fruit supply that the technology doesn't, and get new forecasts based on these changes.
Packing Priority Adjustments: Operators had the flexibility to exclude certain products or shift focus based on evolving market demands or contract requirements. The optimisation engine then took these changes into consideration.
Dynamic Throughput Rates: While the tool recommended throughput rates for each lane based on historical data, operators could modify these settings to run faster or slower. The system updated throughput constraints and provided revised lane setups based on these changes.

Proactive, Data-Driven Operations

Although the tool was implemented late in the season, early results showed significant improvements:

Faster Setup Times: Operators now plan the next run in just a few minutes, compared to up to 30 minutes previously. This time saving allows for more efficient operations and quicker transitions between packing runs.
Increased Operator Confidence: Accurate predictions and optimised setups have reduced guesswork, allowing operators to make quicker, more confident decisions. The ability to modify inputs and revise solutions has empowered them to fine-tune the process, reducing guesswork and reinforcing their decision-making capabilities.
Reduced Downtime: Fewer mid-run adjustments have improved throughput and reduced delays. By anticipating needs ahead of time and having proactive solutions in place, operators are better able to keep the flow of production steady and efficient.
Proactive Decision-Making: Operators can now anticipate issues before they arise, adjusting as necessary to maintain smooth operations. This has resulted in better planning, fewer disruptions, and overall improved efficiency in the packing process.

What This Means for You

This example demonstrates how addressing a single bottleneck with targeted predictive analytics and optimisation tools can lead to significant improvements, without the need for a complete overhaul of operations or data systems.

The principles and tools described here are adaptable to a wide range of industries. Whether you're in manufacturing, logistics, or customer service, the right combination of predictive models and optimisation tools can unlock hidden value within your existing data. In many cases, businesses don’t need more data—they need the right expertise to transform that data into actionable, effective solutions that drive operational efficiency and support smarter decision-making.

1 Comment