PredictFlow — AI-Driven Resource Orchestration for Railway Centre Operations

AI Quick Summary

Planzer's railway centres operate under extreme time pressure. Parcels arrive overnight by train and must be sorted, loaded, and dispatched by 7am. Delays are unpredictable, employee turnover on the night shift averages four months, and any disruption cascades through the last mile. Today, planning is reactive. PredictFlow is an AI powered system that analyses historical parcel volumes, train schedules, delay patterns, and sender behaviour to forecast each night's reality before it unfolds. It recommends staffing levels, sorting priorities, and van loading sequences. When a train runs late, PredictFlow recalculates so supervisors adapt before the bottleneck hits. The pilot validates predictions against Planzer's historical data, then runs in shadow mode alongside live operations. Robotics and automation can transform the physical flow, but they require intelligence first. PredictFlow provides that foundation. The 3am Problem Every working night across Switzerland, Planzer's railway centres come alive while the country sleeps. Hundreds of railway wagons carry roughly 100,000 parcels through the night. Between 3am and 7am, teams of workers unload those wagons, place parcels on the sorting belt at a pace of over 90 seconds per parcel, and load them into delivery vans that must leave by 7am. This happens every single working day, without exception. But this system runs on a razor's edge. Swiss law prohibits trucks on public roads at night, making rail the only viable long distance transport option. The majority of parcel volume moves by train. The SBB timetable is fixed. When a train runs late, the truck dispatch time does not change. The entire downstream chain absorbs the delay, and the people who absorb it most are the night shift workers. Those workers stay an average of four months. Four months of 3am starts, physical sorting, and the stress of a system where one person missing creates immediate disruption across the operation. Planzer invests in training each new employee, only to repeat the cycle a few months later. The cost of this turnover is enormous, and it compounds the very problem it stems from: an operation that depends heavily on human labour in conditions that make retention nearly impossible. Cost per parcel is the metric that drives the business. Labour is the largest component of that cost. Robotics and automation could transform the physical flow from train to conveyor belt and beyond, but deploying machines into a process this complex and variable requires intelligence first. Before any robot can act, something must understand what is coming, when it will arrive, and how priorities should shift when conditions change. That intelligence does not exist today. Where the Data Breaks Down Today, data from senders arrives in bulk at the end of the day. Parcel weight and size are collected at pickup, but the measurements are not always accurate. Parcels are re-weighed later in the process, by which point it is too late to use that information for planning. Across Planzer's network, each hub tends to follow its own process, with no standardised way to anticipate volume or complexity ahead of time. The scan process only begins when the driver picks up parcels at the ramp. Before that moment, the railway centre has limited visibility into what is actually coming. Supervisors plan shifts based on experience and averages, not on what tonight's trains will actually carry. This is the gap PredictFlow is designed to close. What PredictFlow Does PredictFlow is an AI powered predictive planning system that analyses data Planzer already collects, including parcel scan records, sender shipping patterns, SBB schedules, train delay histories, seasonal volume fluctuations, and shift performance logs, to forecast what each night will look like before it begins. The system answers the questions that supervisors currently resolve through intuition. How many parcels will arrive tonight, and on which trains? Which trains are at risk of delay? How many workers are needed on the sorting line? Which delivery routes will carry the heaviest loads, and which vans should be loaded first to protect the 7am departure? PredictFlow also addresses the parcel data quality problem. By learning from historical patterns between initial scan data and actual re-weighed measurements, the system builds correction models that improve weight and size estimates before parcels even reach the hub. Over time, this creates a more accurate picture of what each wagon is carrying, enabling better load planning and sorting prioritisation. When conditions change mid shift, a train running 40 minutes late for instance, PredictFlow recalculates sorting priorities and van loading sequences in real time. The supervisor sees the adjusted plan on a dashboard and can act immediately, rather than discovering the problem when parcels stack up on the belt. Why This Matters for Cost Per Parcel Every minute of reactive scrambling during the night shift costs money. Overstaffing on quiet nights wastes labour. Understaffing on heavy nights creates backlogs that delay van departures and cascade into missed delivery windows. Each new hire who leaves after four months represents thousands of francs in recruitment, training, and lost productivity. PredictFlow attacks cost per parcel from multiple angles. Accurate volume forecasts enable right sized staffing, eliminating waste on light nights and preventing chaos on heavy ones. Better data upstream means fewer surprises at the sorting belt. And by reducing the unpredictability that makes the night shift so punishing, PredictFlow helps extend employee tenure. Even moving average retention from four months to eight months would dramatically reduce hiring and training costs across the network. Why This Matters for Employees PredictFlow contributes to better working conditions by removing the uncertainty that makes the shift so stressful. If the system predicts a light night, the shift can start later or run with fewer people. Nobody stands in a cold warehouse at 3am waiting for a train that will not arrive until 4:30. If it predicts a heavy night, extra resources are arranged in advance rather than scrambled for when the first wagon doors open. The difference between a planned night and a chaotic one is the difference between a job people can sustain and one they leave after four months. A Feasible, Phased Approach PredictFlow is designed to prove itself incrementally, without disrupting existing operations or requiring infrastructure changes. Phase one is retrospective validation. Using 6 to 12 months of Planzer's historical data from one or two railway centres, we build and train prediction models, then test them against what actually happened on those nights. This proves accuracy before anything touches a live operation. It also surfaces data quality issues early, particularly around the weight and size measurements that have been identified as unreliable. Phase two is shadow mode. PredictFlow runs alongside real operations at a pilot location, generating predictions each evening that shift supervisors can view on a dashboard. They do not have to follow the recommendations. We simply measure how often PredictFlow's forecast was closer to reality than the existing planning method. This builds trust through evidence. Phase three is active integration. Once predictions are validated and trusted, supervisors begin using PredictFlow's recommendations to adjust staffing, sorting sequences, and van dispatch priorities. The system becomes a working part of the nightly operation. Because each hub currently follows its own process, PredictFlow is designed to learn each location's specific patterns rather than imposing a single model. This respects the reality of Planzer's decentralised operations while gradually building a standardised intelligence layer across the network. The Foundation for What Comes Next Planzer's vision calls for AI, humanoid robots, and Level 4 autonomous vehicles to converge into a seamless logistics system. PredictFlow is the intelligence layer that this vision depends on. When robotic systems are deployed to automate physical parcel handling, something will need to tell those robots how many parcels to expect, what sizes and weights are coming, and how to reprioritise when a train is late. That something is PredictFlow. When the sender to hub process is improved with better data capture, something will need to ingest that real time data stream and translate it into actionable plans for the railway centre. That something is PredictFlow. When autonomous vehicles are integrated into last mile operations, something will need to decide which vehicles to dispatch first and how to resequence departures when the sorting schedule shifts. That something is PredictFlow. Before machines can act, something must anticipate. PredictFlow is that intelligence.

Project Idea Metadata

• Project Idea Name: PredictFlow — AI-Driven Resource Orchestration for Railway Centre Operations
• Date: 4/4/2026 11:15:55 PM
• Administrators:
  • Elie Maalouli

Project Idea Description

The 3am Problem

Every working night across Switzerland, Planzer's railway centres come alive while the country sleeps. Hundreds of railway wagons carry roughly 100,000 parcels through the night. Between 3am and 7am, teams of workers unload those wagons, place parcels on the sorting belt at a pace of over 90 seconds per parcel, and load them into delivery vans that must leave by 7am. This happens every single working day, without exception.

But this system runs on a razor's edge.

Swiss law prohibits trucks on public roads at night, making rail the only viable long distance transport option. The majority of parcel volume moves by train. The SBB timetable is fixed. When a train runs late, the truck dispatch time does not change. The entire downstream chain absorbs the delay, and the people who absorb it most are the night shift workers.

Those workers stay an average of four months. Four months of 3am starts, physical sorting, and the stress of a system where one person missing creates immediate disruption across the operation. Planzer invests in training each new employee, only to repeat the cycle a few months later. The cost of this turnover is enormous, and it compounds the very problem it stems from: an operation that depends heavily on human labour in conditions that make retention nearly impossible.

Cost per parcel is the metric that drives the business. Labour is the largest component of that cost. Robotics and automation could transform the physical flow from train to conveyor belt and beyond, but deploying machines into a process this complex and variable requires intelligence first. Before any robot can act, something must understand what is coming, when it will arrive, and how priorities should shift when conditions change. That intelligence does not exist today.

Where the Data Breaks Down

Today, data from senders arrives in bulk at the end of the day. Parcel weight and size are collected at pickup, but the measurements are not always accurate. Parcels are re-weighed later in the process, by which point it is too late to use that information for planning. Across Planzer's network, each hub tends to follow its own process, with no standardised way to anticipate volume or complexity ahead of time.

The scan process only begins when the driver picks up parcels at the ramp. Before that moment, the railway centre has limited visibility into what is actually coming. Supervisors plan shifts based on experience and averages, not on what tonight's trains will actually carry.

This is the gap PredictFlow is designed to close.

What PredictFlow Does

PredictFlow is an AI powered predictive planning system that analyses data Planzer already collects, including parcel scan records, sender shipping patterns, SBB schedules, train delay histories, seasonal volume fluctuations, and shift performance logs, to forecast what each night will look like before it begins.

The system answers the questions that supervisors currently resolve through intuition. How many parcels will arrive tonight, and on which trains? Which trains are at risk of delay? How many workers are needed on the sorting line? Which delivery routes will carry the heaviest loads, and which vans should be loaded first to protect the 7am departure?

PredictFlow also addresses the parcel data quality problem. By learning from historical patterns between initial scan data and actual re-weighed measurements, the system builds correction models that improve weight and size estimates before parcels even reach the hub. Over time, this creates a more accurate picture of what each wagon is carrying, enabling better load planning and sorting prioritisation.

When conditions change mid shift, a train running 40 minutes late for instance, PredictFlow recalculates sorting priorities and van loading sequences in real time. The supervisor sees the adjusted plan on a dashboard and can act immediately, rather than discovering the problem when parcels stack up on the belt.

Why This Matters for Cost Per Parcel

Every minute of reactive scrambling during the night shift costs money. Overstaffing on quiet nights wastes labour. Understaffing on heavy nights creates backlogs that delay van departures and cascade into missed delivery windows. Each new hire who leaves after four months represents thousands of francs in recruitment, training, and lost productivity.

PredictFlow attacks cost per parcel from multiple angles. Accurate volume forecasts enable right sized staffing, eliminating waste on light nights and preventing chaos on heavy ones. Better data upstream means fewer surprises at the sorting belt. And by reducing the unpredictability that makes the night shift so punishing, PredictFlow helps extend employee tenure. Even moving average retention from four months to eight months would dramatically reduce hiring and training costs across the network.

Why This Matters for Employees

PredictFlow contributes to better working conditions by removing the uncertainty that makes the shift so stressful. If the system predicts a light night, the shift can start later or run with fewer people. Nobody stands in a cold warehouse at 3am waiting for a train that will not arrive until 4:30. If it predicts a heavy night, extra resources are arranged in advance rather than scrambled for when the first wagon doors open. The difference between a planned night and a chaotic one is the difference between a job people can sustain and one they leave after four months.

A Feasible, Phased Approach

PredictFlow is designed to prove itself incrementally, without disrupting existing operations or requiring infrastructure changes.

Phase one is retrospective validation. Using 6 to 12 months of Planzer's historical data from one or two railway centres, we build and train prediction models, then test them against what actually happened on those nights. This proves accuracy before anything touches a live operation. It also surfaces data quality issues early, particularly around the weight and size measurements that have been identified as unreliable.

Phase two is shadow mode. PredictFlow runs alongside real operations at a pilot location, generating predictions each evening that shift supervisors can view on a dashboard. They do not have to follow the recommendations. We simply measure how often PredictFlow's forecast was closer to reality than the existing planning method. This builds trust through evidence.

Phase three is active integration. Once predictions are validated and trusted, supervisors begin using PredictFlow's recommendations to adjust staffing, sorting sequences, and van dispatch priorities. The system becomes a working part of the nightly operation.

Because each hub currently follows its own process, PredictFlow is designed to learn each location's specific patterns rather than imposing a single model. This respects the reality of Planzer's decentralised operations while gradually building a standardised intelligence layer across the network.

The Foundation for What Comes Next

Planzer's vision calls for AI, humanoid robots, and Level 4 autonomous vehicles to converge into a seamless logistics system. PredictFlow is the intelligence layer that this vision depends on.

When robotic systems are deployed to automate physical parcel handling, something will need to tell those robots how many parcels to expect, what sizes and weights are coming, and how to reprioritise when a train is late. That something is PredictFlow.

When the sender to hub process is improved with better data capture, something will need to ingest that real time data stream and translate it into actionable plans for the railway centre. That something is PredictFlow.

When autonomous vehicles are integrated into last mile operations, something will need to decide which vehicles to dispatch first and how to resequence departures when the sorting schedule shifts. That something is PredictFlow.

Before machines can act, something must anticipate. PredictFlow is that intelligence.

Planzer's railway centres operate under extreme time pressure. Parcels arrive overnight by train and must be sorted, loaded, and dispatched by 7am. Delays are unpredictable, employee turnover on the night shift averages four months, and any disruption cascades through the last mile. Today, planning is reactive.

PredictFlow is an AI powered system that analyses historical parcel volumes, train schedules, delay patterns, and sender behaviour to forecast each night's reality before it unfolds. It recommends staffing levels, sorting priorities, and van loading sequences. When a train runs late, PredictFlow recalculates so supervisors adapt before the bottleneck hits.

The pilot validates predictions against Planzer's historical data, then runs in shadow mode alongside live operations. Robotics and automation can transform the physical flow, but they require intelligence first. PredictFlow provides that foundation.