Dr. Chris Caplice, a senior research scientist at MIT and a chief scientist at DAT Freight & Analytics, talks about how data science is transforming the transportation industry.
Here's a summary of how shippers and 3PL can use data-driven insights to answer four key questions and make proactive, informed decisions.
Background
The market is huge
$700B in 2020 (3% of GDP)
1,000,000 shippers
100,000 carriers
10,000 intermediaries
Networks are complex
Wide range of different lanes
Little overlap on lanes
Less than 5% of a shipper's network is showing up in another shipper's networks
Many lanes don't repeat
A lane that has less than 12 loads per year will not be there next year
Every shipment is full of data
Continuous communication
EDI, API
Temporal and spatial changes
Spot vs. Contract rates
Spot Rates
Like "weather": day-to-day state of the atmosphere
fluctuating
Yellow line
Contract Rates
Like "climate": weather over a long period of time
stable
Blue line
Notice Spot rate rises first and Contract rate lags a little bit and rises afterward
4 key questions answered by data Science
1. Descriptive: What happened in the past?
Knowing what happened in the past is particularly important when it comes to 3PL, where high variability at any given time can make it hard for shippers to predict the future.
With data science, you can:
Break down lane rates into contract and spot over a set period of time.
Further break down the market contract rate to see what the rate was for certain types of loads (FTL, LTL), trucks (Dry Van, Reefer, Flatbed), etc.
You can ask questions like:
How did contract rates differ for this lane by different shippers
2. Diagnostic: Why did this happen?
Once you figure out what happened, you want to find out why.
Did something cause past behavior?
Can you change it for the better?
Understanding the underlying issues from the past helps you address them proactively for the future.
Lane rates can vary widely based on a number of factors including distance, equipment type, lane characteristics, macro-economic conditions at the origin and destination, the difference between forehaul vs. backhaul rates, etc.
With data science, you can:
understand how lane works
estimate a contract rates
monetize the impact of different factors
gain new insights to adjust your approaches in the future
You ask questions like:
Why do these rates differ so much
The truckload lanehaul cost is a function of many variables like:
Equipment type (Reefer, Dry Van, Flatbed)
Distance
Macro-economic conditions at origin and destinations
Lane from Chicago-Miami is more expensive than from Miami-Chicago
Lane characteristics
A lane with steady-state freight vs. lane with high variability
Shipper behavior
If you have a bad payment term or dwell time, you are likely to pay higher rates
Why was the sky blue lane paying higher contract rates?
Maybe that shipper had a bad payment term or dwell time
Inbound vs. Outbound Regional values
Size: # of inbounds or outbounds
Color: The bluer, the more inbounds to the region. The redder, the more outbounds in the region.
The region in LA is red because there is a lot of stuff that comes out of the port and get distributed across the states.
Regional values are stable over time -- very little relative change
useful in network design and facility location
This is very similar to what I did for a side project on Seoul bike trips to cluster counties by their inbound-outbound ratio by time!
Segmenting lanes by demand behavior
Demand Cadence
frequency of demand by weeks
Demand variability
how variable is demand when it materializes
Spot / Dynamic Rates
fluctuating
Price is determined at the time of the transaction like Spot rates
Recently, more shippers are starting to move into dynamic procurement
It's up to business strategy whether you want to procure spot/dynamic rates to give yourself the flexibility to go with the market rate or push everything to contract as much as possible and only use dynamic only when things fail
Data science allows you to separate those by analyzing characteristics for a specific lane
3. Predictive: What is most likely to happen in the future?
With data science, we can:
forecast contract or spot rates by including potential explanatory variables
shift from lane-level forecasts to shipment-specific forecasts, giving us more control with a specific load
e.g.) Predict churn (bounce load)
Customer Churn: Customer ends relationship with the shipper
Carrier Churn: Carrier does not show up for pickup ==> Refer to the bounce project I did
Can we identify customer/carrier churn ahead of time?
Combined with human judgment, data science can help us identify customers and carriers at risk of churn (likely to bounce), so they can then take proactive steps to keep them engaged.
4. Prescriptive: What actions should be taken?
Once the data has been analyzed and data science has offered predictions for the future, it can offer some possible actions to maximize positive outcomes. This is where human judgment comes into play.
Bounce Load Project can be utilized to provide warnings for loads that are likely to bounce, so that sales reps can take proactive actions (checking with high-risk carriers more often via phone call) on those loads. It's not taking away decisions from humans but providing extra information to them to make more informed decisions.
While predictive models may identify customers and carriers at greatest risk of churn and even offer some possible actions to take, it’s up to us to decide:
1) if we'd rather have false positives or false negatives mixed in with that data
2) what specific intervention to take with each customer or carrier.
Reference
https://www.youtube.com/watch?v=XrwzEX8UHPI
Comments