Algorithm to Assign Work to Truck Drivers based on Bidding Data

Project Description

Project Overview:

We are looking for assistance in developing an algorithm, which will assign work to drivers in a competitive bid environment.  Drivers will have the ability to indicate their interest in working specific projects on a daily basis.  Drivers may be interested in bidding on multiple opportunities on a daily basis, but will only be able to work on one project at a time (if the project is expected to last an entire work day); drivers may be able to work several separate "smaller" projects on a daily basis. 

Company Profile:

We are a local DFW, TX sand and gravel-trucking broker primarily serving the construction industry. We do not own trucks nor employ drivers. We work only with owner/operators.  

Data Source:

A MySQL database hosted on AWS contains all the bid information and relevant historical data.

Deliverable:

An algorithm that integrates with our existing platform and assigns work to drivers considering the following scenarios:

• No shows - % chance the driver “wins” a bid and does not show up to the complete the order
• There will be a certain percentage of drivers that will win a bid, but still won’t show up
• Specific drivers will be worse than others
• We can “over allocate” the # of drivers we assign to a specific order on each day based of historical “no show” averages and/or specific drivers that are assigned, but poor performers
• Over time we’ll want to eliminate these drivers from the system
• The larger the # of requested drivers (for an order by day) may allow us to better account for no shows
• Ex. if only one driver is requested, it’s a little more difficult to assign another driver just in case the original driver does not show up
• Ex.  if 20 drivers are requested, it’s easier to assign one or two extra drivers
• Timeliness is another thing to take into consideration.
• If a job starts at 7 AM, we want someone to be there at 7 AM – not 10 AM
• Maximizing revenue vs. Maximizing performance
• The lowest bid might not always be the most optimal outcome for completing the order
• It may be more beneficial to pay a slightly higher overall price to guarantee the driver will complete the job, considering the competitive marketplace
• This may prove seasonal, both from a monthly (summer = busy) or even daily (rain = slow) cycle and should be flexible to handle new MSAs (as we expand to new cities):
• Slow times = less competition for drivers from other trucking brokers
• Busy times = more competition for drivers from other trucking brokers
• A/B testing – Max Bid 
• For same day “bids” we will show the driver the maximum amount they can bid to be assigned the order.
• This “max bid” will equate to our current cost allocated for the trucker and is directly relatable to the another project we are working on to create customer-specific Pricing Algorithm Based On Historical Pricing Data.
• For orders tomorrow (and beyond) the maximum amount may or may not be shown to the driver
• We’ll be able to control this on a driver by driver basis to test how bidding is affected if the driver sees our max bid or not
• Est. Loads per Day – we will give the drivers an indication of the duration of orders, both in terms of A. expected time spent working per day and B. expected number of deliveries per day so that he may better frame his bid:
• Time – do we except for the drivers work day to be a quarter, half, or full day.  Work lasting an entire workday may be more attractive than work only lasting a few hours.
• Pay – given all variables (load time, drive time, dump time, traffic, time of day, etc.) how many loads can we except for that driver to complete during a work day
• Load time – will vary by customer's equipment (if a haul off) or pit (if a haul on)
• Drive time – highway vs. in city driving, toll roads used or avoided
• Dump time – will vary by customer's equipment (if a haul off) or dump (if a haul off)
• Time of day – the first load of the day / right after or during lunch may experience longer wait times compared to the average as truck bottlenecks occur
• Ex.  Lots of trucks will line up to get loaded for the first load of the day each morning.   If driver A is 20th in line, he’ll wait longer than if he had arrived sooner
• Traffic – construction on the roads?  Rush hour?

In your proposal please share more details on the solution, milestones and your previous experience in optimization algorithms for supply chain or other industries.