Delivery windows
A promised two-hour window is only as good as the drive-time estimate behind it. Rush hour belongs in the plan, not in the excuse.
Case study · Routing
Closer isn't always quicker.
A case study from my delivery and field-service routing work at Lowe's and Redbox, rebuilt with public data: how traffic around Washington, DC and Richmond rises and falls through the week, and why the "nearest" stop depends on how you measure: a straight line, the road, or the clock.
The story
Miles aren't minutes.
At Lowe's and Redbox I worked on delivery and routing optimization, and two ideas drove most of it.
Use real road distance, not a straight line. Straight-line (Haversine) distance is quick to compute, but a river or a missing bridge can turn a short hop into a long drive.
Add historical traffic. Being 20 miles closer doesn't always mean quicker once time of day and traffic are factored in.
At Lowe's it was delivery, overwhelmingly appliance deliveries. Redbox was far more than DVD kiosks: ecoATMs, Pokémon card vending machines, charging stations, and more. Merchandisers restocked the machines and service techs repaired or serviced them. Tech routing needed more work, because it factored in what each tech was working on. A screen replacement takes longer than a regular cleaning, and a long job pushes every later drive into a different hour of traffic. Try the service tech mode in the demo below to see it.
Part 1 · Real counts
When the roads fill up.
Average vehicles per hour on major corridors, by day of week and hour, from 2025 roadside counters. Drag the slider or press play.
Each strip is one direction of one corridor: 7 rows (Sun to Sat) by 24 hours, shaded against that corridor's busiest hour. Volume is not speed: a jammed freeway can carry fewer cars per hour than a busy, moving one, which is why some peaks look flat. Powhite Parkway is missing because there was no continuous counter on it in the 2025 federal data.
Part 2 · Real routes
Which stop is nearest?
One start point and a handful of public landmarks. Switch how distance is measured and watch the ranking change, or plan a service tech's day.
Illustrative
Straight line is Haversine distance. Road miles and drive time come from OSRM routes on OpenStreetMap roads, precomputed and stored with the site. OSRM drive times assume free-flowing traffic. Illustrative: drive times by departure hour are estimated from real 2025 TMAS weekday counts via a standard volume-delay formula with assumed parameters, not measured speeds. Service tech mode uses example job durations, not real data, tries every visit order for one tech, and ends the day at the last job.
Beyond the map
Why this matters for operations.
The same two ideas, real road distance and time of day, show up anywhere people or trucks have to be somewhere.
Field service
Techs aren't interchangeable stops. A long repair pushes every later visit into a different hour of traffic, so the job type should shape the route.
Restock routes
Merchandisers visit the same machines on a cycle. Small, repeatable savings per stop add up when the route runs every week.
Territory design
Drawing service areas by straight-line radius splits neighborhoods that a river or highway already divides. Drive-time areas follow the roads.
Honest ETAs
Customers notice when "15 minutes away" turns into 40. Time-of-day aware estimates set expectations people can trust.
Better questions
Once distance and traffic are modeled well, you can ask the real question: not "who is closest?" but "who can get there soonest?"
Data: traffic volumes from the FHWA Travel Monitoring Analysis System (VDOT continuous count stations, 2025, all months). Routes from OSRM on OpenStreetMap data; landmark coordinates from OpenStreetMap Nominatim. Map tiles © OpenStreetMap contributors; map by Leaflet. Built by me with AI. Updated Oct 2026.
Talk shop