A study in Transportation Research Part E quantifies something every delivery driver already knows: finding a place to park is often harder than finding a route. Giacomo Dalla Chiara and colleagues (University of Washington, University of Groningen) analyzed GPS data from a beverage distributor’s delivery operations in downtown Seattle (472 manifests, over 2,400 trips) to ask what happens to route planning once you stop assuming parking is instantly available.
The answer: quite a lot. Using a Time-Dependent Traveling Salesman Problem with Time Windows, the researchers compared “parking-blind” planning (optimizing on travel time alone, the industry default) against “parking-aware” planning, which folds in expected parking delays estimated from a regression model combining Google Maps travel-time predictions with curb-space data (loading zones, bus zones, no-parking areas) and time-of-day effects.
The gap is striking. When a parking-blind plan is confronted with real parking conditions, tour duration balloons by 52%, average lateness jumps from under 2 minutes to nearly 170 minutes per tour, and the share of tours running late rises from 7% to 86%. Parking delay alone ends up eating a third of total tour time — roughly matching earlier Seattle estimates (Dalla Chiara & Goodchild, 2020) but now shown to actively distort planning, not just execution.
Parking-aware planning doesn’t eliminate the problem, but it manages it well. By anticipating where and when curb space will be scarce. Parking delays peak in a double hump, around 7–8 am and again at noon–2 pm. The optimizer trims total tour duration by about 3.4% (17.4 minutes) on average, trading a small increase in driving time for a much larger cut in parking delay (–11%).
The bigger win is reliability: lateness drops by over 95%, and time-window violations fall by roughly 18 percentage points. Interestingly, the mechanism is modest: routes are barely restructured (median edge-overlap with the blind solution is only 20%, yet overall stop order correlation stays high), and most of the gain comes from re-timing (shifting departures earlier and nudging arrivals into lower-delay windows) rather than wholesale re-sequencing.
For practitioners, the takeaway is pragmatic: you don’t need a new optimization engine to capture this value, just better inputs. Historical GPS and curb-inventory data can be integrated into existing routing tools to adjust timing, yielding outsized returns in service reliability relative to the effort.
For cities, it strengthens the case for digitizing and publishing time-dependent curb information; not just adding capacity, but making existing capacity more predictable, since it’s the predictability of curb access, not just its availability, that carriers can actually plan around.
Supported by ClaudeAI.