Using Predictive Modeling to Guide Enforcement in Berkeley, CA
Berkeley and Turnstone first partnered to optimize parking the way most cities start: by pricing to demand. In the first six months, the city adjusted rates twice across its paid curb. But the same analysis that informed those rate decisions surfaced a challenge that pricing alone could not solve. A large share of curb use went unpaid, and no rate change reaches a vehicle that never pays. The rate work revealed a new lever: enforcement.
Berkeley's curb was busier than its payment data implied. When Turnstone modeled actual occupancy against paid occupancy, the gap was substantial: across the city, a significant portion of curb use went unpaid, from expired sessions to vehicles that never paid at all. Rate adjustments could shift the behavior of people who pay, but they had no effect on the share of the curb that was not paying in the first place.
The harder question was where to act. Enforcement spread evenly across 37 Curb Areas is expensive and mostly wasted, because noncompliance is not evenly spread. Without knowing which Curb Zones actually drove the problem, the city would be patrolling on instinct.
Turnstone built Berkeley's view of the curb from the city's own payment, permit, and LPR data, with no new sensors in the ground. The model estimates actual occupancy on every Curb Zone and compares it against paid occupancy, so the unpaid share is visible block by block rather than buried in a citywide average. That comparison did two things at once. It gave the city the demand picture it needed to set rates, and it exposed how concentrated the unpaid usage really was. Roughly one-fifth of Berkeley's Curb Zones accounted for more than half of all noncompliance. That finding turned enforcement from a citywide guess into a guided approach: direct officers to the specific blocks where unpaid use is concentrated, and leave the compliant majority alone. Enforcement and rates complement one another, because the data showed pricing could not solve the problem on its own.
“Turnstone gives us one place to see how the curb is actually behaving, so we can put enforcement and pricing where it will do the most good.”
The concentration is the finding that made enforcement actionable. The top fifth of Berkeley's Curb Zones drive more than half of all unpaid curb use, so directing enforcement to that fifth reaches the bulk of the problem at a fraction of the effort of citywide patrols. Alongside the enforcement work, the demand-based rate adjustments shifted curb behavior toward turnover: short stays up 13.6% and long stays down 26.1%. The curb now serves roughly 4,000 vehicles a day.
The enforcement payoff itself is still being measured. Berkeley recently began directing resources to the highest-noncompliance Curb Zones, and the citywide effect on noncompliance is a result the city is tracking, not yet a settled outcome. This case reflects the program as it stands: the guided approach is proven, and the reduction is in progress.
Berkeley set out to price its curb and discovered something more useful: where the curb was not paying at all, and what to do about it. The lesson is not unique to Berkeley. The same data that sets rates can show a city exactly where enforcement will earn its keep, so resources go to the blocks that need them instead of being spread thin across the ones that do not.
