Case Study – Parking at the Universal Studios Lot
Challenge: Oversubscribed parking on the Universal Studios Lot, LA, CA.
NBCU is located in Universal City, Los Angeles, CA. It is the home of the main West Coast offices for many NBCU business units, the production lot (sound stages, costume department, transportation, props, lighting, production offices, post-production facility, etc.), as well as the Universal Studios Theme Park. The theme park operates entirely separately from the production lot.
When Comcast bought NBCU, one of their first orders of business was to consolidate all of their facilities in the Los Angeles area onto the Universal Lot. That required a great deal of office reconfiguration Facilities were able to find office space for all Comcast/NBCU employees. What they didn’t consider was parking.
It is possible to shrink cubicles and offices, but there are regulations on parking space size they could not avoid. It became a huge problem.
The initial, brute force solution was hiring a company to provide “valet assist” services (which I will refer to as “valet”). When the parking spaces for any lot were filled (there were about a dozen parking lots on site, with three main large ones, valet personnel would guide arrivals to park behind the other cars and leave their keys. If someone blocked in needed to move, they would move the blocking car(s) and then put them back, so the owner knew where the car was when they needed to leave.
They fully staffed every lot every day to handle a peak load. Of course, each lot didn’t have a peak load each day so they were massively overstaffed, but couldn’t afford for any of the lots to be understaffed.
It cost NBCU $4M the first year, so they asked our team to find a better solution and I ended up with the project.
Over the first few months, I had the valet company provide me with a reasonably accurate estimate of the number of cars arriving in each lot each day, allowing me to start building a model. I quickly realized that there were three populations on the lot:
- Regular employees – a virtually steady and predictable number of cars daily
- Visitors – a relatively easy population to predict with daily averages (Mondays and Fridays were light with the middle of the week heavier) and some seasonality.
- Production personnel – the people who worked on the sound stages. This population varied wildly depending on the status of a production.
Once I’d gathered enough data the first two populations were predictable within an acceptable range. The production population was impossible to model, but reasonably accurate data was available, on paper call sheets. Call sheets were created every day for the following day for every production on the lot and included the number of expected cars and to which lot they were assigned. Also, each film production and TV season created an initial full schedule, which could be used for estimates. Full schedules were created at the beginning of every production, which were helpful, but often changed on a daily basis.
I also had to account for seasonality and the occasional anomaly, such as an irregular event on the lot or a holiday, which needed to be filtered out of the data.
Long story short, by introducing digital call sheets the data from which could be fed directly into the model (including the initial full production schedules), I was able to create a model to predict the number of cars coming to each lot each day, and therefore we were able staff accordingly. Once the model had been trained, with a potential 8,000 vehicles arriving at the lot each day, the model was able to predict the next day’s parking volume to within 30 vehicles.
In year two, valet costs were reduced to $1.5M, saving NBCU $2.5M annually.