Jonathon Barkl discusses dynamic pricing experiments for parking marketplaces. He explains the objective function of optimizing long-term property value and revenue, the key metrics tracked like driver retention and parking duration, and the experimental design: 6-8 week tests with four treatment buckets at 25% allocation each, plus an 80/10/10 rollout to test against seasonality.
I have a question for you about what your kind of objective function in your dynamic pricing exercises, like what are you trying to maximize? Is it revenue? Is it some sort of utilization? Is it some sort of long-term revenue? So what are you, what are you kind of learning when you're wiggling the price around for, you know, 2 customers that look very much alike?
Yeah, what we're optimizing for is basically long-term value of the property for the owner, which then implies on our end the long-term revenue of the property and in the reason I say long-term value of the property for the owner is that you have to look at each property on an individual basis where if it's a property with parking, but it's also retail and office and residential, you can't just jack up the parking rates because then you're going to piss off the office tenants and the retail tenants and their customers and the residential tenants. So it's about for them looking at the holistic picture of how are we going to maximize the long-term revenue of this property and looking at every case on a case-by-case basis to do that. And the thing I always tell owners is the classic phrase of you don't want to cut off your nose to spite your face. So we don't want to jack up your parking rates 3x and then find out that all your tenants are now upset. So it's really about long-term value of the property. And in a property that is pure paid parking, there's no other thing attached to it. Long-term value of the property is still a good indicator of that because really for long-term revenue to be healthy there, you could, sure, you could triple rates in month 1 and then now you've increased your revenue by 3x for the first month, And then the question is, are people going to keep coming back, or are they going to be turned away because now your rates are not competitive for the area, or because you're just charging so much that people are disincentivized from parking there? So it really is about long-term revenue, and there's a lot of metrics that go into that about retention of drivers, the rates that are being paid, how long people are parking for, the enforcement rates, things like that that flow into that. But it really is about like long-term sustainability of the revenue that we're increasing, because we've run experiments before. Where to find the optimal price, you inherently have to go a little bit past where you think you should be. And then you start to see what is the point of diminishing returns on that price that you're now charging. And so, you know, we've seen in our experiments that you can see, hey, at this point, this price point, that is when you start to see all the downstream metrics take a negative hit, even though you are getting more revenue. So you need to dial it back to this range that is actually the optimal range.
But It sounds like all your experiments are sort of kind of short run and they're measuring like a short run price response. How do you get at the long run stuff?
Yeah, normally the experiments that we run are run for like 6 to 8 weeks at a time. So you'll see basically we'll have 4 different buckets and during those 4, those 6 to 8 weeks, 25% of drivers are getting put into each of those buckets. And then we're seeing over those 8 weeks, how are people responding? And the other thing I'll point out is that once you get put into a bucket, we then are like keeping you in that bucket. So you'll always get shown that rate. So over those 8 weeks or over the long term, we'll see which rates you were being shown and compare that versus the control group that exists. So we're always comparing against some sort of control group. So when we take over a parking facility, say they were charging $4 an hour before, up to $30 for the day, that's the control group. So 25% of people are going to get shown that rate And then 25% are going to get shown like a $5 rate that increases to $8 per hour with still that $30 per day cap. And then the next group is going to get shown $7 to $9, and the next group is going to get shown with different functions and things like that in terms of how they scale with occupancy. So then we're measuring over a decently long chunk of time, but then it's really like this ongoing experiment. So even after, out of those 4 rates, we pick a winning rate, we'll then run a next experiment for another 8 weeks. That basically is that experiment versus new experiments. And then once we kind of decide after a few of those experiments that this is kind of the optimal rate to be charging, even in the background at all times, there'll be, you know, 80% of drivers will get shown that new default rate, and then 10% and 10% will get shown 2 other rates that we're constantly testing against those in order to make sure that this is truly the optimal rate. Because, you know, seasonality is a thing and events are a thing and weather is a thing. So you need to always be testing your pricing. Which is something that's very new to the industry because most companies in the industry, it's just, let's pick a price and stick with it for a year and it never changes.
