Recommending future franchise locations

I grew up in Margao, Goa – a place known for beaches, greenery and a happy-go-lucky lifestyle. As kids, my friends and I loved junk food. Every summer we’d talk about how a friend of mine went to an aunt’s house in Mumbai or some other big city and how he’d eaten at Domino’s pizza or KFC while he was there. We’d always ask the same questions: “How was the food?” and “Is it better than pizzas here?”, and we’d always get the same answer: wide eyes and a hum of approval. More than that even, I think we liked the idea of being part of a big city. Being big time. Eating pizza at Domino’s was definitely worth bragging rights as a middle schooler in Goa.

Margao was a sleepy town. There were no Domino’s or McDonald’s or Inox movie halls. On some weekends, my friends and I would take an hour long bus to Panaji just to get to eat at Domino’s pizza and watch a movie at Inox. We would plan this trip for weeks and spend a lot of that time wishing that Domino’s would someday open in Margao. This got me thinking about ways by which I’d know when and where a new franchise would open. I needed to know when Domino’s was coming to town.

Bogmalo beach, Margao

A map of Margao

Now, 5 years later, I finally have the skills to try and do this. The next few blog posts are going to be a slight bit different; I’m going to be exploring the possibility of future franchises using recommender systems and R. I found a course offered by the University of Minnesota and this link quite helpful in understanding how recommender systems work. There’s plenty of recommender algorithms out there. I studied recommender systems and found that User-based collaborative filtering would best serve my purpose of, well, predicting the future.

User-based collaborative filtering basically means this: If you like pizza and I like pizza, we have similar tastes. If the both of us also like Thai food, we’re more similar. This means that if you like Chinese food, I’ll probably like it too. The more foods we like in common, the more likely it is that we have similar tastes. User-based collaborative filtering does exactly this: it predicts what I’ll like based on what similar users (you) have liked. This system is used by Amazon, Facebook, Twitter and many other heavyweights in the tech world.

Instead of using recommender systems for ratings and ‘you may also like’ products, I decided to use it to map companies and their store locations. To recommend to companies where they should open their next store. Stay tuned for what’s to come.

 

Does Uber charge more in bigger cities?

You’d pay about twice as much for an Uber ride in Nashik than in Ahmedabad. I’ve always written about franchises of companies that I see around – Inox, Cafe Coffee Day, Domino’s Pizza –and companies that intrigue me, like V Mart and Equitas Microfinance. Apart from two exceptions (Hero Motocorp and V Mart), the general ‘rule’ I’ve learnt is that more GDP equals more franchises. My friends in Pune say that Ubers are more expensive than auto rickshaws, but I’ve found that they’re priced about the same in Hyderabad. This got me curious and I decided to take a look at Uber pricing in different cities.

For a company like Uber which doesn’t have franchises, I looked around for a parallel to this ‘rule’. I couldn’t get my hands on the data of the number of cars they have per city, but I found what I think is a pretty close match: the amount they charge customers. Franchise or no franchise, as long as a service isn’t a luxury good, businesses should want to reduce prices for their poorer customers right? If they didn’t, people would just switch to booking other cabs. Or taking auto rickshaws.

So why isn’t this ‘more GDP equals more franchises’ rule true all the time? Well, the argument I make in the blog posts linked is that V-Mart and Hero Motocorp specifically cater to poorer rural markets. Does Uber cater to rural markets? Nope, not yet at least. There’s not one small town in Uber’s 27-city empire.

Uber’s absence in towns makes total sense. For a town to interest Uber, the people of the town would need smartphones with an active data connection to book a cab (that’s changing a little now) and there would need to be enough employable youth who own cars.

Here’s a graph I plotted that shows the lack of correlation of the base fare for Uber Go cabs on the GDP per capita in tens of millions of dollars. I found the price data on their website here. The correlation coefficient of this graph is 0.06; no correlation.

I thought that maybe base price isn’t a good parameter because some cities are larger than others, and smaller cities, on average, could charge higher base prices. That’s why I decided to take another parameter: cost per kilometer. This is the price you pay above the base price after a certain number of kilometers (generally 4 or so). I expected to find something that showed correlation – evidence that people in Nashik wouldn’t actually be paying twice as much as people in Ahmedabad – but I didn’t. I ran the correlation with cost in rupees per kilometer instead and got an R value of 0.06 again.

So Uber is a company that doesn’t seem to charge it’s richer customers more – something that’s pretty unusual. On average, maybe Uber still does charge customers from bigger cities more, but it’s not something that increases progressively with GDP. The only other times I’ve seen this pattern are with companies that cater to rural markets. I want to find out if this is exclusively an Uber thing, and to do that I’m going to need to take a look at Ola cabs, Uber’s closest competitor in India, and see what’s up. That’s what I plan to do next week.