Craigslist, Freshness, Bait and Switch Problems
When Lawrence and I gave our original Demo Day pitch in the Summer 2009 Y Combinator batch, we cited Craigslist as a large part of the NYC apartment finder problem. Back then, you could not sort by price, location, or photo gallery. The only thing you could do was “grep for text” that suited you (like the word No Fee or Luxury Highrise). Many predicted Craigslist would not last so long, while others, including Harvard Business School professors Peter Coles and Ben Edelman argued the network effect would ensure their survival.
But the one thing Craigslist got right was their focus on Freshness. They mimic a real-life bulletin board, because you can’t sort by anything reasonable (price, distance from a landmark, etc). As new postings come in, they obstruct older postings. Everyone is welcome to continue browsing the older items, to a point, but savvy renters know that anything old is probably already gone – otherwise the agent would have been reposted it again.
The MOST important component in our HopScore ranking system is the Freshness. Given how quickly the market moves, and how quickly inventory can come off the market or change in price, Freshness measures how likely the listing data is still accurate. Our algorithms can see how many other renters have already inquired on a particular listing, when the agent or landlord most recently updated the listing, and whether the account has a track record of accuracy.
While designing a sorting algorithm for apartment listings, there is a counter-intuitive logic we face. We can use other factors such as Quality and Manager scores to determine whether an apartment will be popular. However, after crossing a critical threshold, the MORE people we have seen click on and inquire about a listing, the LESS likely we should recommend it to others. Either the apartment is no longer available, or there is something wrong with it after 20 renters have scheduled showings and none have taken it.