Artificial IntelligenceCryptocurrencyMetaverseCybersecurityTech Forward

Uber Burned a Whole Lot of Money to Make UberPool Work

May 31, 2017, 7:21 PM UTC

Bleeding cash is an understatement.

In 2015, Uber burned a whopping $1 million per week in San Francisco alone at times on its UberPool carpooling service in order to compete with rival Lyft, according to internal documents obtained by BuzzFeed.

The money was spent on subsidizing rides to try to attract customers with low fares. As the company rolled back the discounts around the end of 2015, it reportedly lost 26% of its riders to Lyft.

An Uber spokesperson would not comment on the figures but told Fortune that “as a new product, we offer discounts to encourage riders to try it out.”

UberPool debuted in San Francisco in 2014 as a carpool for riders who are going in a similar direction to be matched together in one car. It made rides more affordable for passengers and the routes more efficient for drivers.

Because Uber’s algorithm wasn’t very sophisticated initially, the match rate was low, according to BuzzFeed, and many passengers ended up getting private rides at discounted rates. As a result, the service bled cash.

For example, during one week, Uber offered carpool rides for only $5.

Get Data Sheet, Fortune’s technology newsletter

By June of 2015, the tech behemoth had reportedly spent $6 million per month just on UberPool in San Francisco.

“Every time we did something to reduce burn, our ridership went down,” a former Uber employee told BuzzFeed News. “The riders were just so used to getting discounts.”

To help make the numbers work, Uber has created partnerships with transit agencies and local governments to subsidize rides for residents, although the financial benefit is unclear. UberPool is currently available in 34 cities and 12 countries.

Uber, which is valued at $69 billion, continues to make changes to its carpooling service to keep up with its rivals’ pricing. Just last week, Uber debuted a “smarter” UberPool service in Manhattan, meaning that it tweaked the algorithm to better match riders and drivers for an optimal route.