Last month the tech world oohed and aahed over a new Facebook virtual assistant service called M. Like Siri on steroids, M can help you plan a birthday party or wait while the cable company puts you on hold. It uses a mix of artificial intelligence and Facebook-employed human “trainers.” And like the on-demand app Magic, it promises a magical experience.
But when evaluating the latest shiny object to come out of 1 Hacker Way, it’s important to consider one thing: Facebook launches a lot of things that utterly flop. It launches a lot of things that become blockbuster success stories, like its video player or its Facebook Messenger app, but it also has built a lot of duds. M could be forgotten in a matter of weeks.
Such was the fate of Poke, Facebook’s Snapchat knock-off. And Slingshot, Facebook’s second Snapchat knock-off. (And after that, its reboot of Slingshot.) Same goes for the much-hyped Rooms, a collaboration app. And don’t forget Paper, the news reading app which quickly fell from the top of the App Store charts. Or Riff, the collaborative video sharing app, or Gifts, its ill-fated foray into commerce.
But the most relevant flop to consider when evaluating Facebook M is Graph Search. Facebook (fb) launched it in 2013, and it was hailed as “amazing” and a Yelp-killer. A Wired writer declared that Graph Search “made him a Facebook addict.” Except for most people, it didn’t work. CEO Mark Zuckerberg even admitted as much, telling Bloomberg Businessweek in 2014 that it would be “generous” to say Graph Search worked half the time. Eventually, the company dropped the “graph” part and retreated from its grand plans to make everything on its network searchable, noting that “search at Facebook is a long-term effort.”
The Graph Search stumble is important, because it was Facebook’s big attempt to dethrone Google’s most profitable business, search advertisements.
M is its latest. On the surface, it appears that M competes with Apple’s Siri and Microsoft’s Cortana. But those virtual assistants aren’t attached to advertising businesses and they’re not money-makers for their parent companies. The rhetoric around M’s launch made it clear that Facebook’s virtual assistants are very advertiser friendly: Facebook vice president Davis Marcus told Wired quite directly that M is “capturing all of your intent for the things you want to do.” He explained further: “Intent often leads to buying something, or to a transaction, and that’s an opportunity for us to [make money] over time.”
The I-word—intent—is key. It’s the bottom of the so-called marketing funnel, and it’s a really valuable place to be. If you know a customer intends to rent, say, a hotel room in Tupelo, hoteliers in that Mississippi town are willing to part with a lot of money to get in front of that customer. Intent doesn’t require hundreds of millions of eyeballs, or hours of attention, or “engagement,” or “brand awareness,” or any flavor of advertising fluff. It is straightforward, transactional and efficient. It’s why $66 billion-a-year Google (or, really, Alphabet) can enjoy healthy 24% profit margins while investing in self-driving cars, Internet balloons, and life-extending biotechnology.
But Google’s (goog) money-printing machine has come under threat with the shift to mobile, where people search for things inside discrete apps and not in Web browsers. (You might instead find that Tupelo hotel room using your Kayak app, where booking is easier than in a mobile Web browser because you’ve already entered your credit card information.)
Facebook’s M, with its aspirations of planning birthday parties and delivering groceries, is a shot fired in the battle for mobile intent. The company wants Facebook messenger, where M will live, to become your go-to app for everything.
This battle is worth watching closely, because Facebook is already winning the mobile advertising game. The company commanded 36% of mobile ad spending last year, while Google grabbed just 11% of it, according to eMarketer. Even though “mobile jitters” played a big role in Facebook’s IPO, those days are long gone. Facebook’s transition from a company that makes no money from mobile to a company that makes 76% of ad revenue from mobile was quick and painless.
Its latest bid is a big one. Moreso than many of Facebook’s past launches, M is not a cheap experiment built in a hackathon and thrown into the App Store for fun. Facebook M required investment in a team of human “trainers” who manage user requests. There’s nothing a Silicon Valley tech company dislikes more than throwing expensive, imperfect, irrational, unpredictable humans at a problem. That’s how big a payoff Facebook is expecting from M. (By contrast, Google Now, its own artificially intelligent personal assistant, is all algorithm.)
If M flops, it won’t be Facebook’s last stab at mobile intent. The stakes are high, and with just $9 billion in annual spending, the money is only beginning to migrate from desktop and TV advertising onto our phones. The fight for mobile dominance is just getting started.
Sign up for Data Sheet, Fortune’s daily newsletter about the business of technology.