Arccos thinks it has figured out how to turn golf statistics into a machine learning-powered Caddie feature that can improve your game. So I decided to take the Connecticut-based startup’s system of sensors and services for a whirl to see if it could save my golf game.
See, I’m not a good golfer. I like hitting the driving range for a little stress relief every now and then, but a journalist’s schedule combined with a journalist’s salary means that I very rarely end up on the course. Yet I’m also a huge sports stats nerd, and Arccos offered to give me a ton of data about how I played the game, with the hope that it could help me get better.
Caddie is designed to take that data, feed it into a machine learning model and then suggest the best course of action when tackling a hole. The feature is designed to provide users with assistance that would traditionally come from a human caddie. According to Arccos CEO Sal Syed, the nature of golf makes this a particularly useful assist.
“Before every shot, you’re spending 45 seconds to a minute making a decision that generally is biased about your own views about your game, and your own bias, optimism, frustration, and whatever,” he said.
In theory, the assistant should help people perform better on the course by mitigating some of those biases. From what I’ve seen, it’s not going to save the game of someone who’s truly abysmal, like me, but for golfers who are slightly below average (or better), the recommendations will likely give them a boost.
I didn’t get to try Caddie out in full, since users need to track five rounds of golf using Arccos sensors before it will start generating recommendations. (That’s because the machine learning system requires enough data about a user’s golf game to start generating solid advice, according to Syed.)
I did have a chance to look at how Tom Williams, Arccos’ senior vice president of sales and marketing, got recommendations from Caddie on his phone when I joined him to play through the Presidio Golf Course’s front nine on Monday morning in San Francisco. Here’s how it went.
On the links
Williams showed up for our 7 a.m. tee time with a package of Arccos 360 sensors, which make up the backbone of the company’s tracking system. The sensors (there are 14 in a package) screw into a hole at the end of every golf club’s grip. The hole is needed to let air escape while the club is being assembled, Williams explained, so Arccos can take advantage of that for its sensors.
Arccos’s sensors were a perfect fit for the decades-old clubs that my grandfather handed down to me, so they’ll likely fit any set purchased recently. While my clubs technically got slightly longer as a result, I wasn’t able to notice any difference in my swing, or in the weight of my clubs.
Next, each club needs to be paired with the Arccos Golf app so that the system knows when a user is swinging a particular club. It took about a minute to work through all of the sensors with Williams’s assistance, and the process was surprisingly straightforward.
Once the hardware was set up, I just had to tell the Arccos Golf app to start tracking my round, and slide my iPhone into my front pants pocket. The system uses high-frequency audio to send signals from the sensors to the phone for processing, which means users have to keep their phones in their front pockets while swinging, which some may find uncomfortable. (I didn’t.)
Trackers set up, we were off to the races. Or rather, Williams was off to the races, and I was off on an adventure. I managed to slice my ball out of bounds two strokes in, and the rest of my game was a comedy of errors. Case in point: I lost a pair of balls in a pen full of goats doing brush control on the course.
Arccos happily kept up with most of my abysmal strokes, though. The system seems to be fairly aggressive in trying to cut out false positive swings, which means that you need to tell the app if you hit a ball out of bounds or into a water hazard. I didn’t mind getting a few free strokes, but if you want to get the best advice, you’re probably best off tweaking it manually.
The Caddie feature provided Williams with a plan for how to address each hole. While it won’t yet update its recommendations over the course of a hole, it provides a template for addressing each one off the tee to help eliminate guesswork about club choice.
For example, Williams would have ordinarily used a pitching wedge off the tee on the fourth hole, a short, downhill par 3. Following Caddie’s advice, he used a nine iron to get a bit more distance, and landed on the green within easy putting distance of the hole.
I’m not good enough to get the most out of Caddie, since hitting a shot straight off the tee is typically a bridge too far for me. But I could see it being useful if I actually corrected my slice to the point that other deficiencies in my game became more pressing problems.
How it all works
Arccos’ team takes data from all golfers using its sensors (more than 70 million shots hit to date) and then uses that to build a model of individual golfers’ performances, using machine learning.
The company relies on Microsoft’s Azure Machine Learning service to handle training and deployment of the algorithms it uses to power Caddie and other services. And Arccos says it is always searching for better algorithms that can offer improved suggestions.
“To test the models, we’ve developed a test harness that runs against hundreds of thousands of holes of golf for users at varying handicap levels and tee shot distances,” Colin Phillips, Arccos’ senior vice president of software development, said in an email. “For any hole, we can see whether a user went with the advice given by Arccos Caddie or opted against it. We can then aggregate performance with and without Arccos Caddie advice to understand its exact benefits.”
What’s next for Arccos
While sales of the $250 bundle of sensors is the company’s primary revenue stream at the moment, Syed said Arccos’ long-term plan is to work with manufacturers to build the sensors into every new club that golfers buy. Arccos has already started down that path with Cobra, which has integrated sensors into its standard line of drivers for 2017.
Arccos Caddie is currently free, while it is still in beta testing, but the company is looking to make it a paid subscription once it’s ready for prime time.
There’s a lot more than just shot tracking in Arccos’s future, as well. The company has over 368 million geotagged data points about where users are making shots on more than 40,000 golf courses. Syed pointed out that the company could use that information to help golf course operators make better decisions about where to focus their maintenance efforts, as well as providing insight into how changes to a course would affect play.
Update June 19, 2017: An Arccos spokesperson explained that its sensors will come pre-installed in all of Cobra’s three standard driver models for 2017, which makes up the overwhelming majority of what the club maker will sell. The article has been updated to clarify that.