June 23, 2003
Ron Antinoja, Part I
Ron Antinoja founded and runs a service called Tendu at www.tendu.net. The firm--so named for teams' desire to understand player and coaching tendencies to do certain things in certain situations--expects to track the velocity, location and result of about 90% of all major league pitches this season. Tendu tracks those pitches and their outcomes and stores that information in an Internet database that allows users to discover pitcher and hitter tendencies in any given situation. Antinoja recently chatted with BP about teams' neverending quest to gain the upper hand on the opponent.
Baseball Prospectus: What was the genesis of Tendu?
Ron Antinoja: It came into being because I have a background in software engineering. Prior to that, I had a mediocre baseball career through high school. I was never never good enough to be a star, but I felt like I knew how to win. I became a software engineer, but I was dreaming all along that one day I would do something involving baseball.
One way I was driving along the freeway on my way to a consulting assignment, and I just snapped. I turned off to go to Stanford University, went straight to the baseball field to talk to their coach, Dean Stotz. I described what I wanted to do--it had to do with collecting data, studying tendencies of players and coaches. I believed that if people looked closely at pitches, they could see tendencies. Human beings repeat patterns, they go with what works. When they find something that works, they repeat it; when something doesn't work, they avoid it--it's an underlying philosophy of human behavior. That's the underlying philosophy of my software. If a person had information on what a player does in different situations, he can make a good deduction as to what they might do in the current situation.
BP: How did you apply your software background toward actually getting the company started then?
RA: First we had to create tools that were powerful and flexible for collecting all the data. After we had the data collection tools in place, we started building in the analysis aspects. The key was to isolate game contexts. It all comes back to being in a given context, a human being will think about what's worked in the past in this context. That includes who's on base, how many outs there are, the size of the lead or deficit, how many pitches he's thrown, etc. I never believed I had so much knowledge that I could figure out what would be the right contextual factors to look at in any given moment. So I built tools that could be used by baseball experts, so that they could use their expertise to think of as many scenarios as possible--they look at the situations that they think are important.
BP: How do you collect the pitch data you need? Who does it, and where do you find the people who do it?
RA: It started as word of mouth, but we've also done some advertising. When people come to us, there's an interview process they have to be go through, where they sit in front of a TV with an expert. We'll pick out pitches, and ask them 'what do you see?' If they don't know what they're looking at, they don't work for us. If they do, we ask 'why do you see that?' Do they know about grips, catcher signals, follow-throughs, catcher setups, all factors we need to track if we're to correctly identify every pitch that gets thrown. They don't even get to try out if they don't know these kinds of answers.
BP: So you're tracking pitches watching game tapes instead of in person. Why take that approach?
RA: Everyone does it off of TiVo, not live at games. It works better this way--it's a very difficult job, one that requires a lot of focus and patience. We'll stop and rewind every pitch. The more people do it, the better they get, the less they have to stop and rewind. But after they're done we'll still compare our results to the game box score of at espn.com or mlb.com.
BP: It's still got to be tough working off of TV broadcasts, where announcers can misindentify pitches, or they come back late from a commercial break and the inning's already started...
RA: That's what drives us really crazy, when they miss a pitch by coming back late from a commercial or during replays. Hopefully one day we'll be able to get more video straight from stadiums--it's already happening, just not enough yet. On average we might miss a couple pitches a game. Of course our goal is to chart 100% of all pitches in a major league season. Last year we did 80%, this year we're looking at something higher, around 90% hopefully.
BP: OK, so what do you do with that kind of information to transform it into something a major league baseball team will want and pay money to get?
RA: Once we've compiled the data by tracking pitch-by-pitch results, we use our analysis tools to make the data usable. Tendu has three analysis tools at the present time. There are more on the drawing board. We have the Pitching Chart tool, the Hitting Spray tool and the Pitch Sequence Analyzer.
The Pitching Chart tool is the one that has the red-yellow-green zones that are sometimes called hot and cold zones by other people. We designed this tool as a pitch effectiveness measure. To measure effectiveness, we use the at-bat outcome, which indicates whether the hitter got on base and how many bases did he get. The measure of effectiveness can use any of four well-known baseball algorithms:
The Pitching Chart tool can be used from either the pitcher's perspective (pitches thrown by a pitcher) or the hitter's perspective (pitches thrown to a hitter).
BP: What about the Hitting Spray tool? Aren't there already other services out there that track where a hitter hits the ball against certain pitchers or in certain situations?
RA: Another way of looking at your question is to compare the style of baseball game preparation in the past, to what Tendu's tools offer. In the past, analysis was based on a set of canned reports. For example, a hitting spray showing the location of the batted ball for each time that a hitter put the ball into play. There might be more than one hitting spray for a hitter, with one or more control parameters changed.
One of my competitors delivers four hitting sprays for example. Those are:
They claim that these four situations are the most valuable and have been specified by years of interactions with major league teams. I don't question the value of those reports. But I believe that there are more than four hitting sprays that could have value. What I have seen is that for any given player that is being studied, if I use Tendu's Hitting Spray tool, I can view five, 10, 20 hitting sprays in a few minutes, changing control values for each one, and I may find an unusual pattern of batted ball locations.
We call these unusual patterns red flags. A red flag is a tendency that may be exploited. For example, if a hitter hits a lot of non-fastballs low and away to the opposite field, the coaching staff can position its fielders in such a way as to give better coverage, when they know that the hitter is going to get mostly non-fastballs low and away.
Tendu's tools have been designed explicitly for exploration. If a user only wants to look at a specific set of canned reports, then he will only be able to find red flags that are clearly identified right on those reports. Our philosophy is that by exploring the data, by using a baseball person's natural instincts to change the control parameters to produce spray charts with different sets of data, a smart user can find more red flags that can be useful during the upcoming game.
BP: So it sounds like the Pitching Chart and Hitting Spray tools are supposed to improve on what's already out there. What then is a Pitch Sequence Analyzer and what does it do?
RA: The purpose of the Pitch Sequence Analyzer is to enable a user to discover tendencies in the form of patterns and frequencies hidden in a pitcher's sequence of pitches. You want to be able to do several things: determine how frequently a pattern is used by a pitcher; search the database for each occurrence of a pattern; predict the next pitch in a sequence from the frequency of pitches that follow the pattern; discover a pitcher's "out" pitch and learn how the pitcher sets up his "out" pitch; and find red flags in the sequences of pitches.
Users can browse the database of pitches, which are organized as groups of pitches from an at-bat. You can look at the frequencies of seven characteristics of a pitch for all pitches that a pitcher has thrown or only those pitches that match a pattern. Those seven characteristics are:
You can define patterns to search for. You can apply the Pitch Sequence Analyzer to all of the pitches that a pitcher has thrown during a given year, or all the years that Tendu has collected data (2002 and 2003) to find tendencies that are there for all game situations. Or, a user can load only subsets of the data for analysis, comparing the results to determine if a pitcher's tendencies are different for the different situations. You can compare early innings to late innings to see if a pitcher shows different tendencies. You can look at the lead size, the number of outs, or the runners on base, to see if those factors change a pitcher's tendencies.