Here's What Happens When You Apply Data Science to Dating

I was putting on my favorite lacy red dress when I got the text. He wasn’t coming.

Correction—he, my date for the evening, a smart and funny writer, was coming, but he was going to have dinner with his college friends first, before driving the two hours to Manhattan to see me . . . also for dinner. The same dinner for which he’d told me to find someplace “delicious” and “quiet” where we could have a “nice conversation.” Now he’d get here at 10:00 p.m.—maybe.

The day before, I’d spent 30 minutes optimizing the location for what would be our second date: elegant but not fancy, romantic but not cheesy, intimate but not too sexy. I had canceled plans with a girlfriend in order to make this happen. (I know. I felt terrible about it, but he was in town visiting from Los Angeles, and I’d felt such potential.)

The worst part? Apparently, he didn’t feel bad. No apology. I sent my girlfriend a screenshot. “I have to end this, right?”

It didn’t really matter what she said back—I couldn’t invest in something where I felt such disrespect for my time. Immediately, I composed a text that began, “James, I really like you but I don’t think we should continue seeing each other,” and ended with, “Good luck with everything!” I canceled our 8:00 p.m. reservation and ordered delivery strawberry rhubarb pie for dinner.

A few days later, flowers showed up at my apartment. On the card: “The future is uncertain, but this uncertainty is at the very heart of human creativity.” (That’s Nobel laureate Ilya Prigogine. I know.)

“You have no way of knowing,” he said on the phone, the first call after texting for weeks, “but this is not how I am normally. Will you give me another chance?”

I was a math major in college, so I tend to see patterns everywhere I look. He was 12 minutes late to our first date and, if it had happened, would have been over 120 minutes late to the second: not a good pattern. That said, he was right—I had no way of knowing if this was a normal pattern or merely a sampling error. What if this was a two sigma (translation: about 5 percent likelihood) event, and it had just happened to occur on our second date?

Dating, I have always believed, is at its essence, all about data: You spend time together. You use that experience as a sort of data collection. You build data sets, you analyze them, and you make decisions. I didn’t have enough data on James to make a confident decision. So, I thought, “Why not?”

The first time we saw each other after the no-show, he showed up at my apartment 30 minutes early. I’d just showered. He poured a glass of Bordeaux, sat in my bathroom, and watched me put on makeup. (Yes, we made it to dinner.) At the restaurant, our waitress invited us to the hotel guest–only Library Room for dessert, on the couch, because apparently, the way we smiled at each other made it seem like we were from out of town. We fed each other gelato and made out.

The next weekend, he took red-eyes to and from New York to spend Saturday with me, from 7:00 a.m. to 9:00 p.m., before getting back home for Father’s Day with his two young kids, who live in L.A. with his ex-wife. I made brunch, which meant I mashed some avocado and smeared it on toast. We took a two-hour bubble bath to digest the food we didn’t eat, listening to the xx and wrapping therapy around each other à la Julia Roberts and Richard Gere in Pretty Woman. On his way out, he left a tiny pillow in the shape of California on my bed.

This, in case you’re wondering, is a great reversal of the previous trend. But any good data scientist knows that you should always make sure your data quality is high, which means collecting from as many sources as possible. Bonus points for low correlation—you need unrelated sources to build the best signal.

I met a cluster of his friends when he invited me to visit his turf a couple weekends later. He hadn’t seen some of the people in decades. They laughed and made fun of each other with love and abandon. James held my hand the entire time.

The next day, I met his kids—against court orders. (As part of the divorce, he wasn’t supposed to introduce them to anyone he had been dating for less than six months; our first date was seven weeks prior.) I taught his son, age 7, permutations using scones at the bakery. His daughter, 4, whispered secrets in my ear, like how she had seen an adult movie—Captain America: Silver War. Later, his son asked if he could sleep in my bed. We felt like a family.

So far, so good.

In addition to multiple sources, you should ensure quality through various methods of collection. Observational data is great: How many times a day does he call? Text? What is his Uber rating? (On average 1.5; ranges from 1 to hundreds; null, since he shuns Uber and uses Lyft.)

My favorite method of data collection, by far, is surveys. If I want to know something, I ask.

“James, why did you major in American history?”

“Because I just love America, and all things related to American culture, and wanted to contribute to it.” Now he writes for a hit TV show. No big deal. (I wanted to rip his clothes off.)

“James, when did you start falling for me?”

“Before I met you! When I saw the photo of you and Olaf.” I had joined Bumble on a whim, the first weekend I became single after spending years 19 and up as a serial monogamist. Every recent photo had an ex in it. I managed to find a recent one of me at my previous company’s family day, sandwiched between sing-along Olaf and off-key Elsa. James was the first person I met off the app and I, too, fell for him before we met, when he texted me, “What was your rock bottom?” (A man who wanted to plumb my depths? I was so in.)

“James, so . . . are you in love?”

He was driving me to LAX when I conducted this survey. I looked at him looking at the road. He’d told me his first relationship after his divorce didn’t work out because, though he loved her, he was not in love with her. Big difference. He had not shied away from telling me how he loves my voice; the way I make him feel; my lips; the texture of my hair; every minute we spend connecting our minds, bodies, souls. But he hadn’t said he loved me, not exactly.

He inhaled. Then exhaled. Then: “I am.” Stillness. Seconds. Cars whizzed by. “I am. I didn’t think it would be possible to open my heart again after the divorce, but you make everything seem possible. You’re my black swan.” (As in Nassim Taleb’s The Black Swan, not Darren Aronofsky’s. Again, I know.)

His voice broke. I felt like I really saw him: forty-seven years of stories and of sadness, anger, and fear of not having ever lived true to himself, and of playing hide-and-seek with his heart. We got to LAX. I knew I loved him. I tried to avoid overanalyzing in the moment, and pushed away the feelings in my gut that this love wouldn’t work, at least not now, because his heart was still, clearly, knotted with his past. We said goodbye.

Data in its raw form gets messy, quickly, so the first step in analysis is to clean. This was surely a two-person job. While the plane was taxiing for takeoff, I contracted the consulting services of my best friend for the very next day. Eight hours. In person. With lots of rosé.

Generally speaking, any data collected while your subject is under extreme stress, under the influence, or in bed (with you, hopefully) should be thrown out. Exception to the bed rule if it’s while cuddling.

“You’re amazing.” In bed, at night—out.

“You’re everything to me.” On FaceTime, in the afternoon—in.

“Fascinating!” On text, as an answer to a direct question of mine, in the morning, while his kids were jumping on him—out.

Beyond this, you need a systematic way of pruning your data. The good part of pruning is that you get to decide what you do and do not care about. (You want to let your family have 50 percent say? Great. You care if he taps it back at SoulCycle with you? Sure. Anything. Entirely up to you!)

The bad part of pruning your data is that it’s entirely up to you—and this is the hardest, most important part, because having a good and meaningful data set will allow you to make the best decisions later. It doesn’t matter what your rules for pruning are, only that you have a learned set of them formed through direct experience (trained on your dating history) or theoretical knowledge gained (programmed from, say, your mom, or dating books like The Rules).

Personally, I put 100 percent weight on learning by experience. My training set includes one six-year relationship where I was twice engaged (and planned a wedding, which I canceled two months before for no other reason than wanting big love—and nothing short of it), one three-year relationship where he started saving for a ring, and a bunch others in between. Zero marriages.

So here’s what my model says as it applies to James.

Data I do not care about: our age difference. His divorce. His L.A. to my New York City. His young kids. Job or career. Looks (although, I have to say that I find him deliciously cute).

Data I do care about: whether we have a strong connection in mind, body, soul. Whether we want the same things in the future. If he has passion—topic agnostic, but about something. If he has drive—an intense will to produce rather than consume. If he is emotionally available. And, whether we have a deep desire to co-create (ideas, projects, potentially babies).

My best friend and I pored over the data. Highest weight went to how he made me feel: happy. Silly. Seen. Like someone finally understood how to read the book of me. Except that someone also lets the book get buried beneath coloring books and iPads and too many commitments. Unintentionally, but still.

After some additional analyses, time-series (how does he compare to my past, or I to his?) and cross-sectional (how does he compare to current viable options?), she and I agreed: strong data, suggesting a real match. Everything I care about is there, except one—emotional availability.

I had to go collecting.

The following weekend, he invited me to Boston, where a close friend of his just had a baby. The parents were so exhausted from parenting they repurposed our trip as nap time for them. We got to play house—me, James, the baby.

Here’s additional data: James loves babies. James loves his friends. James really loves other people, including me. But a question remained: Does he love himself? Why does he say yes to every social engagement, even the things he knows he can’t do? Why this fear of letting someone else in?

On Sunday, he wanted me to give him a personalized tour of my alma mater and to go to my favorite restaurant for dinner. After all that playing house, I was excited for some alone time. I was putting on my shoes as we were leaving his friend’s house when I heard his friend say, “So you guys will tour MIT and then we can meet back up for dinner? Maybe we’ll grill, or something?”

“Yeah, that sounds great!” said James.

Really?

I said nothing. Unlike the aborted first date, I was not upset. I let the curiosity of a true data scientist wash over me, wanting to see how this would play out, not letting emotions bias my data collection or analysis. I was watching him overcommit, in front of my very eyes.

It was a hot summer day. We held hands and walked around campus, up dorm row, and through the Infinite Corridor. Earlier, we had rosé and oysters, and suddenly, I was in the mood for ice cream.

“James, want to get some Toscanini’s?”

“Well, aren’t we going to your favorite tapas place in just a bit?”

We were—or at least, that had been the plan. And so we did. Maybe his friend got the same text I did before our cancelled date. Maybe his friend knew that when James made plans, they weren’t really plans. I still don’t know.

The gut—intuition—exists for times when your mind is operating on data overload. It often functions as a way of telling you to stop, to slow down, to process. The bad data I had previously tossed out—that first incident with his college friends, his occasional answering of texts with non sequiturs, the articles I’d send him that were brushed off under five minutes of receipt with a “Wow!”—became suddenly significant, because they exposed a hidden variable. (The good thing about hidden variables, if you can uncover them, is that they have great explanatory power.) They told a story: not of how James didn’t respect me, or my time, but of his overcommitment, his lack of bandwidth, emotional or temporal, for even himself. He overcommitted to avoid being alone, to not have to deal with his own data.

The day my mind caught up with what my gut already knew, I walked around in circles in Soho, talking to him on the phone, trying not to pass the same sidewalk vendors for the umpteenth time in tears, as I returned the bandwidth he gave me back to him. He needed it more than I do.

The funny thing about data, though, is that it’s never-ending. There’s always more to gather, and it’s always changing. But no matter which data set you’re examining, current or new, the process is the same: collect, analyze, adjust model, repeat. The most important data you have is on yourself. So be honest. Don’t be afraid to explore, to dig deep, and certainly, don’t curate what you show others. You’re the whole set.