Transactions are people

The most basic information about customer transactions tells you what someone bought, when they bought it, and for how much. But if that’s all you see, you’ve pretty much reduced people into rows in your spreadsheet and you’ve put to bed any ambition of understanding the relationships you have with customers. This is a post about coffee, but it’s also about waking up to the meaning and motivations behind transaction data.

To anchor this thinking, I’m going to summarize a bunch of quantitative transaction data with you: 1,973 of my purchases with coffee shops across the last seven years. This will take us a little bit into acquisition and upsell, but the main point is to humanize 1’s and 0’s — to show how an AI system might replicate some of the first-person insights here. If I were to encapsulate the key to that thinking, it’s that you focus on mutual maximum lifetime value. That means understanding what matters to customers, which changes depending on what kind of customer they are and where they are in their lives.

Welcome to Verve

Integrate.ai is based in Toronto, but most of my time is spent in the western outpost of San Francisco. Every morning, the team is used to seeing me take calls from one of my favorite local coffee shops, Verve.

The first time I went to Verve was three days after they opened: February 16th, 2017. Between then and the end of 2017, I had 175 different transactions with Verve. It’s my main coffee shop, as you can see by this table of 2017 coffee shop transactions. Also I’m actually writing this at Verve right now.

If you were trying to acquire me as a customer, here are the things that seem to matter:

  • The major competitors for my cafe dollars are Verve and Reveille. These are both in my neighborhood (Verve is a 6 minute walk from my front door and Reveille is an 8 minute walk; 0.3 miles vs. 0.4). I would say these also have the highest quality of coffee/tea/food. So being close to where I am and having products that I like are probably the top elements.
  • If you look at the last line, you can see that my transactions/month went up a lot starting in June. That does, in fact, correspond to a life transition: I started at Integrate.ai in May. Before that, I was working down the peninsula and often not getting back home to San Francisco til 9:30pm.

If you’re a coffee shop, you’re going to have done some market research about where to open your store and already have some understanding of who is the an area and what they are like. If you look at the pictures of Verve above and Reveille below, you’ll also get a sense that I really like modern design with glass and wood. My furniture purchases and restaurant choices also would show this if you had access to them; my social media posts would also give a lot of clues about my aesthetic. And of course, there’s a lot that you could assume about me based on very simple aggregated preferences — odds are that I like what people in my neighborhood like.

But it’s really learning or detecting a life transition that probably is your biggest opportunity. Becoming part of a new habit or ritual is easiest at life transitions since they open space to new kinds of choices. Your own first-party transaction data often shows when someone is starting to look like a different-kind-of-customer.

Counting and trusting

In order to understand whether I’m becoming a reliable customer, you’ll need to have some way of knowing that “I” am “me”. A barista doesn’t have to be a super-recognizer to identify a person who comes in each day (more on human-human relationships in a moment). But at scale, cash transactions are the worst. I almost always use my credit card, so let me tell you about the two problems with having the table above adequately represent me:

  • Prairie Lights is bigger than what shows up here–my family knows how much I love it and so when I’m visiting them in Iowa, they often have gift certificates for me.
  • The table also doesn’t report my love of Boxcar Social coffee in Toronto, which was really close to Integrate’s first office. They only take cash, so they don’t show up in my data and they don’t have “me” in their spreadsheets except as disconnected individual sales.
My favorite coffee spot in Toronto (it looks a bit different now)

Boxcar Social is good to pause on because I would say that even though their accountants can’t know anything about me-as-me, their baristas saw me as a person first and foremost. That is, they got my unwavering loyalty by simply trusting me.

It was my first day in Toronto for the new job and I went to them without any Canadian dollars. They said, “No problem, just pay us back next time you’re here.” If we go for coffee in the summer, this is where I will always suggest us going.

This is probably not quite the right strategy for a store in an airport, but the barista on the corner of did the emotional+financial calculation and probably made a good call on risk vs. reward. There’s an opportunity for AI systems to figure out when and how to be a bit more human. (Now! A very important caveat on making sure not just to learn bad kinds of bias.)

Customer rhythms and mutual maximum lifetime value

The next table shows the top cafes with data starting in 2010. The table answers the question of how regular my visits are. What you can see is that a place like Weavers shouldn’t be surprised if I go ten days without stopping by. But if Ryan at Verve doesn’t see me for ten days, he should check the hospitals. Reveille would also worry if they didn’t see me for ten days, but it wouldn’t be as unusual (see the top quartile values).

If Verve or Reveille were an Integrate.ai client, we’d want to know how many Tyler-like people they have. Is it a situation like Caesars casinos where you want to make sure you treat high-rollers specially? Or is that a tiny amount of your business, so no need to create a white glove service? This is something measurable: transaction rhythms and lifetime value calculations tell you what your main customer patterns are.

The person who stops by very occasionally may never be able to (or want to) transition to be a daily customer. And you could waste a lot of time and money trying to upsell or retain someone who won’t be worth it: in the chart above, you see a number of businesses that I rarely go to anymore, Coffee Bar and Mazarine were places I took folks who worked at the Natural Language Processing company I co-founded after grad school. Equator was one of those, too, but it happens to be closer to where my boyfriends lived — until last week when they moved to a different neighborhood. (Equator, I really like you, but do not spend a bunch of money trying to get me back. It’s not worth it!)

What matters to a customer with a low slow rhythm may be different than what matters to your most loyal customers. You can’t play to your strengths if you don’t know what different kinds of customers value, what they tolerate, and what they think stinks.

Meanwhile, if you find that 100% of your customers are one-time only, well, for many businesses such little retention suggests one course of action: panic.

As a corporate entity, you can only go so far treating each individual as an individual, but you at least want to know the basic categories so that you don’t keep trying to corral everyone on to the same path. Mutual lifetime value means figuring out for any given individual, what the best-but-also-most-appropriate customer journey is.

Verve vs. Reveille

If Verve were an Integrate.ai client, we wouldn’t expect them to know about my transactions elsewhere or certainly not with Reveille. And vice versa if Reveille were our client. But since I have all the data about me, let me paint a picture.

The position “Tyler Schnoebelen” occupies for the two companies is different. I’m guessing that I’m a fairly high-value customer for both. For Reveille, the question would be, “Why have we lost Tyler” because before Verve opened, it was Reveille where I was spending most of my coffee dollars.

Reveille, my second-favorite neighborhood haunt

The immediate qualitative answers are (a) the space in Verve is more compelling, (b) the folks at Verve are super friendly and fun. What’s interesting about (b) is that Reveille probably has the third-best service (Weavers ranks second mainly for one super-duper-friendly barista).

A big shift is that the first round of baristas that I fell in love with at Reveille when they opened on May 19th, 2014 are no longer there. Their replacements are nice but they just aren’t the same. There hasn’t been much in the way of turnover at Verve yet although one of my favorites is about to move to Southern California because it’s better for his other career: professional skateboarding.

What either of these companies should hear, then, is that if they value their relationship to me, then they need to pay attention to the fact that I really value the interactions with their staff. The people are not fully separate from my experience of the spaces and the products.

From Verve’s perspective, why can’t they take all of the Reveille interactions? One aspect is that I don’t actually want 100% of my time to be in one cafe, I value variety, so there’s always a ceiling. But more particularly, Reveille is a much better match for me when I want to eat–they have a delicious shakshuka for mornings and a great kale salad for lunch. These are things that match my desire not to have many carbs in my diet.

Tempt me with a shakshuka but hold the bread

Most of what Verve offers is bakery stuff: eggs on biscuits, avocado toast, muffins, quiche. I sometimes get their chia seed pudding in the morning but it has delicious jam in it and for the last few months I’ve been avoiding sugary stuff. So if they really wanted more of my dollars, they’d need to change/add to their menu (Reveille has a kitchen staff). There may or may not be enough Tyler-like customers for it to be worth it. In the case of the Verve location near me, I don’t think they have room for a real kitchen, so it’d be a more creative solution.

Even though Verve doesn’t know that I want shakshukas and kale salads, they could probably find indications of opportunities in their existing data. How many of their most frequent customers rarely buy food? What times of day and what products do they make exceptions for? For big food buyers, how do their purchases cluster? Looking at the data this way can show how big the “gluten-free” population is, to justify offering more or cutting them altogether.

The main point here is decidedly not “cater to Tyler-like customers”. That only makes sense if there are enough of them, but you don’t know that unless you’re measuring transactions in terms of customer-types and then asking what it is that matters to the different key customer types. To simply try to scoot everyone along the same path or to imagine that everyone cares about the same things means you haven’t looked at your data with the keen eye of a pattern-matcher nor the keen heart of an anthropologist. Transactions are not disembodied words and numbers, they are gateways to understanding what matters to the people who buy your products and services and the people that might.

Tyler Schnoebelen (@TSchnoebelen) is principal product manager at integrate.ai. Prior to joining integrate, Tyler ran product management at Machine Zone and before that, founded an NLP company, Idibon. He holds a PhD in linguistics from Stanford and a BA in English from Yale. Tyler’s insights on language have been featured in places like the New York Times, the Boston Globe, Time, The Atlantic, NPR, and CNN. He’s also a tiny character in a movie about emoji and a novel about fairies.