Making Good Use of Data: Jaime van Schyndel of barismo
Interview by L. Valena; photography by Canyon Twin
Jaime van Schyndel is the President of this company, but he is also the founder of barismo coffee roasters , which he started in 2008, and co-founder of Draft Coffee Solutions. I sat down with him recently to talk about his opinions on data collection in coffee roasting.
How did you learn how to roast?
I started out as a barista/ home roaster, and thought I was roasting really good stuff until I was actually trained by someone who really did roast good stuff.
Where did you meet this person?
A friend and I started putting together a top list of the most impressive coffees that we had seen. Not the specific coffees- we were really interested in the people roasting the coffees. Coffees that we shouldn't like but we did, or coffees that we found truly exceptional. Of that list, there were four people on it, and we went down the list and there was one person on the list who made coffees that I shouldn't like, but were really good. They were dry processed, they were pulped, they were things that I didn't like normally, that were just exceptional. He was a roaster from Taiwan. In first training me how to roast he had me hand-sort coffees. If I was going to roast a coffee he was going to make sure that I didn't blame the coffee for the flavors that were in it, so that I could truly evaluate what I did to it. So I'd go through it and turn every bean over three times while it was green, and two times when it was roasted to remove every defect possible. His style was not to teach me how to roast, so much as how not to roast. It was learning about defects and the mistakes that you've made, and how to not do that. What you want to do is your problem.
And what did you want to do?
I decided I wanted to light/medium roast coffees that had consistency and ease of use. I had previously been a barista who struggled with other roasters’ wandering profiles. Roasts that were light were suddenly dark, or unevenly roasted- these things were very confusing to me and I really just wanted consistency first. Barismo originally was coffee for baristas.
When you became a production roaster, you came up with your own framework for figuring out how to make sure it was good every time. Can you talk about that?
We initially played a lot with roast logging, and having very fine detail. We had side-scrolling data loggers that were meant for other things, and that were high resolution, connected to a probe and logging the profiles. But we realized that we had been trained to roast in a manner that was built on consistency, so we had points within the early phase of the roast, and we would measure the distance between those two points to estimate what the drying was, to know if we had dried the coffee enough or not enough. We had a rise over run that we followed- how much did the temperature go up during this specific time? We would look at those segments, and we knew that if it went above or below a certain point, it was bad. If it doesn't get enough drying time or it gets too much, that's bad. So it was really about narrowing it into what we wanted it to be.
We realized that the logging was actually giving us a lot of misleading information. We would be making changes to the gas input so the airflow changed, but the data logger looked the same. We realized that just because the profile looks good doesn't mean it is good. And it's actually not indicative of the things we think it is- it's just an empirical observation but it doesn't necessarily relate to the data points that went into it.
Is part of that because it's an organic product that is constantly changing and shifting?
Yes. All coffees are technically blends. Even a single estate coffee is almost always a blend of different parts of the estate, or different lots- or at least of different plants and different harvesting days. But also we roast in an environment with slightly different temperatures, different humidities, you open the door and humidity gets in or gets out. First crack kind of wanders. So, your first crack can be in different places- the question is, do you pretend it's always in the same place even when it isn't audibly, or do you respond every single time to when it changes?
How do you do it?
We set up theoretical points of where it's supposed to be, and then when it isn't we go back and find out why. Most people have been leaning on the fact that the profile looks good, and so therefore it must be good. Our attitude is that it was actually supposed to be 'A', so why isn't it 'A'? We make adjustments as the coffee ages or in days of spectacular humidity. If it's too humid or too dry, then we have to make adjustments to our theoretical profile. We don't get stuck looking at the pretty picture that the roast logger puts out.
We could have deviations- let's say that the roaster was using a roast logger- they can miss an adjustment that was really pivotal, and still the roast profile can look like they hit the adjustment. It's really important to just measure the things that are worth measuring, and not necessarily fall in love with what looks good. That's like looking at the color afterwards and saying that if it's the right color it must be good. But how did you get there? That's what's important. A roast profile can have a long, slow bottom out- it might mean nothing. Or it can have a really short bottom out, and a quick rebound which means nothing. It's about the trajectory that it takes later on. And becomes that rise over run that hits those two later data points that effect drying. I think when you fall in love with the whole profile, you say that it has to drop the same way every time, and I say actually it doesn't. It just has to hit the drying at the right time. It needs to hit first crack at about the right temperature/time. It's fundamentally trying to predict what you should do rather than reacting to what happened. Roasting should be predictive, not reactive, to be repeatable.
The same thing with Agtron, which measures colors, tracking roast profiles, and even TDS meters. Really these things measure observational data, and we have to understand what's behind it. Those are tools to make it approximately repeatable, not necessarily to assure us that it's quality. A TDS meter just measures what's in it in terms of solids. Even though TDS means 'total dissolved solids', it can also measure suspended solids, and we have to really know what we're doing, not necessarily just say our chart 'looks good, feels good, must be good'.
So the data is still an important thing, but it's not the whole story.
You can't fall in love with collecting lots of data to validate your results. Back when I was a home roaster, roasting in an apartment on a Stir Crazy (which is basically a popcorn paper), I thought I was roasting really good coffee. Those profiles were pretty, but meant very little once I really learned what is wrong, what is right, and what the boundaries are of the roast errors, and separating that from the physical defects in the coffee. You start to realize that the line on the screen is just a byproduct of the equation, it is not the equation itself.