Mammen Dairies: Production data changes operations, tactics and business strategy | Ep. 9

Henrik Kæmpe (00:00):
You cannot just take a management decision that, okay, we want to improve yield with 5%. That’s not possible, right?

Mikkel Svold (00:08):
They don’t bring a lunch pack.

Henrik Kæmpe (00:09):
That day, they don’t bring a lunch pack. When I was young, we tasted 120 cheeses and then we had a cup of coffee and then we tasted 120 again.

Mikkel Svold (00:21):
Perfect.

Henrik Kæmpe (00:22):
So, no lunch that day. Here it’s important to invite people that’s not blind because it’s their own daily work, right?

Mikkel Svold (00:31):
How do you make people trust these data? Because I would assume that people don’t necessarily just trust what they see.

(00:53):
Welcome to The Food Tech Podcast. I’m Mikkel Svold, and today we are talking about how production data can not only change, of course, the production and the operations in a daily factory, but also how it pushes in and how it moves into the tactics of how to run the factory but also, of course, the business strategy itself. And that I’m looking very much forward to talk to you about. And with me, I have Henrik Kæmpe from Mammen Mejerierne, which is the second-largest dairy in Denmark. Henrik, you joined us in the last episode as well, so welcome back to you.

Henrik Kæmpe (01:28):
Thank you very much.

Mikkel Svold (01:30):
And just for those who didn’t listen to the last episode, I would strongly recommend that you go back and do that, because we might build on some of that information and some of that knowledge in this episode as well. And if you already heard it, you will also know that Henrik is the operations development manager at Mammen Mejerierne. So, let’s try and dive in. And I wonder, a steady flow of production data changes everything. That statement, is that true or false?

Henrik Kæmpe (02:02):
I think it’s true.

Mikkel Svold (02:04):
Why?

Henrik Kæmpe (02:05):
Yeah.

Mikkel Svold (02:08):
That was an easy why question.

Henrik Kæmpe (02:11):
Yeah. I could see a change from looking backwards to be able to trust looking forward. So, it could be a game changer. It is a game changer, if you have a, let’s say, steady flow and you trust your data. Then you can also, at the end, predict data. And that way looking to look forward and then control your production, rather than try to see what went wrong and how do we mitigate.

Mikkel Svold (02:50):
Okay. I think that’s interesting. How much data do you need? Or for how long do you need to have been collecting data and working with data in order to use it proactively? Is that something if I start collecting data and aggregating the stuff across the factory floor, can I predict tomorrow?

Henrik Kæmpe (03:14):
It depends where in the factory you are working. Are you an operator? Are you a team leader? Are you part of the management? If you are, let’s take an operator, that has a sensor, that let’s say it might be out of control, we have some issues measuring the values. Then let’s say you don’t trust it, and you spend your time, and in a unstable environment in order to get in control again. And then your focus is focusing on your issues, rather than you have stability and you can look ahead. So, that could easily change, because if you change the sensor, if you find what’s wrong and get in control again, that could be within minutes that you are in control again. On the other hand, if you have a full, let’s say, cheese production, where you think this cheese is not as it should be, we have to adjust texture or pH or some of the parameters to produce the cheese, then it’s a much long-term issue.

(04:36):
Then it could be weeks. It could be cheese that taste eight weeks old and it has a shelf life for 40 weeks. Then the cycle, where you take the decision, you follow up and you see the results. It could be yes. You know old cheese? One year old?

Mikkel Svold (05:01):
Yeah.

Henrik Kæmpe (05:02):
Then you make a change. And one year after you can see did the cheese come out with the change you have expected.

Mikkel Svold (05:13):
That is such a crazy timeline, right?

Henrik Kæmpe (05:16):
So, the timeline goes from seconds or minutes to years.

Mikkel Svold (05:22):
To be honest, even in your own production, how much is data used for reacting to immediate trouble, rather than planning or improving quality or being better tomorrow than just taking out the fire today?

Henrik Kæmpe (05:45):
I think the number of data is… I think it look like a pyramid, that we have a huge amount of data. And if, actually, goes down to shop floor and look at it, the huge amount of data. And they’re used to control the production, to consolidate the production and so on. And I think, actually, a lot of work going on with getting things right, getting things done and controlling the daily work. And then the higher we get in the… closer to management and management decision, then more the consolidated numbers we have. And the more deviations in the numbers is also disappears in average. If you look at average fat contains in a daily production, then it takes weeks to see if the targets moves. So, here we really have to use the thing that we learned in school, use standard deviations and so on. Where are we in reality with our processes?

Mikkel Svold (07:11):
If you set a new target, how do you then keep focus on that and keep adjusting towards that target so you can actually see the average go in the direction that you want? Because that will take more than just a couple of weeks, I’m guessing.

Henrik Kæmpe (07:30):
Yeah. Let’s say developing a new cheese and make it the first time, it’s not easy because you have to find a taste and so on, that people like. But it’s a reputation of making the same cheese again. And it’s a reputation of getting the same taste as the sole content pH and so on, so the customers buy the same cheese every time. And you have the same yield every time and you can calculate the cost price and so on. So, it’s that way the perspective change.

Mikkel Svold (08:15):
I think you’ve already mentioned all the different layers of where you can use data. So, in the actual production, of course, the day-to-day, the minute to minute adjustments that you would need to do. And then aggregate that up to how should we run the batches. In what order should we run the batches? How large should they be? How small should they be, that kind of stuff? And then again, aggregate further up or consolidate the data further up to how do we set the price. How do we get better deals in incoming supply chain, from the incoming supply chain?

(08:55):
I want to ask you, and I know that this is probably repeating what we talked a little bit about in the last episode, but what kinds of data points, if you were to choose two or three or maybe even a small handful of data points, are necessary to keep track of? And how can you aggregate or consolidate those data points going, I’m going to say upwards in the organization? In the organizational chart, maybe that’s more correct.

Henrik Kæmpe (09:26):
No. Somehow, I need all the data, all the different data for different purposes. Because take the process data, it’s about control, it’s about keeping the production stable. If we take the quality data, it’s data to ensure that we keep the product within specification, regulation, quality. If we take production data, then we look at the yield and the consumptions of our raw materials and so on. And if you look at the equipment data, it helps us optimize production setup, production equipment. If we look at resource data, it’s both costs and sustainability and so on. So, somehow we need to look at everything, all data in different worlds with different people and for different decision makings.

Mikkel Svold (10:49):
What data point or piece of data, I don’t know if you can talk about it, has surprised you the most that someone asked for?

Henrik Kæmpe (11:11):
It could be that sometimes we focus on data, that in an overall context it could be financial, it could be quality that has no real effect. That we look into and use data that we should not use within the group of people we are, that they’re simply too detailed. And on the other hand, sometimes we can forget that what really matters. That the yield or the value of some of the things we measure is really, really high, is really a lot of money, that we have to focus more… we have to focus on as a management. And has to spend more time out the things than we do and then most people do. And it come down to what do we do as leaders. How much time do we do or spend them on different tasks?

(12:24):
Because if you are a production manager, you should use a hell lot of time working with a yield, because it’s really a lot of money. And then at the same times, the same people asking very simple things, emails and so on. And of course, I know that our job is complex, but spending the right amount of time on the right tasks that’s aligned with the value of the task we are doing, I think that’s a huge optimization all over.

Mikkel Svold (13:09):
And it’s also what’s really hard.

Henrik Kæmpe (13:11):
Yeah. It’s so easy because we are not machines. And we cannot just sitting behind of desk, and at my level in the organization, just to look at, let’s say yield from morning to evening. We also have to work with other things. And it’s needed and it’s needed in the organization.

Mikkel Svold (13:39):
I’m thinking about the data sources and the data points that we could have on different levels of the organization. And somehow it seems fairly easy when you’re at an operational level. They may need a lot of different data for all the machines, but it seems to be easier to give actionable data points because they’re closer to the actual production. Whereas, when you take the helicopter a couple of stories up, I wonder how… You mentioned yield, but how do you actually look at yield when you get at a strategic level, at leadership level? What does it mean to look at the yield? And maybe more importantly, how do you make that number actionable? Because that was one of your main points from the last interview and I think that’s a very good point.

Henrik Kæmpe (14:39):
Yeah. By self yield is we’re going to say how much. If you take cheese production, you can also take other production, how much product per kilo raw material or how much raw material to make a given product. And that’s one thing. But you cannot just take a management decision that, okay, we want to improve yield with 5%. That’s not possible, right?

Mikkel Svold (15:09):
Good luck.

Henrik Kæmpe (15:10):
Yes, good luck. It’s really quite fun that we just could take a management decision. Oh, tomorrow, we’ll start. So, here, we have to get our boots on and get things done and involve people. So, we can, because in management group volunteers, there’s a lot of knowledge about cheese production. And so, in that perspective and then involving team leads and operators, depending on the task. Then break down the KPIs and see, look, what’s the underlying things in this KPIs? Because the yield is also, how much salt do we have in the cheese? How much fines a small cheese particle is lost in the cheese production? And then that could also be broken down, because how much is lost in the way, in the cheese production directly in the cheese fat? How much is lost later in the production? How much, let’s say cheese has we lost by breaking a corner of a whole cheese or damaged cheese and downgraded cheese and so on?

(16:34):
So, it ended up being quite complex. And then we have to look into where do we have the most… The biggest issue that we have to solve is one way to look at it. And at the same, looking in where do we have the biggest possibilities to optimize. It could be two different places. Getting the bottom of, don’t make so many mistakes and then you get everything in control. And then on your other hand, getting optimized things that on the surface seems to work very well, but still there’s a lot of optimization possibilities.

Mikkel Svold (17:30):
All of these, when you break up these KPIs, you say, for instance, you look at what happens when you cut the corners of the cheese, what happens or how large a percentage of a production batch is downgraded. How do you find out? Do you somehow obtain data? Do you have sensors out there or is it a manual process? Do you say, okay, now we scrap this, so we weigh it first and then we scrap it? Or how does it actually work?

Henrik Kæmpe (18:01):
It’s a relay combination. And it could be measurements per batch, it could be daily measurements. And also, if we have to optimize it, sometimes also analyzes that we don’t do very much on a regular basis, because it will be fat in the way. We don’t have to measure per batch, but we have digged into what the fat contents in a way that we lose. So, it’s a matter of using, let’s say, standard, lean tools to see what should we dig into, and where should we dig in, and how should we do it. And then see what we find out and then optimize from there.

Mikkel Svold (19:03):
I know that cheese production and I could mention other industries as well, I’m sure, but I know that in dairy there’s quite a tradition from many of the dairy operators that they have so much knowledge, that they can basically… they can bang one of the tanks and then they know if it’s the right temperature, basically. So, there is a lot of, I’m not going to call it gut feeling because that’s actually wrong, but it’s more like there’s a lot of experience talking. And there’s a lot of being at one with the production somehow, as the person walking around, walking the shop floor.

(19:47):
And I could imagine, as with so many other industries, that when you then introduce hardcore data points, objective data points collected by a cold sensor, how do you make people trust these data? Because I would assume that people don’t necessarily just trust what they see; is that correct?

Henrik Kæmpe (20:15):
Yeah, I understand. Yeah. There’s a journey in that also. Productions get bigger, more industrial, and that’s to be competitive. So, when I was educated dairy man, we actually on each batch took the curd and say, “Ah, this has to steer five minutes more.” That was the measurement. And I learned that from, let’s say, the old experienced dairyman who was in charge of the cheese production. “Now it’s too soft, now it’s okay. Now it’s too hard, we have to move on.” And then we started to measure the water contents in that cheese curd we have measured in our end. And that way, we try to standardize, calibrate our feeling of how is the cheese. Now we make the cheese in closed vats and that’s not possible anymore.

(21:32):
So, we also take some compromise, because yes, that feeling of how is the cheese right now, how’s the cheese production going right now, somehow we have to standardize that and get away from that gut feeling, at least in that part of production. And I think you can go to very small dairies that they make a couple of vats every day. They do it that way. But I think also the customers accept that now we’ve bought this from a small farm dairy. Customers also think that, ah, it’s not the same this time. There’s acceptance of that it’s not the same every time, where if you go to-

Mikkel Svold (22:28):
Yeah. It could even bear some kind of quality stamp with it.

Henrik Kæmpe (22:31):
It could. Yes, exactly. If they’re not the same, it could be a quality stamp, where the more industrial the production gets, the more the customers don’t accept that anymore. Then we can measure all the things we want. If people don’t like it, they don’t buy it. When we would taste the cheeses… And we taste all the cheese we make. The one in charge of our cheese production, he or they tastes all the products to be sure that they’re the same every time when we make, let’s say Danbur cheese.

Mikkel Svold (23:14):
They don’t bring a lunch pack?

Henrik Kæmpe (23:16):
That days they don’t bring a lunch pack. When I was young, we tasted 120 cheese. And then we had a cup of coffee and then we tasted 120 again.

Mikkel Svold (23:28):
Perfect.

Henrik Kæmpe (23:29):
So, no lunch that day. And we do the same… But it’s really important part of our job. And that’s also the fun and interesting and working with food stuff, that if people don’t like it, they don’t buy it. They find another brand, they find another products. So, let’s say the Mammen cheese has to work, has to have the right texture and the right mouth feeling and we have to like it. And not just as when we cut it in the cheese production, but also on bread. We have to say, okay, when we cut a piece and put off a piece bread on Sunday morning. And we have to say, this is the best cheese we can get, because otherwise we’re not there and they buy it, a piece of cheese from our competitor. So, end of the day… And also, we don’t get a complaint. It divvies a bit from standard our expectation. We don’t go to the internet and send the claim to where we have bought a product. We just throw it away or eat it and say, “Okay, we’ll find another product this time. I prefer another brand.”

Mikkel Svold (24:49):
Yeah. So, you’ll never know.

Henrik Kæmpe (24:51):
You’ll never know.

Mikkel Svold (24:51):
Basically.

Henrik Kæmpe (24:52):
Yeah, basically. So, that’s really important that we are really on top of the quality of our products. And then convert that quality that we taste, and feel, and agree on. How do we produce the cheese, and to be able to do that down to quality process controls, down to sensors. We’re down there again.

Mikkel Svold (25:21):
I actually have one extra question regarding this trust, how we can trust the data and whether people will trust the data. Because getting people to trust the data, getting people with 25 years of experience to trust a new data point introduced to them, that’s one side of the equation. But I think there’s also another side and that is to ensure that the data we look at is correct. What kinds of guardrails do you set up and how do you handle exemptions and false positives, false negatives, like that kind of stuff? How do you deal with that in your actual data setup?

Henrik Kæmpe (26:16):
We had a system so far for most of the things, where we measured it. Let’s say, what’s within the limit and what’s out-liners? And of course, again, depending on what we are measuring, if it’s direct measuring on the sensor, if there’s that sensors drifting or is it consolidated data, that it could be our yield. Are we in control with the yield? So, we try to set up borders, limits for whatever we measure. And then we react immediately on that, but sometimes it’s hard. Is this really out of control or is it as we measure? And we have to be very careful, if it felt positive or what’s really wrong here. And it’s sometimes really hard. And sometimes it’s easy to convince ourself that, for instance, yields changes in cheese production during the year and it also changes if it’s from milk, from cows on grass or it’s not on grass.

(27:43):
So, there’s a lot of variances. So, sometimes it’s easy, both with the yield, but also a lot of other stuff, to to convince ourselves that these data are right, what’s wrong here? Or these data are wrong, there’s something wrong that we can’t find out. It’s a quite hard task to dig into. But it’s the same way as optimizing things that put a group of people with nuls around, if you can cover 360 degrees nules around an issue. And then from different perspective, challenge what we see and what we measure, because I think here it’s important to invite people that’s not blind because it’s only daily work. You often get blind of the numbers if it’s your daily work.

Mikkel Svold (28:49):
I think somehow it’s reassuring to hear that experience will not go out of fashion.

Henrik Kæmpe (28:57):
No, definitely.

Mikkel Svold (28:59):
And I think also it’s reassuring to hear that there’s a lot of talk in different production areas, I’m sure both in dairy production, but also in other industries, about what’s called lights out production. So, basically, having a factory with no people on them. And it seems like it’s going to be a long way, because there’s so many variables in cheese production. And there’s so much craft and knowledge that comes with producing a good cheese. That is not something you can just completely automate by setting up more sensors.

Henrik Kæmpe (29:36):
And then at time. You can make a yogurt or something and next day, you can taste the finished products and you can adjust it. You can add more flavor or another flavor and so on, but there’s really a time perspective in cheese, because really it’s weeks and months and up to years.

Mikkel Svold (30:00):
Yeah. It’s like with any good wine as well.

Henrik Kæmpe (30:03):
Yeah. It’s the same.

Mikkel Svold (30:06):
You only know it’s good when years have passed.

Henrik Kæmpe (30:10):
Yeah, exactly. And then you can say-

Mikkel Svold (30:13):
Just hope you took notes back then, right?

Henrik Kæmpe (30:15):
Exactly. And then is it you that had changed or is it the cheese or the wine?

Mikkel Svold (30:25):
Henrik Kæmpe, thank you so much for enlightening on this. And I think it’s very interesting again to talk about the levels of where data comes in at different levels and how you can activate your, well, whatever, how you run your daily job, based on what kinds of data and what the current data tells you. I think that was very interesting. Thank you so much, Henrik, for joining us again.

Henrik Kæmpe (30:50):
You’re welcome. Thank you very much. It was a pleasure to be here.

Mikkel Svold (30:54):
And of course, to the listener out there, thank you for joining us also to you. And if you have any questions, reach out to us at podcast@au2mate.com. That was podcast@au2mate.com. If you have any questions or people you want us to talk to or topics that you think would be interesting, reach out and let us know. It’s always nice to hear from you. And other than that, I just want to say thank you so much for listening.