From gut feeling to guided zero-waste dairy production | Ep 4

Mikkel Svold (00:00):
AI is also at the same time a bit undervalued and I guess misunderstood in most production lines, so getting it right might be actually one of the biggest business opportunities that we see right now in the processing industry. Is AI hyped?

Anna Olsson (01:57):
I wouldn’t say hyped, but it’s… After ChatGPT, AI became on everybody’s mind, and it would be the problem to everything. And there are huge amounts of pilots out there that is leading our work.

Mikkel Svold (02:09):
How can they benefit from AI?

Anna Olsson (02:09):
I typically simplify it and I say that you can learn, act, and detect.

Mikkel Svold (02:09):
When is the right time to introduce AI into your production facility?

Anna Olsson (02:09):
So the first thing is, do I have the sensor data? Have I stored it?

Mikkel Svold (02:09):
If you’re the production manager at a dairy and you want to start your AI journey, what’s the first step?

Anna Olsson (02:09):
I would say…

Mikkel Svold (02:09):
AI is on everyone’s lips. And in the dairy and the processing industry as well and all over food and beverage, it’s no different. From dreams of, I would say, fully automated plants to the stark reality of what can actually be done with AI, there’s no doubt it’s hyped, in my opinion at least. But AI is also at the same time a bit undervalued and I guess misunderstood in most production lines. So getting it right might be actually one of the biggest business opportunities that we see right now in the processing industry. So today, we are talking AI and we are talking about what food and beverage companies can actually do with AI.

(02:09):
Welcome to The Food Tech Podcast. I am Mikkel Svold. I’ll be your host. And with me on an online link, I have Anna Olsson, who is the co-founder of Intelecy, which is a non-coding platform for industrial AI.

(02:09):
Welcome to you, Anna.

Anna Olsson (02:09):
Thank you so much.

Mikkel Svold (02:09):
Let me just start out by asking you straight away. Is AI hyped?

Anna Olsson (02:09):
In a way, it’s I wouldn’t say hyped, but it’s… After ChatGPT, AI became on everybody’s mind, and it would be the problem to everything. And there are huge amounts of pilots out there that is leading nowhere, which is a lot.

Mikkel Svold (02:10):
Okay.

Anna Olsson (02:11):
And I can come into why I think that is. And there are a lot of AI strategies, a lot of hypothesis created from central digitalization teams. And it’s really important in my view that you actually get started and do something concrete and actually know what you’re going to solve, not just a big AI general pilot.

Mikkel Svold (02:37):
Yeah, because when I’m thinking AI, I’m thinking ChatGPT, I’m thinking Grok, I’m thinking Gemini, all of these, Copilot. So the language models, that’s what I’m thinking about, but also the image generators. And I’m just wondering. If that’s what on everyone’s minds, is that the same that you would then go use in industrial processing?

Anna Olsson (03:06):
Typically, not. So if you imagine the LLMs, ChatGPT, its algorithms behind the scenes, and they’re typically called something LLMs, large language models trained on, say, the all of Internet’s information. But in this industrial setting, think about it as if your dairy data from your plant is what the algorithm is trained on. So kind of AI. And it’s the same, but AI is such a broad term, right? Anything from generative, any AI like the images and chats. But today I think, or hope, we’re going to talk about more the application for the industrial process operations. How can you operate your dairy more efficiently?

(03:52):
And of course, an LLM could be an interface for you to interact with the data, but it’s a different kind of algorithms inside that is trained on the industrial data, which typically is time series data, like the sensor data that comes from your pumps, your filters, your pressures, the temperature. That is typically the data that you talk about in industrial AI. And we’ll come back to why that is important.

Mikkel Svold (04:22):
Yeah. I think so because I think one thing what would be nice to get clear from the very beginning here is what can a processing plant… A dairy, food and beverage, those kind of people, how can they benefit from AI? What is it that you can do?

Anna Olsson (04:47):
Yeah. So I’m going to focus now on the process optimization of a plant. So how can you optimize your dairy to run better? And I typically simplify it and I say that you can learn, act, and detect. So you can learn from everything that has happened, like why did it happen and what happened. And then the next area is act. So AI can help you detect anomalies before they happen so you can act in time. And that’s typically often talked about as predictive maintenance.

Mikkel Svold (05:22):
Can you try and give an example of that just for me to visualize what you mean?

Anna Olsson (05:26):
Yes. So imagine that you have a part of your production line where you have stops often, and you don’t really understand why and your maintenance crew cannot inspect and see it. And then an AI model or a machine learning model could let you know hours before it happens because it’s been learning on stable production for a long time. And then when it finds that something is about to go wrong, it will give an alarm so your maintenance team can go and fix it before it happens. And of course, then you will reduce waste, you will improve uptime, and of course it has a lot of… Economically, you’ll get an ROI on that typically rather quickly when you can hinder a stop in the production.

Mikkel Svold (06:20):
Also, just the fact that you can then control when the stop is going to happen. You can plan the stop rather than having just the part suddenly not working in the middle of a batch, for instance, or just in the middle of where you are the most busy in production. I’m guessing that’s quite significant as well.

Anna Olsson (06:40):
And the last part, yeah. So let’s tackle-

Mikkel Svold (06:41):
Yeah, you mentioned the last part as well.

Anna Olsson (06:44):
Yeah. So that is maybe the most common thing people think about when they say AI. We can do predictive maintenance because that has been on everybody’s mind. But then what if you can predict the future and the production? So that is something we see huge increases in, which is process optimization. That’s where you can predict something that’s going to happen, like in an hour, the wastewater will have this pH level, or in an hour, if I keep my production at this stage, the water percentage in my butter will be X.

(07:25):
And this is where it’s really interesting because then you can shift your processes not on gut feeling because you know how to produce cheese, but actually you can shift the process and you will know what kind of all these thousand things I can change in the process, should I change to optimize the best quality by example? Because quality, as you know, in food and beverage and dairy industry is hugely important because if you’re producing protein powder, the protein level in that protein powder have to be optimal. Same with butter. The water percentage in the butter has to be optimal.

Mikkel Svold (08:03):
Yeah. And obviously we’ve talked about this in previous episodes as well, but you also, in all processing industries, you have the raw material problem. So you have the problem that you don’t really control the quality of your raw material, but your output from your factory should be within small margins, the same quality or have the same attributes as well.

Anna Olsson (08:28):
Yeah.

Mikkel Svold (08:31):
I’m just-

Anna Olsson (08:32):
And that was… Yeah, go ahead.

Mikkel Svold (08:32):
I’m just curious. It’s so wild that you can already say that, but before AI, how would you do this? Because you would still have to control the output. You would still want to have some sort of preventative maintenance. You want to know when the pH level of your water becomes too sour.

Anna Olsson (08:58):
Yeah. So typically the only way it’s possible to do right now is to take lab samples. If you look at process optimization and you’ll see that the water content in your butter is too high, and then you change the process, but you can’t do it until then it’s too late. And most likely, quite a bit of that product will go to waste. So that’s the whole point here, that you can make the changes to your production before it’s too late so that actually you produce on top quality all the time.

(09:29):
And when it comes to maintenance, typically it’s being going or you’re going rounds physically maybe on the plant floor and inspecting things, maybe measuring that now this pump has been running for X hours and then we should plan maintenance.

(09:52):
I was actually at a plant not long ago where I talked to the team and said, “How do you know when something is about to go wrong?” And this guy, he was a senior engineer, and he said, “Well, I talked to the pumps.” And I’m like, “Okay, can you explain that?” And he said, “Yeah, I knock on them.” So he was trained for so long so he could knock on the pumps and hear when things were about to go wrong, which is great, but that knowledge only exists in his head and it’s not very scalable. So imagine now when you-

Mikkel Svold (10:23):
It’s very hard to pass on as well.

Anna Olsson (10:25):
Yes. And now we have a huge wave where plant workers are retiring. And how do you get the new generation, the digital born native generation to go and knock on pumps?

Mikkel Svold (10:39):
Yeah. I mean, what’s the solution there? Is it to put up a sensor, a sound sensor that will then knock the pump every hour just to hear the resonance? Or-

Anna Olsson (10:50):
I would rather-

Mikkel Svold (10:51):
What would you do?

Anna Olsson (10:52):
I would rather use all the sensors from both the pump and the process line and train AI models on it. And what is important is that you actually train your AI models on your whole production process because the problems with the pump might start earlier in the production process. So maybe it signals that it’s not coming from the pump that can tell you that something is about to go wrong. So there we are with the data again.

Mikkel Svold (11:20):
Yeah. I just came to think about, when do you think it’s… And you may be a little bit biased here, obviously, but when is the right time to introduce AI into your production facility? I mean, how digitally mature should you be to get bang for the buck? How do you get value? Because I would expect that a lot of dairies, a lot of other food processing plants, some of them obviously are high-tech but many of them would also be very reliant on these senior dairy engineers.

Anna Olsson (12:02):
And they should still be there. We’re not replacing them. So we’re taking their knowledge and scaling it. So that’s why it’s so important that this… This is kind of two answers. I’ll come back to the senior engineers and why we really want their knowledge and that’s why no-code comes in. But let’s first start with the first one.

(12:21):
Before you start thinking about an AI project, you need to look a bit at your dairy. Do I have the data? Because it’s no use to just jump into an AI project just because everybody’s doing AI. So the first thing is, do I have the sensor data? Have I stored it? And we typically say you should at least have 6, 9, 12 months of stored data from more-

Mikkel Svold (12:45):
Why so long?

Anna Olsson (12:47):
Because the machine learning models are trained on historical data and needs to understand how typically… what kind of data is coming when the production is running normally 24/7 or however you run the plant. And you need enough data to train the models so that they can predict failures and how you should optimize the process.

Mikkel Svold (13:16):
Would there be any seasonality that you want to take into consideration as well?

Anna Olsson (13:19):
Yeah. Yes. And then the next thing is how is your dairy being run? Do you have a control room where the operators are monitoring the plant or the dairy? I would typically recommend that you have that in place first, that you have control systems that there are operating the plant through a DCS or a control system. Because if you don’t have those first steps in place, let’s start there. And if you haven’t started saving your production data, start with that now.

Mikkel Svold (13:52):
Even the smallest amount of data or just any data that you can somehow collect.

Anna Olsson (13:57):
Yeah, start there. And then you can start training models, or you should never start training models before identifying your problems and what you can solve. But then when you have a hypothesis on a use case that’s valuable and viable, as we typically talk about, something that we guide our customers through, identify the use cases where there’s value in them, they’re rather easy to solve, to get some ROI or start positive on the project. And then you’ll pretty quickly understand if it’s viable if you have the data.

Mikkel Svold (14:35):
Yeah. I’m thinking just from your experience and from previous projects that you’ve already done, how is the ROI? What should one expect?

Anna Olsson (14:49):
Oh, it can be huge depending on… So look at your biggest waste factors, by example. Say that you have problems with cheese becoming contaminated by bacteria by example, and you have to cultivate. This is actually an example from one of our dairy customers, TINE, who’s using our product to solve that.

Mikkel Svold (15:16):
And TINE, for those outside the industry, is the largest dairy in Norway.

Anna Olsson (15:20):
Yes.

Mikkel Svold (15:20):
Am I right? Yeah.

Anna Olsson (15:22):
So maybe I should do this as a use case instead. But yeah, so I’ll do that. So they had a problem with bacteria leaking into the ready product to the cheese. And when that happened, it was not quickly to detect. You had to do your lab sample, cultivate it, and it took several days. And imagine the cost of that, all that product that had to go to waste, both from a money side like the cost of it, but also from sustainability angle. So it would be days that was wasted. So you can just imagine that the sums were this can be if you could avoid that.

(16:07):
So they’re engineers, and this is where we haven’t really covered why no-code. So we’re jumping a bit back and forth here, but no-code is important so that the engineers that have the problem can build the machine learning models themselves. Because if they have to call IT, they wouldn’t really have a hypothesis on what might be happening here and what should I look at. But the senior engineers do. They have a hypothesis. They know the process and they know that this is a really a big problem.

(16:40):
So that’s when they created an AI model in Intelecy, was their quality engineer, so it was not IT. They could do it on the plant themselves. And it turned out it was micro fractions in a filter that the human eye couldn’t see. So on the inspection rounds, it wasn’t possible to detect it, but the machine learning models could detect it because there were signs that would happen before it happened, like pressure increasing in a filter and things like that that the human eye could never see. And now that machine learning model actually picks it up every time it’s about to happen. So that’s a really, really exciting part. And that is just one example.

(17:28):
Another one is that a big, big cost for many of these dairies is energy. And imagine if you could run your process in a more efficient way so that you just save small, small percentages of energy use, but 24/7. Imagine the ROI on such a project as well. And that’s I think why the shift is moving from more traditional predictive maintenance more into process optimization. Because when it comes to ROI, just a breakdown might or might not happen. It might happen once a year, and then you of course save that money and didn’t have the downtime. But imagine when you can optimize your production process. So you increase the production with just small tiny percent, or you decrease the waste with just a little bit, then you actually have savings 24/7.

Mikkel Svold (18:22):
And I think also what people outside, especially the dairy industry, but also I would guess a lot of other food processing companies, I think what is astonishing is the volumes that go through those factories. And it’s with the volumes that you get… That’s also why we talked in an early episode about decimal optimization. So if you can increase productivity or efficiency or yield or whatever, if you can increase that just by half a percent or even less, that becomes $70,000 or-

Anna Olsson (19:05):
Yeah.

Mikkel Svold (19:05):
… whatever, right? Over a year. I think I want to come actually a little bit back to the beginning of our talk. AI is a little bit hyped. I think one of the reasons that it may be hyped is because many projects fail. Can you maybe talk about why AI projects fail? And, too, maybe if you have examples, that’d be… It’s always fun.

Anna Olsson (19:37):
Yeah, there are a lot of reasons. There are a lot of reports around this. And Gartner, they talk about pilot purgatory where successful AI pilots go to slowly die. And there are a lot of reasons. We already talked about the data. That’s kind of a prerequisite. And-

Mikkel Svold (19:58):
So poor data, basically.

Anna Olsson (20:00):
Poor data, that’s the first one.

Mikkel Svold (20:01):
Or no data.

Anna Olsson (20:02):
Yeah. Then the second one I would say is that you get ROI from it. So you can prove that you have value from it, and in a fast enough way because if you run your pilot and you have no ROI but it’s promising, then it won’t be that good.

(20:27):
And one of the big problems that I truly believe is that the AI projects typically are run as a central projects with coding everything from scratch. And that is a good way. You have data scientists that code machine learning models and then deploy them. But the thing is it doesn’t really scale. So it’s a good way potentially to start see the insights about AI, but it doesn’t really scale.

(21:02):
So I have an example here, a dairy that we work with that started that way. And they had a central team building a machine learning model for energy optimization of a Pasteur. And they were really happy because it really worked, but it took them months to code and build that one model. And then they started thinking, “Okay, this was a really great pilot, but what do we do now?” And they realized if they’re going to do that for the next Pasteur and the next type of equipment and maybe the next dairy, it’ll actually take the same amount of time as the first one, and then somebody has to maintain them and so over or and over again. It’s not-

Mikkel Svold (21:45):
Maintain the code and-

Anna Olsson (21:45):
Yeah, and run it. And it’s actually, it’s not working. So that’s why they were shifting from build to buy, which is really something that is about to happen now. SaaS offerings or software is of course I represent Intelecy, but where you have a software that you can build these models in and just deploy it within minutes and then maybe copy that model, adjust it for the next Pasteur, retrain it, and have the second model within minutes after you build the first one. So that is one of the biggest problems, I think, scaling the successful pilot. Yeah, you created a good model, but what do you do then?

(22:28):
And then the next one is not to be minimizing to the central teams, but often they do not have the same kind of knowledge as the senior operators and engineers at the plant. And also under-

Mikkel Svold (22:45):
And so the processing knowledge, we’re talking about.

Anna Olsson (22:46):
Processing knowledge. They’re really good from the IT side, coding, all of that. But actually what kind of problems are to be solved?

Mikkel Svold (22:56):
They can’t make the cheese.

Anna Olsson (22:57):
No.

Mikkel Svold (22:58):
Basically.

Anna Olsson (22:58):
And they can’t knock on the pumps.

Mikkel Svold (23:02):
Yeah.

Anna Olsson (23:03):
So the thing is here that we need to bridge that. We often talk about something called the IT-OT gap where there’s a big divide between the IT side and the OT side. But to get this to work, they actually have to work together. And we need to have a solution that’s put into the hands of the people with the knowledge and the problem, the frontline workers that you typically could say.

(23:25):
So in my view as that is also a product we deliver, is that that’s why you need a no-code tool. So yes, IT is really important because they help secure, ensure that the security settings is set up right, about the data is secure, and they are setting up the solution in a good way. But then it’s actually the people in the plant that are using it, building the models. Because they know what the challenges they have. They know that it’s a fluctuation in the quality of the protein powder. They know where their challenges are and they can easily solve it, so…

Mikkel Svold (24:06):
Also, they would also probably have a better understanding of where are the impactful problems to solve, not just how to solve them, but also prioritizing the problems.

Anna Olsson (24:17):
Yes, for sure. And that’s where you should start. So I’ve actually met central teams that confessed to me that they’ve never been to a dairy.

Mikkel Svold (24:25):
Right.

Anna Olsson (24:25):
So they are building the machine learning models, but they were like, “I think you know more about dairies than we do.” And then when we start talking together about, okay, if you could provide a tool to the dairy where actually the people with the problem and the knowledge can solve the solution, that’s a win-win for everybody, right?.

Mikkel Svold (24:48):
Absolutely.

Anna Olsson (24:49):
And I think you need to have the data. Otherwise, the pilot will not scale. You need to have a solution where the people who understand the challenges can build the models, and then you need to be able to scale it. So it doesn’t help if you just can build one model. You need to be able to scale it and use it.

Mikkel Svold (25:08):
Yeah. And I think also an important point you also made is about the ROI. So you have to be able to show the return on investment. And just considering my own daily life basically, it’s the fact that you can showcase a return on investment early on in the process can sometimes be a lifesaver and be that little bit that then gets, say the board, joining in on the decision or get everyone to back the decision on moving forward. How do you secure those fast return on investments, those signals that you would need to showcase early on in the process?

Anna Olsson (26:03):
That is something we have actually looked and realized, that we need to maybe educate or support our customers on in the first phase. In the beginning, we typically just put the hands into the tool of the engineers, and engineers being engineers often solve really exciting cases, but might not have the ROI hats always on.

(26:30):
I once met with a customer and they were so excited because they had solved or built the model for a really complex problem and they hadn’t thought about ROI. So I started asking a bit stupid questions like, okay, let’s see if we can document their ROI for this case. And I was like, “How often does it happen, this breakdown? What does it cost when it happens?” “Oh no, it never happened, but if it did, it would be terrible.”

(27:00):
So then we realized, okay, we actually have a blueprint that we help our customers with as the first step. And that’s typically where partners as Au2mate often also work together with us and towards the customer because they can guide the customer around these things. So typically we gather the people. And with the people, it would be maybe somebody from quality, somebody from maintenance, engineers and operators into workshop where they actually come up. Think about it like an exercise with sticky knots where they come up with the problems they can think about that could be solvable.

(27:45):
And then we typically rank them. I talked about it already, valuable and valuable. So both of the cases are, number one, easy to solve, low-hanging fruits. We have the data. It’s not that typical and valuable. So we quickly can showcase. And then we help plan this roadmap for what kind of use cases should we prioritize first, get the quick wins, show our why, and then start attacking the more big, complex one, or identify that this is a hugely important case, but we need some more sensors. We need some more data. We need to invest in that while we solve other use cases. And-

Mikkel Svold (28:26):
Is that a typical problem?

Anna Olsson (28:28):
Often, especially when you work with quality, because a lot of dairies, quality is the key thing that you need to optimize. And you mentioned already. If you cannot measure the quality of the product coming in, the raw product, because that will most likely affect how you run the process, and cannot measure the end results of quality, then using AI to optimize that process will be difficult.

Mikkel Svold (29:04):
Of course.

Anna Olsson (29:04):
And so there are so many new interesting sensors. So TINE that we worked with, they’ve been in quite a few of these projects also, in research projects. And they, by example, use NID sensors, near-infrared sensors, where you can measure things like I’ve discussed a lot like water content in a product, things like that. So you can measure it so you know what the optimum. What do you call it? The golden batch. So what does the golden look like? And can I optimize it? And you can still do it. You can solve use cases if you have lab tests because the lab data can be an input into the machine learning model.

Mikkel Svold (29:56):
Yes.

Anna Olsson (29:56):
Yeah. But then typically, how often do you do a lab test? Maybe once a day.

Mikkel Svold (30:02):
Yeah. So you’ll have all the deviations that are outside of that specific… If you do the lab test at 10:00, everything outside of 10:00, actually it’s probably not a black box but it’s gray.

Anna Olsson (30:17):
Yeah. And with those sensors you can get live-streaming data from the plant that runs up into the machine learning model. And that means that the predictions, if you continue as is, and this value is going to go up or down, if you have live-streaming data instead of lab tests, that result, the prediction from the AI model, can be delivered within seconds as well to the operators.

Mikkel Svold (30:42):
Of course, yeah.

Anna Olsson (30:43):
And this is an area we haven’t covered yet, because what happens after AI does a prediction? Somebody needs to consume that.

Mikkel Svold (30:52):
Yeah, but actually it’s probably a pretty good segue into our last topic here. So Industry 4.0, it’s much talked about, less implemented, lots of aspiration, maybe fewer results. Is the dream to have a fully human-free factory where everything is just automatic processes, AI-driven? And if so, how far off are we?

Anna Olsson (31:25):
I wouldn’t say human-free. You would still have humans there, but maybe not tweaking the knobs. But so in Industry 4.0, you talk about closing the loop, which is exactly this. We let AI automatically update the settings in the production. And we are there when it comes to technology-wise. What is the most important step before we get to that is getting trust in AI, getting trust in the AI predictions so you actually can trust that, yes, AI is able to make the right suggestions on how I should optimize my process. And I think that’s where you have to start. And the most successful customers we work with, how they do it is that where the operators are, they are typically using a control system to operate the plant and the dairy.

(32:23):
Let’s make the AI predictions available in the control system because then they can see that in real time, when I change something, yes, the curve flattens out on the prediction or it goes up or down. So they can actually see that it actually works. And that is a hugely important step to first create that trust in AI that actually when I do changes to the production, that is making a difference. And then you could move on to the next one.

Mikkel Svold (32:55):
And could-

Anna Olsson (32:55):
So then the humans still are running, turning the knobs in the factory, so they are still tweaking it. But imagine in a production process of, I don’t know, any product in a dairy, there are so many things you could turn on and tweak and settings to optimize the quality of a product. But if you let AI help you and say that you need to keep this pressure or temperature within this set of range, or you need to turn up the temperature, if you then create that trust so that they see that when I tweak up the temperature, actually the quality of my cheese is so much better. It’s always on par, then you create the trust in that the AI model actually recommends the right things for you and in the right time.

(33:52):
And the next step would then be to use those AI predictions to automate your automation system. And we have customers that have started with that. So it’s not a dream. Typically, they choose to start with maybe optimizing. We talked about pH level in wastewater. That could be an area that you start and maybe not in some other more complicated process, but I truly believe we’re getting there. It’s just taken a bit longer than people thought. And maybe that’s because of all the stalled pilots as well, that you haven’t really gotten to fully operationalized scalable solutions yet.

Mikkel Svold (34:37):
And it could probably also be if you have lots of those engineers knocking pumps, there’s a lot of processing knowledge. There’s a lot of how-to knowledge that is very hard to capture and therefore very hard to predict, I’m guessing.

Anna Olsson (35:03):
Yeah. But AI models will be able to predict, right? Because they know at what time, what kind of product were you producing, in what batch did you have, what was done in that process by the experienced engineer? And then the AI model will learn that when the raw material is X, this batch turned out optimal because the super engineer or operator put all the settings like this.

(35:37):
But you’re right. And on the part that there will be change management needed because if you’re used to going out and knocking on pumps, moving over to trusting AI will take a bit. So it’s not like, oh wow, we have this great AI tools, we can build models and now everything’s solved. You need to kind of get that into the daily workflow, using it, change management.

(36:06):
And something we see that many forget is actually allowing time because it’s always, nobody has extra time, typically, when you’re running a plant. There’s always you don’t have a lot of time extra to sit down and learn new technology. So what’s important that we see that some people forget or some teams forget is to ensure that you actually hold time for the engineers to work on process optimization and try to build new models, solve use cases, and not have 100% of their work time running the daily operations. So ensure that you take time.

Mikkel Svold (36:50):
It sounds like a problem that any industry across the board can recognize, taking your time to push forward development instead of just being in full operation all the time. I think that’s something that any company in any industry could probably be better at. And I have just one last question before our time is up. If you are a dairy or if you’re the production manager at a dairy or food and beverage, other food and beverage company, and you want to start your AI journey, to use a phrase that’s heard before, what’s the first step?

Anna Olsson (37:41):
I would say just start and get going. So reach out to Au2mate. Au2mate is a trusted partner of ours and have a lot of knowledge in this area and most likely know the dairy very well as well. So just get started and identify the data, start looking at the challenges, and don’t make it huge. Just choose to get started and kick it off and identify the right problems. Test with some data from the plant or even all the data. And of course you need a strategy and a plan, but I see a lot of companies get stuck in that strategy phase.

(38:25):
So yeah, you need to have a strategy and a plan, and you need to have this anchored centrally with the digitization teams and with IT because otherwise I’ve seen a lot of clashes where plants and dairies have been going rogue. Not going rogue, but ensure that you have alignment internally. And then just start looking at the use cases. Where do you have challenges? Where could AI be a potential solution? Do I have the data? And then let’s start looking at those cases. And I know both Ultimate and we are here to support on that.

Mikkel Svold (39:04):
Of course, of course. Anna Olsson, thank you so much for joining the conversation today. It was really interesting. It’s always, I think, interesting too to follow the AI trend and also to just check in on hype versus reality. Where are we right now and what’s the way forward? So thank you so much for joining in.

Anna Olsson (39:25):
Thank you for having me. It was a pleasure.

Mikkel Svold (39:27):
And to the listener now out there, thank you so much also to you for joining this conversation. And if you want to join the conversation even more, please do reach out to us at podcast@au2mate.com, and that’s automate spelled with a number 2, so it’s A-U number two. So that was podcast@au2mate.com. Reach out to us and let us know what you think about this episode. Let us know how far you are in your AI processing journey or if you’re not started at all. And of course, if you have any ideas or anything that you would like us to discuss on this podcast, let us know as well.

(40:05):
I think the last thing I want to just ask from you is, if you like this episode, give it a like and subscribe to the episode. That really helps us spread the word and keeps us going, sharing this knowledge into the processing field. So please do that and share it with a friend or a colleague or whoever you think it might be relevant for.

(40:26):
Last thing I want to say is just thank you so much for listening.