Modernizing feed production with digital, data and AI
Discover how to transform digital buzz into bottom-line results in feed manufacturing, cutting through the artificial intelligence (AI) hype to reveal practical use cases delivering measurable value today.
Drawing from real client implementations, this session presents a framework for approaching digital, data, and AI investments — from digital twins that optimize throughput and reduce cost per ton to real-time quality monitoring that minimizes shrink to predictive maintenance models that prevent costly downtime.
Pelsoci explains the building blocks of digital maturity essential to AI success and argues that waiting isn't an option — stagnation carries risk. Feed manufacturers can systematically progress from foundational infrastructure through data integration to AI enablement, creating a technology flywheel that turns digital tools from an operational hindrance into a true competitive advantage.
PRESENTATION TRANSCRIPT
David Pelsoci, managing partner, Captios Partners: Good morning. Thanks for joining us today. As Jackie mentioned, I’m here today to talk a little bit about how all the different digital technologies come together to support and optimize and ultimately improve the bottom line of different businesses in the agribusiness space. Before we jump into the details, a little about Captios. Won't spend too much time here, I promise.
But we are a data, AI, and digital consulting and services firm. We help companies both from an advisory perspective as well as implementation perspective, and we primarily focus in the agribusiness space. That includes grain, animal feed production, as well as livestock production and ag lending services sitting on top of that whole area. So questions like "Where does AI fit into my business?" "How do I think about cybersecurity?" "How does this whole picture come together to serve those outcomes?" And ultimately build roadmaps for our customers and then work with them throughout that journey. Whether that's implementing or operating as a trusted advisor and project management team, we go with them all the way along the journey until they really realize those business outcomes.
Today, we're going to talk, like I mentioned, a little bit about how that whole picture comes together. And first kicking off with some different AI use cases that are really relevant now. We try to stay grounded in what's actually happening and available in the space.
Oftentimes picking apart some other adjacent businesses that have taken some of these steps. Agribusiness, as we know, is sometimes trailing some other industries, and that's a good indicator as to some of the areas that can be taken. A digital twin is one of those concepts. Imagine all the different data inputs you have from a feed mill production operation. So you have your ERP, you have your automation systems, you have dispatch, you have scheduling, maintenance, all of those different data points that can come in, in a real-time way into an AI neural network model.
With all that live data, including the physical layout, you can do a couple of things. One is getting a real-time view of what production actually looks like in a sort of consolidated dashboard manner, and you can actually watch products being made from ingredient all the way through finished goods and shipping out. And the idea there is that the tools enable you, in addition to observing that in real time, you also get the ability to perform what-if analysis. So what if you move a production line over 10 feet? What if you change ingredients that are priced differently?
What if you have alternating power sources and different pieces of data that are changing? If you change that without actually physically changing in the world, but in the tool, what bottlenecks does that create? What impact does that have on the bottom line? And it really enables you to put all those different data points together in one view that ultimately is focused on improved optimization and utilization of the plant. It's actually technology that's been around for about 25 years, especially in other industries.
You see it very prevalent in heavy manufacturing. But now it's making its way into animal feed space as it's come down in affordability and one of those areas that actually you can get clear ROI on. Another example, and this is just one of many, back office activities, so HR, financial activities are really good candidates for a lot of what has come as a result of the introduction of large language models. So the use cases fit nicely into that space. So here's an example of one where we're looking at accounts payable.
This particular example is one in which you've got all your different customer contracts, the terms in those contracts, receiving and pricing rules built into the business. And as opposed to manually reviewing each invoice to see if it fits into all those different terms that I just described, you can match that up with an agent who will actually check against those terms and say, "These are all the ones that fall outside of a normal range or outside of the definition of what you want to see, and these are the ones that require your reconciliation or approval," as opposed to going through thousands and thousands of detailed invoices. We'll come more thematically back to this, but the obvious statement that a lot of people think is this is primarily going to be a labor cost focus in terms of the efficiencies that we find. Many of our customers in the animal feed space rarely pursue headcount reduction as a result of these implementations of tools. Instead, they can direct their people toward more value-driving activities as opposed to spending half their day doing manual reconciliation of, in this case, invoices.
So these are just a couple examples of some of the realistic AI technology that's available right now and rather than possibilities several years down the road, which is something we see quite a bit when we watch the media and watch what a lot of the leaders of these technology companies are saying. So I won't read all these quotes, but basically, we've got some pretty aggressive statements around artificial general intelligence, which is basically AI levels that are equivalent or better than human abilities. And we've got statements from some of these folks from last year saying that it's going to be available in this year or the next year, three to five years away. I think it's important to remember that a lot of the leaders of these companies, publicly traded companies, have a vested interest in that message. Obviously, from a share price perspective, they aren't exactly objective in the way they think about that.
That's not to say it isn't coming. The timeline is uncertain and just how much it'll impact the business and when. So then where does that leave us? Do we do nothing? Do we wait for AI to just solve all of our problems?
There's a lot of other technologies that have to be considered to support AI applications, but before we get into that, way too many companies go through the motions of implementing technology in general and don't realize the outcomes, and they don't really have a good why. They've implemented these tools, all this money's been invested, and it's still unclear what the outcome actually is. So we really boil it down to more of a business-centric approach towards thinking about these technologies. A lot of firms get really excited about the tech, and you can probably hear my enthusiasm. I do too.
However, I want to be careful to not have a situation where you have the tail wagging the dog. The business outcomes are really what's supposed to drive how we think about the technology that we choose to use and implement in the space. So starting up front, far too often we work with customers that don't really have clear goals stated. There's really no North Star about how they're supposed to be operating as a business. And more specifically, and if they do have those, it's not quantified.
Very common, we see companies with loose goals that are pretty unclear and broad in the sense of what they actually mean. So that's really where we begin this conversation and how we think about this whole picture coming together. And then the second part of that is often if there are KPIs and quantitative goals that have been stated, it often doesn't get communicated appropriately down to the team, such that some of the folks at the more tactical levels have a clear idea and a shared view of what success will look like across the organization. So that's really the starting point. Again, like I mentioned, more on the business orientation side of how that drives what technologies we implement and when, where, and how.
A couple examples: When we think about feed specifically, throughput is one we hear pretty commonly. One that's a little less common is thinking about the full life cycle of an order to delivery. So what is the timeline that it takes for someone that comes in and places an order, and what is the ideal timeline to get all the way through delivery and billing? Some of these other ones are pretty straightforward in terms of what they actually are looking to achieve.
But actually hardening those and making sure that they are understood throughout the organization is really the two parts that we would argue need to be in place. So now moving over to technology. The way we think about all these different digital areas is there's four domains that we want to be concerned with. So you have infrastructure at the bottom, core applications right above it, data, and then AI. So working our way up from the bottom, infrastructure is where the physical world meets the digital world.
So where are you actually hosting and storing all of your technologies across this whole pyramid framework? Is that in the cloud? Is that stored locally? Does that sit in some sort of hybrid model? How do you think about that from a cybersecurity perspective?
And how are you thinking about disaster recovery? I'll come back to some of those details in a moment here. But basically, that's the table stake situation really to operate at all as a feed business today. Most companies, aside from cybersecurity, I would argue, is really an evolving, never-ending emerging technology that always has to be addressed. Some of the other ones, as we look at a lot of customers in the space, they have a pretty good hold on that.
The next layer up, core apps, is where we see a lot of companies struggling in terms of how they think about that. So that's all the tools from CRM, ERP, HR management systems, all the other operational tools. How does that all come together? And yet the real goal here is to make sure that the workflows in your software, this is the key word, reflect the actual standard within the business, and it's properly integrated throughout. So often we see companies that, because of the tight margin nature of the business, of the space, they customize everything for everyone.
Software technology is going to hit a lot of roadblocks when you can't standardize your process throughout. Of course, there's going to be different ways of doing things, but having a different process for each customer really won't work well from a tech perspective when you really get down to trying to build something that's reflective. So the goal is to make the workflows and integration reflective of what's the actual real-life standard process. And the reason that you want to do that is to let you get, I'll call it, transactional coverage of your whole business. All the data is being then created, and you don't have gaps in that.
There's no shadow IT, there's no Excel, and you actually have that full coverage of data about your business. And we do typically think about decoupling these different areas. Of course, some of these tools have reports, and there's specific data for that particular tool. Not saying to do away with that. Those serve an important role as well.
But the next layer up, data, is really when you want to aggregate different data sources from all these different tools to then bring back into your KPIs that I just mentioned. And you have all those transactions flowing into your different data, we'll call it data lake house for the moment, your data aggregation layer, and then you can implement analytics tools sitting on top of that aggregation layer. Using your KPIs, you now basically have your scoreboard using data that's actually reflective of your business. The first big win companies usually get is getting there, and we're seeing in the space, is getting that core apps transactional layer in place. That's usually the big first rock that we've got to address and get through.
The next one is truly getting that data visibility layer. So many companies are really still operating on intuition and not really operating from a data-driven way. So once you have that scoreboard, you can now test things in the business and actually have an understanding from a data perspective of did that actually work? Did that grow the customer base? Did that improve efficiencies?
And without this sort of approach, it's all intuition, which is a dangerous game, especially in such a low-margin space. So you've got your scoreboard now. AI can now come in and start optimizing on these three different layers. So there's many different applications of AI, but the scoreboard is what actually gets you the visibility into understanding if you're driving those improvements throughout the business. The other key part to think about AI is if you're building anything custom, and we'll get into more of those details in a moment, you're going to be heavily reliant on high-quality, detailed data that's flowing in real time to your system.
Without that, building anything custom AI is really not realistic. With that said, you can purchase off-the-shelf tools that have been pre-trained and have their own AI embedded. So we'll talk a little bit more about how that comes together. But this is really the full digital pyramid that we think about when we think about the path forward for a lot of the folks in the space. So we'll jump in a little bit more detail around what each of these means, but I mentioned earlier that the infrastructure layer is really where the digital world meets the physical one.
So where are you physically storing all these technologies? Is it sitting on your own servers on your local location? Is it sitting in the cloud? Is it a hybrid situation that you're working with? Networking connectivity is basically making sure that we have secure networks communicating with one another.
Starlink is one of the emerging technologies in that space that's really enabling, especially a lot of more remote locations, which we see in feed pretty commonly, more and more common. Disaster recovery is one that is often overlooked. So this is basically a situation where, imagine you have on-prem servers and you are not backing up that data remotely. If you have a tornado or a cyber attack or some sort of extreme event, and the servers are destroyed, you will not be able to get that data back, which is really a crippling situation for a lot of companies. This can even mean weeks, months of downtime, which, as we all know, no one can really afford that at any production location.
Cybersecurity is really thinking about it from two perspectives. One is more of if you think of a castle, standing up that castle and having the walls as high as possible, and that no one can breach your ecosystem. The second part, which is often overlooked, is how do you think about the recovery process and very closely related to disaster recovery, how do you mitigate and recover in the event that there is a cyber attack? A lot of firms, and we'll recommend that you actually think about it from two perspectives and make the assumption that you will get hacked at some point in time. It's a terrible thought to think about, but it's amazing how many attacks are happening, especially in the feed space lately.
And there just has to be mitigation plans in place that can quickly get you stood back up, lock out the intruders, and get your business stood back up in hours, days, as opposed to weeks, months, which I'm sure financially, we can all understand and appreciate the value of that. So I'm not going to spend too much time on this slide, but this is just an overview of the different technologies that sit in that core application layer. So you've got your enterprise resource planning. Most of the folks here are probably pretty familiar with that. That's kind of the heartbeat of a lot of these businesses.
You have your CRM, which is uniquely managed in the feed space when you think about some of the other industries that use CRM in more of a traditional sense, given the tight connection to the geographic way that feed businesses grow. And then you've got supply chain logistics, so everything from transportation management systems to warehouse management systems, and then other different parts of operational software. So formulation, more detailed inventory management, and on. And then custom apps to bring that all together. Of course, HR and payroll, I think is a pretty straightforward concept.
So- How this all should work together is, really, there's a couple of things. Like I said, it's got to be, there first has to be standards in the way you're doing business and standard workflows, and integration throughout. So when you have that, you avoid duplicate data entry. When we think about labor costs and time spent for expensive resources, manually entering transactions in one system, in system one, then manually entering the same transaction in system two, then manually entering transactions in system three, the sheer amount of time alone makes that really challenging, not to mention it erodes users’ confidence that all of this is supposed to help their day-to-day work. Manual workarounds is another one that's really common and pretty challenging thing that we want to stay away from.
The idea here is the biggest problem with manual workaround is there's two risks that evolve out of that. First, the people doing that work can almost hold the business hostage because it all lives in their heads. The second one is that you're getting gaps in your transactions that are supposed to be reflective of your business. So if that's sitting in manual, also to be clear, also includes Excel-based items as well. I love Excel as much as the next person, but would not recommend you treat that as a transactional system.
It is an analytical system that you can play with data and learn more and basically run more detail analyzing those transactions, but should not be used as an actual transactional tool. Same reason as that. If you do that, it’s difficult to drive standards and reduce gaps in your data that's reflective of your business. So this is a view of how this can come together from a data perspective. You have your different sources of tools on the left, and then you have a data lake house, and there are a lot of terms around data warehouses and data lakes and data lake houses.
So I'll clarify that one real quick. So, the data warehouse is the more traditional data aggregation tool that really stores data in a more like a database fashion. So, you have rows and columns, much like an Excel sheet, representing tables in databases, and that's for structured data. You have data lake, which is a way that you can store data that is unstructured. So unstructured would be image files, it would be long-form content, it would be audio files, those sorts of things that, or video files is another one that might be a little bit more relevant if you're thinking about cameras on your site that are tracking safety and quality and things like that.
Having those in one place, you have data warehouse for structured data, data lake for unstructured data, bringing those together, a lot of leaders in the space will now have data lake houses, which combines the two of them and enables you to connect those up so you can run analytics collectively across both. Another thing to note about this is that each one of these tools, your ERP, your automation system, your HRMS, each one of those is going to store the same data fields, likely in different formats. So my ERP might track a date, for example, using 12-28-88. My CRM might go no dashes and just straight numbers. The next one might be slashes.
So when we want to run consistent analytics against all of that data, it all has to be transformed via these pipelines that come from these tools into that data lake and transformed into a way, and standardized in a way that you're comparing apples to apples. So that then enables you to run analytics against the full data set and not just components of it. Once you have all that data standardized for apples-to-apples comparisons, you can then do all kinds of different things with that data. Whether that's integrating with different partners, Excel, manipulating that and analyzing that using Excel or building more traditional reports like Power BI or Tableau that sit on top of that data. So now we're moving on to AI, and how do we think about that?
So very much would recommend a crawl, walk, run approach when thinking about that. We work with a lot of companies where each individual in the company has their own account with a large language model, not a company-wide account or an enterprise account, which is a very dangerous game to play because they are likely sharing proprietary data with the large language models. If you don’t have that arrangement sorted out with an enterprise account, they may reserve the right to train future models off your data, which is a very concerning thought. I think everyone would agree with that. And the starting point is to really define that secure account and then have clear playbooks about what leadership expects people to use.
Simply just handing the tool to someone and saying, "Go." Especially, a lot of folks are going to have issues in understanding how do I use this tool in the best way? So simple stuff like playbooks and different use cases that you expect the people to use is really a key component, really more of a change management issue to get the most out of those types of tools. The next step up in maturity is thinking about agentic AI and different task automation tools. So I gave that agentic AI example earlier, but there's literally thousands of similar ones that would be, some are more or less applicable to your business. But the idea there is that you can really start driving some automation and shortening and finding more efficiencies in the way those transactions and workflows are moving through your business.
This is very dependent on your data being clean, usable, and timely. If you don't have that, agentic AI is going to fall down very quickly, and you’re going to have more problems than with the regular workflow. More advanced is thinking about things like digital twins and some of the more enterprise AI type tools. So you're really now getting into that optimization step of the business. Some of the partners that we've worked with on the digital twins are looking at, these are companies that have been around 20-ish years, and they have every single project's been under a six-month ROI, which is pretty impressive when you think about, number one, the cost of that and how quickly they're able to get that ROI back.
This is one where the data is very important, but there are ways around it. So I wouldn't necessarily say you want to wait till your data is totally in pristine shape. You will literally never get there. So it's very much encouraged to take this in a stepwise approach. In doing these types of agentic and digital twin projects, you're going to identify the data areas that are unclean, and then you can go fix those in the source system as opposed to cleaning them up after the fact.
So, by source system, I mean if it's in the ERP or the control automation system or your quality system, that's where that transaction and process needs to get cleaned up, such that when it flows into these more advanced data and AI tools, you've kind of got that fixed from the source. And then looking at fully custom AI is heavily dependent on clean data and a lot of it. That's the other part about anything really custom is you really need years and years of data to do that well, or you'll be spending a lot of time training models unnecessarily, not really getting the outcome you're looking for. So that is very much use case driven. Some ambitious companies look to start with fully custom AI.
The key part of doing that is having a clear idea of what outcome it is that you're seeking, as opposed to the buzzwords and excitement of what this can bring. Because these are often more expensive projects, and the clear business case really has to be there before you take that on. So when these pieces aren't in place, you can have what I'll call the reverse flywheel. Your users don't have clean data in anything they're looking at. They're forced to enter everything multiple times in each system.
Trust decreases. Management doesn't have visibility into what they're trying to understand about the business. You're kind of getting the worst of all worlds. And then on top of that, you're making investments in these tools and wondering where all this investment's going and why aren't the outcomes there. So when you bring all these pieces together in a way that works, first making sure your tools are connected and represent real-life workflows builds trust in the tools.
Users are getting more familiar from a training perspective. The tools can enforce correct data entry, building trust in the data. Leadership's now getting visibility into the business. They're becoming now a more data-driven company where we can try new things and understand what actually is working and what isn't. And now you get this faster decision-making loop, and it's a very highly competitive advantage when this whole thing comes together in a way that enables you to really, at the end of the day, move much more quickly and get those outcomes from what you're looking for.
So some key takeaways today. I think I just talked through about the first part there of really becoming more of a data-driven organization from both a tactical level and a more leadership level. The second takeaway is that, even though AI is exciting, we don’t know how fast it's going to move. It's very important that we take one step at a time and build those foundational tools in place. Stagnation is a risk.
Anyone who thinks that you can just wait for AI to solve every problem is playing a very risky path to choose. So we would very much recommend that you start on that journey of those earlier steps of getting your ERP in place, getting those standards in place, getting that visibility in place so you have the scoreboard before you start taking on some of those more advanced things. So that's it for me today. Thanks for the time, and looking forward to any discussions later.
















