Turning Computer Vision Into Real‑World Value at Enterprise Scale – with Joseph Nelson of Roboflow

Summary

---

Episode Summary: Turning Computer Vision Into Real‑World Value at Enterprise Scale

Podcast: The AI in Business Podcast
Host: Daniel Faggella
Guest: Joseph Nelson, CEO and Co-founder, Roboflow
Date: April 15, 2026

---

Overview of the Episode

This episode centers on the real-world deployment of computer vision (visual AI) at enterprise scale, with insights from Joseph Nelson of Roboflow. The discussion identifies why computer vision often succeeds in controlled lab settings but struggles to deliver similar value in messy, variable real-world operations. Joseph explains the practical steps, best practices, and biggest obstacles organizations encounter when scaling computer vision, and shares stories of successful deployments and lessons from the field. The conversation provides non-technical business leaders with guidance on finding use cases, aligning strategy, building executive buy-in, and moving from experimentation to scalable ROI.

---

Key Discussion Points & Insights

1. The State and Promise of Visual AI in Industry

• Real-World Value Creation:

  • Visual AI brings human-like interpretation to physical environments, making sense of image and video data for manufacturing, logistics, retail, and healthcare.
  • As Joseph highlights:

    "What is new is the way with which you can do and get to value faster than ever before. Cameras are proliferating, compute is getting better, there's more data than ever... if I'm a manufacturer... every part of my process for producing that product needs to be done with visual certainty." [02:11]

• Benefits of Progressive Quality Assurance:

  • Moving from end-of-line inspection to visual verification at every process stage allows earlier detection of issues, preventing wasted materials and unnecessary slowdowns.
  • Quote:

    "With something like visual AI, the barrier to entry of teaching a system to understand how something should look... is getting ever easier." [03:49]

2. Barriers and Bottlenecks in Deploying Computer Vision

• Main Challenges Identified:

  • Acquiring the right data: Enterprises often lack sufficient, relevant visual data for their specific problems.
  • Model adaptation: Off-the-shelf models rarely fit unique products/processes; models need fine-tuning or bespoke training.
  • Operational integration: Insights must trigger real business actions—tying intelligence into systems like MES (Manufacturing Execution Systems) or logistics platforms.
  • As Joseph summarizes:

    "You have the technology... but you still need to roll out processes and bring people along in that journey." [09:11]

• Technology is Only Part of the Solution:

  • The most successful deployments combine technology, process design, and people management, often through a “center of excellence” or specialized team to champion and scale projects.

3. The Shift in Approaches: From Task-Specific to Generalized Intelligence

• Edge and Embodied AI:

  • Traditional machines performed single, repetitive tasks. The new promise is edge-deployed, learning-enabled systems that adapt across varied scenarios.
  • Joseph explains the spectrum:

    "On one end... you have very task specific, single job... On the other end... you have general embodied intelligence... it can learn even the jobs to be done." [12:00-13:15]

• Real-World Use Cases Approaching “General” Intelligence:

  • More enterprises now run real-time computer vision at the edge—on oil rigs, in logistics centers, on railroads, etc.
  • Safety, throughput, and predictive maintenance are dominant drivers.

4. Concrete Use Cases and Industry Examples

• Railways:

  • Preventative maintenance, derailment prevention, and container tracking—visual AI provides operational safety and protects multi-billion-dollar assets.
  • Quote:

    "Keeping track of all those containers, keeping the trains on the train tracks, these are all visually intensive problems." [14:15]

• Manufacturing & Medical Devices:

  • Example of IV bag production:
    • Traditional methods failed to detect tiny contaminants; new vision systems using modern AI can inspect bags at production pace, detecting particles as small as 200 microns.
    • Quote:

      "Now they're able to process bags at machine speed... and identify particulate matter that's as small as 200 microns..." [20:46]

• Consumer Goods and CPG:

  • Real-time visual validation of packaging and products has replaced slow, manual, or barcode-driven checks.

• Environmental Monitoring:

  • Visual AI for oil & gas flares reduces environmental risks and improves regulation compliance.

5. Why Lab Successes Don’t Translate Directly to Production

• Real-World Unpredictability:
  • Computer vision systems encounter endless variability—lighting, camera positions, changes in production, etc.—that labs can't simulate.
  • Continuous learning (active learning) is critical for robustness.
  • Joseph:

    "It's not a question of if something's going to go wrong, it's what you do when it does... the longer the visual system is running, the more accurate and better it gets." [23:01]

6. Unlocking Ecosystem Value and Scaling Computer Vision

• Systemic Approach:

  • Companies gain maximum value when they treat visual AI as a living, learning system fully woven into business processes—not as one-off deployments.
  • Example: Manufacturing models reused in remanufacturing lines to immediately create value at scale.

• Returns to Scale:

  • The greater the rollout, the more cross-functional and compounding the value becomes.
  • Joseph’s challenge:

    "What's the foundation model you want to have—the AI system that's yours that understands your slice of the world?" [25:00]

7. Lessons for Enterprise Leaders

• Don’t “Boil the Ocean”—Pair Big Vision with Quick Wins:
  • The classic failure is trying to scale everything at once without a solid proof point.
  • Joseph’s “barbell strategy”:
    • Get executive buy-in and ambition at the top.
    • Pair this with a “material first thing” (a focused use case that delivers visible results).
    • Use rapid success to build momentum and credibility for broader rollouts.

  • "You want to have the executive buy in and vision paired with the true ability to operationalize on the ground what's possible." [27:11]
    "When you're in the real production environments and you're willing to go to those places... that's where you really get the opportunity to unlock really impressive value." [28:00]

8. Step-by-Step Advice for the Executive Suite

• Build a Center of Excellence:

  • Assemble a center of excellence (COE) to collect and prioritize use cases. Embed these champions into relevant business units.

  • "Commonly we're seeing... creating the center of excellence and developing a hub and spokes model..." [31:13]

• Collaborate Across the Organization:

  • Engaging plant managers, regional managers, and directors ensures initiatives have grassroots input and practical impact.

• Value & Effort Analysis:

  • Score opportunities based on value potential and feasibility, then execute and scale based on wins.

• Partner With Vendors and Internal Champions:

  • Both internal and external expertise are usually needed for successful, sustainable deployments.

9. Urgency and Adoption Curve

• Window of Differentiation is Shrinking:
  • Joseph describes the adoption cycle:

    "The good news is there's still time to be part of the early majority and create differentiation. Maybe the bad news is the window is waning." [37:10]

---

Notable Quotes & Timestamps

• On where visual AI breaks down:

  "It really comes down to having the right tooling, the right models, the right team, the right expertise to guide and produce value..."
  – Joseph Nelson [07:44]

• On the evolution of AI systems:

  "You have a system that perfectly identifies... now the next thing is: how do I create action? How do I take and improve my business with this new insight?"
  – Joseph Nelson [06:55]

• On scale and learning:

  "The greater scale you operate at with visual AI, the more benefit you get because your system has better context of your full business."
  – Joseph Nelson [24:55]

• On living, learning AI systems:

  "It's a living system. It's not a static, just eyes watching something. It's living, it's learning and it is very much the ecosystem."
  – Daniel Faggella [24:23]

• On executive buy-in and strategy:

  "All starts with a really strong advocate... who is able to bring people, processes and technology together to realize business outcomes."
  – Joseph Nelson [10:15]

• On the risk of non-adoption:

  "If we don't invest in AI, if we don't become a part of the modern industrial revolution, our business will not exist in as short as a decade."
  – Joseph Nelson [30:53]

• On the urgency to start now:

  "The cost and risk of doing nothing has never been bigger because competitors already are."
  – Joseph Nelson [36:30]

• On the adoption cycle:

  "Now you're seeing early majority get on board. And the good news is there's still time to be part of the early majority and create differentiation. The bad news is the window is waning."
  – Joseph Nelson [37:10]

---

Timestamps for Important Segments

• [02:11] – Why value is becoming accessible faster in computer vision
• [04:50] – Benefits of progressive QA throughout production
• [06:00-08:20] – Three core deployment challenges: data, models, operationalization
• [09:11] – People, process, and tech: the three pillars of enterprise AI success
• [13:15-16:01] – Use cases in railways, manufacturing, and safety
• [19:45-22:21] – Specific vision examples in consumer goods/manufacturing
• [23:01] – Why lab results don’t always translate; the importance of continuous learning
• [24:55] – Scale advantages and exploiting model reuse
• [27:11-29:53] – Barbell strategy for rolling out computer vision at scale
• [30:53-35:25] – Step-by-step for executives: COE, champion selection, value analysis, scaling
• [36:30-37:10] – Urgency in adoption; early majority vs. laggards

---

Memorable Moments

• Railroad Use Case:
  BNSF Railways uses visual AI for preventative maintenance over 30,000 miles of track, saving money and lives—a showcase of vision in high-value, high-risk environments.

• Medical IV Bags Breakthrough:
  Machine learning now inspects for particles the size of a grain of sand at full production speed, overcoming a decade-old inspection bottleneck and significantly improving product safety.

• Barbell Strategy:
  Successful companies pair ambitious executive vision with small, grounded, high-impact use cases to prove value and build momentum for wider rollout.

• AI as an Ecosystem:
  Scaling value isn’t possible without treating computer vision as a living, learning, organization-wide ecosystem—spanning hardware, models, operations, and people.

---

Conclusion

Joseph Nelson’s insights highlight that enterprise computer vision has reached a pivotal, mature stage—but scaling it beyond labs to real facilities still requires careful orchestration of technology, people, process, and business change. Visionary leadership, smart use case selection, and organizational flexibility to iterate are all key for unlocking ROI and competitive advantage. The time to act is now—before the window closes and industry leaders surge ahead.

---

Transcript

[A]

Foreign.

[B]

Welcome everyone to the Emerge AI in Business podcast. Today's guest is Joseph Nelson, co founder and CEO at roboflow. Roboflow is a computer vision software company that provides tools for collecting, annotating, training and deploying machine learning models for image and video analysis. Its platform is used to build and run vision applications in cloud, cloud and edge environments. Joseph joins us on the show to explain why visual AI works in control settings but falters in real operations, largely due to missing visual data models that don't reflect real world variability and the difficulty of tying insights into existing systems. He lays bare how companies that solve these basics see earlier detection, fewer slowdowns and safer facilities, and how pairing big executive ambition with a single well chosen first deployment accelerates meaningful progress. Today's episode is sponsored by roboflow for our solutions partners. Position your brand alongside the Fortune 500 leaders defining the enterprise AI roadmap for the opportunity to showcase your solution to the executives currently funding and scaling global initiatives. Partner with Emerge. Secure your [e-mail] partner that's go.emerj.com V-A-R-R now the conversation with Joseph. Joseph, welcome. Thank you for joining us today.

[A]

Thanks for having me.

[B]

So computer vision is nothing new to our audience. It's been around for a long time but I think we leaders are struggling to understand the importance is they're struggling to see the journey, the real clear picture of real maturity and where it's working, where it's not working, how the value can be created in new ways. And across our past conversations on Emerge around computer vision we've seen a similar trend where it's the computer vision is there, it works in the lab, but deployment has become difficult and I'm very interested in seeing what you seeing as the points where it gets complicated and it breaks down Running computer vision today

[A]

so computer vision and more broadly now visual AI, physical AI is all things that have to do with making sense of the real world. The way that people can see and understand the world is what computer vision and visual AI enables us to do. And vision tends to be most useful in places where you have some physical environment so you can think like visual inspection, defect detection, process understanding in sectors like manufacturing, logistics, retail, healthcare and where things are most challenging. Perhaps we should talk about in fact how broadly deployed things are today and then we can talk about maybe the things that inhibit that going further. So vision right now is pretty broad scale. It's actually not super super new. But what is new is the way with which you can do and get to value faster than ever before. Cameras are proliferating, Compute is getting better, there's more data than ever. And all of that data lives in unstructured video or images. And so if I'm a manufacturer and I'm making a set of products, I'm making a set of goods and services, maybe I'm a vehicle manufacturer. Every part of my process for producing that product needs to be done with visual certainty. Traditionally, that's done at end of line inspection, and there are inspections throughout the production process. But inspecting every single part perfectly at each step of the process may either be cost prohibitive or challenging, or maybe it's not economic. With something like visual AI, the barrier to entry of teaching a system to understand how something should look, should be built, what the raw materials look like, Are folks safe? Are they being made in the right amounts? Our parts fit together the right way is getting ever easier. And so you're starting to see things like progressive quality assurance instead of just end of line qa, meaning each step of the process, you're like validating that the right raw material was in place.

[B]

And that saves obviously money earlier. Because if you only test at the end of the line, you might have wasted a lot of raw materials where you could have stopped the process earlier.

[A]

Yes, 100%. I mean, some of the biggest things that you see for value tend to be things like, I want to increase my throughput without dramatically increasing my costs. I want to optimize labor. I mean, in a lot of the sectors where vision is most useful, they're the physical world. And these are jobs where there's concerns about will there be workers of the next generation at the right amount, at the right skills, at the right scale to do this work that there is to be done. In a lot of ways, automation is the only answer we have.

[B]

And we hear that because tribal knowledge is an issue at the moment, it's not easy to get that transferred. There is just not the workforce. So this actually solves for quite a few problems. And what I'd like to know is computer vision, at the moment it's being used, but where is it breaking down? Where is it not creating the value that you obviously envision and have seen created.

[A]

Some of the biggest challenges to setting up vision in production tends to be with first and foremost ensuring you have data and eyes on the problem of interest. So do you have video footage or images of the thing that you want to improve? So let's return to the manufacturer. If you're making electric vehicles and you're producing batteries as a part of that. Do you have eyes on the cross section of the battery? Do you have eyes on the installation at each step of the process? Do you have eyes on your stamping presses? So one of the first things is ensuring that you have the right data prepared the right way for a machine learning model. Then once you have the data, you actually want a model that's going to understand the scenarios and situations that you need to operate in. This is where you might use an existing model. As machine learning models get better, even more likely though, you'll likely fine tune or train and create your own model for your slice of the world. You make a car that no one else makes, and so your products are unique and special and differentiated. And so training a model to know what your products are meant to look like is kind of a second part of a challenge. The third part is, okay, great, I've got eyes on the problem. I've got a thing that understands what I want to know. How do I run it, and how do I get value and insights? In other words, if you have a system that can perfectly identify, hey, there's, I don't know, four screws, and there's meant to be eight, the next natural thing is how do I create action? How do I take and improve my business with this new insight that I have? So you want to run that model close to the problem that you're analyzing and connect it to your downstream systems. Whether that's a manufacturing execution system, maybe that's a transportation operating system, maybe that's a product catalog for keeping logs of existing inventory and returns. All of these systems allow you to run your business better, but they're starved of knowing the right insights from visual intelligence that could take place. So bringing all of those things together, it really comes down to having the right tooling, the right models, the right team, the right expertise to guide and produce value in terms of the places that one's business can most benefit.

[B]

If I'm hearing you correctly, it's a matter of, number one, understanding your use case, understanding where you where there is data to awaken, to actually create value. Number two is the edge. Do you have actual eyes on it? Do you have cameras? Do you have sensors? Number two, do you have the models trained to actually evaluate and give you insights on that? And then the actual application or implementation? What are you doing on the insights? You might have the insights, the intelligence, but how do you now actually action on that intelligence? So with that in mind, in those three processes, is it across the board that you're seeing, let's say bottlenecks in all three places. My instinct is that the models are there, it might be on the edge, and then the action side, that there might be bottlenecks at the moment.

[A]

Yeah, I mean, any sort of enterprise change, you have people, processes and technology and the technology are a lot of what we're talking about. Of there's been a stepwise change of what's possible. You know, previously you might not have been able to build a system that could understand the things that you want to understand in your business. The technology has made that possible. Well, you still need to have the processes in place to be able to roll out that technology, to connect your operational intelligence team, your operations research team, your manufacturing engineering team. And then the third piece is bringing people along in that journey. And so we commonly will work with standing up or existing centers of excellence within enterprises because they realize that visual AI and artificial intelligence in general is one of the largest transformative opportunities. It's the modern industrial revolution. And so to stand up a center of excellence and to be the person inside the enterprise to recognize that and to capitalize on it and to rally folks around that vision usually consists of a really strong advocate who then also is able to bring people, processes and technology together to realize those business outcomes. So we were talking about the technology piece, which of course is a requisite component, but you also need to have a good grasp of how the business should work or how it could work in the processes and the people who are already aware of and need to be brought along the journey to deploy these solutions at scale. So we see like folks that when they make this bet, they have rapid acceleration and that's exciting both in their business and in their career. And that's a really fun transformation to be a part of.

[B]

Have you seen a shift in how organizations actually approach vision and physical and even embodied AI?

[A]

There's been a shift, yeah. So when you think about things like physical and embodied AI, you're referring to systems that run often on the edge, rather than like cloud computer. And so here the common thing to picture is humanoid robots and so forth. But really even a single stamping press machine, or a single, in some ways very. What we'd consider simple, repeatable machines that exist on factory floors are also embodied intelligence of taking something and making it be more easily usable. And so what change is happening? Well, what you're seeing is the rise of visual AI is making it so that you can think about a spectrum where on one end of the spectrum, you have very task specific, single task, single job to be done. And that's like your stamping press where a piece of aluminum comes in, large machine stamps it, and now you have an aluminum door for a vehicle, does that same thing repeatably, reliably. There's still maybe some error sometimes with like if there was necking or problems in the stamping drift. But nonetheless there's like a very task specific. And we could call that embodied AI or embodied intelligence. It's a very simple, repeatable, maybe low IQ machine. On the other end of the spectrum you have general embodied intelligence. And this is like if you just took a machine and you dropped it into an environment, can it learn even the jobs to be done and then learn how to position itself to be successful in completing those jobs? And in some ways this is like the holy grail, like somewhat pie in the sky future that we're all kind of trending towards. But that's the spectrum. Those are two ends of the spectrum.

[B]

And I guess, I guess that embodied AI would really solve for problems where it's dangerous to have a human in certain situations. And you can have a human in the loop, ensuring that the robot is still. But the robot needs to be visually aware of his space spacing, his distance from things, and that might change. It should learn and it should be able to adapt. I'm trying to think, do you have specific use cases where you already, I mean, I know it's pie in the sky, but use cases that are quite close industries that are quite close to that.

[A]

So that's exactly what I was going to describe is if there's two ends of the spectrum and one end is very task specific work and then one end is very general work, what we see our customers able to do is take steps from the very task specific work towards the general work. That's what visual AI and intelligence is. If you have a more intelligent system, it can be more generally useful. And presently a little over half the Fortune 100 build with us today. And these are industrial enterprises with global operations. And the steps that they're taking are previously we didn't have a way to know and validate unstructured scenarios running at the edge real time in our facilities, but with higher quality intelligence running at 30, 60 plus frames per second on video feeds, even offline in remote environments, on oil rigs, in factories, in logistics centers, in freight forwarding, on railroads, you have the ability to really transform a given system to operate unlike it could before. So I mean, I think about there's A number of Class 1 railroads throughout the United States. BNSF is a Class 1 railroad and they have 30,000 miles of track and they move containers, millions of containers annually. And keeping track of all those containers, keeping the trains on the train tracks, these are all visually intensive problems. Rr anything from the wheels on our trains to the tracks themselves. Can we do preventative maintenance? Can we even find a problem before it results?

[B]

And it's safety and the value is pretty clear there. And it's across the board, it's money, it's safety. And especially when you're working on railways, trains aren't not expensive, they're extremely expensive. So if you can prevent a derailing, that's not just saving money, it's saving lives. The value is pretty clear.

[A]

You have billions of dollars deployed in these CAPEX projects and you can think about like the visual intelligence of ensuring you get the most out of that capital expenditure. And yeah, that could be an inspection to prevent derailment. It could be ensuring that you have the right production throughput so you have the right amounts of raw materials, preventing waste. It could be making sure you meet your production targets in the first place, which increases revenue and optimization. You mentioned safety. There's things around labor optimization and costs of ensuring operational expenditures are as efficient as possible. There's lives at stake. There's dangerous environments and dirty jobs that exist that in a lot of ways those jobs are tough to fill. And providing intelligence so that people can work in places that are safer. Or if you have to work in an environment that risks being unsafe, you have systems that can alert triage, prevent catastrophe from happening in the first place. That's things as simple as ensuring folks are in the right zones or wearing PPE as much as it is things that a human may not be keeping an eye on of a machine, a process that goes awry. And if everything's going according to plan, then there's no need to intervene. But at times systems are unpredictable and having the ability to know one of our customers has cameras they're placing on cranes to see from below, ensuring that where those cranes are operating in the products that they're moving is a safe time to do so. And if not, you should have automatic shutoff or you have customers that are doing environmental monitoring with an oil and gas where you need to have specific flare stack monitoring taking place which reduces the amount if you burn off a given material, it reduces the environmental impact of what can otherwise be a catastrophic process. So really, everywhere you look on a job site, on a production facility, in A logistics center. All of that is based on can we add automation, can we make things safer, can we make things cleaner, can we increase our throughput? And so many of those operational bottlenecks rely on having eyes on the problem, measurement of the problem, and then acting on when something is out of place or could be better in place.

[B]

And then thinking about this, people especially let's say in manufacturing, in warehousing, cameras have been used for a very long time. What is the shift now? Is it the possibility? I guess it would be a range where sometimes the edge is the problem. You may have many warehouses and your edge devices aren't all the same. Now you're trying to use one system, but you've got a problem with the data you have because it's not structured or it's not actually all in the same format. Describe for me all the, all the spots where this new AI capabilities is actually bringing, I guess, time to value a lot faster and also enabling companies to either take what they have and actually create value or implement things that weren't possible before.

[A]

Visual AI expands how you can leverage existing video or even adding new cameras to a scene because the value is so great that it justifies doing so.

[B]

It's worth it. Yeah.

[A]

The types of use cases. You know, if you're a consumer packaged goods company previously you would have to do everything with barcode scans because that was the only way to know to create regimen around what is a given product that was produced. With visual AI you can have much richer validation of the packaging or you can just see a product as it goes down a line in real time. And in fact, you don't even need to necessarily scan the UPC to know which product is being produced or you're matching and validating that like all parts of the packaging are correct. And that's just in cpg. In manufacturing, there's so many places where you have high accuracy, high velocity and sometimes even high variability problems. You know, one example that we solved recently was there is a company that produces IV bags. And when you produce an IV bag, of course you don't want there to be any particulate matter on the inside in that bag you're producing. And this company is bottlenecked by the rate at which they can inspect and ensure that there's not any particulate matter. And you might think, how hard a problem can that be? Vision's been around forever. Interestingly, they've tried to work on that problem with traditional machine vision approaches for the better part of a decade. But the big Unlock is with the advancements in transformers and real time vision, now they're able to process bags at machine speed, 30 plus frames per second. They produce a bag a second and identify particulate matter that's as small as 200 microns, which is like the size of a grain of sand, if you will. And so if you think about it, the advancements of machine learning and technology, now you can run this line at machine speed the whole time and you can meet greater production targets. And you can do that, which is not only saving lives of folks that are going to use this given medical product, but also increasing bottom line, because you're able to do greater amounts of production without having many, many, many people hand review and in some cases imperfectly review to try to find this tiny, tiny defect. So that's another place where AI now allows you to run faster and find smaller objects with higher accuracy. That's a good example of high velocity, high precision.

[B]

And is that technology on the edge side that's just improved or is it actually edge? And obviously it must be processing because it's a lot faster. So it's not just the edge, it's the processing as well. So it's just technology has really sort of caught up to the volume and scale that companies are running at and it's now being able to deploy that across. On scale. We hear this a lot across industry. Something works in the lab, but in the real world it's just there's too many variables. Is that still the problem, that it's working perfectly in the lab where all conditions are perfect, but in a real world conditions aren't perfect?

[A]

That's the thing where a lot of our technology shines. The world is varied and messy and different and you cannot predict all the ways that something will look or where things will go awry. And so something I actually like to remind people is it's not a question of if something's going to go wrong, it's what you do when it does. When you have a given system, you're rolling out and you're building automation. And so you want to build a system that responds well and is resilient and learns from a given failure. And so this is so called active learning, where if you deploy a visual system, the longer the visual system is running, the more accurate. Yeah, the more accurate and better it gets. Which is what's new in the AI era is that all these examples are continuously learning and continuously getting better. And so you have use cases that previously weren't possible to be solved, being solved Faster than was previously thought to be possible, automatically improving. And it just, it is a, the confluence of those things unlocks so much potential for businesses that we're seeing companies save hundreds of millions in wasted product, in warranties and in increased revenue because they're able to meet production targets that previously they were constrained by being able to do.

[B]

And it sounds to me that this is a living system. And I guess that's the shift as well. It's not a static, just eyes watching something. It's living, it's learning and it is very much the ecosystem. And if you don't say scale and deploy and implement across from the edge all the way to the decisioning and assigning, tasking, if you don't look at all of these spots where you need to increase scale and technology, you're not going to get that full benefit.

[A]

That's a really good point. The greater scale you operate at with visual AI, the more benefit you get because your system has better context of your full business. One of our customers, for example, that produces 3 million units per year, started building models for the production of those units. They had a really smart revelation which is that they have a remanufacturing part of their business too. And so customers will send in parts that might be damaged or could be returned or could be resold. The models that they trained for doing the manufacturing process at the first part could be reused for the remanufacturing process. And so by having scale, by having the system rolled out at so many facilities, they could use it in remanufacturing and have immediately understood all of the products that are being sent to them and be able to build smart workbenches and deploy that immediately and in high scale. And so if you think about these companies that are building kind of so called foundation models, I challenge enterprises. What's the foundation model? You want to have the AI system that's yours that understands your slice of the world. And what's really amazing is the platforms and technology that we're seeing inside our customers, built with where we will deliver for them, realizes that dream. So you make a very good point that the scale at which things operate allows you to not just, you know, realize more value because you're doing something at a higher scale, but you actually kind of get the returns to scale where you have an increasing onfair advantage.

[B]

And I guess if I'd had to push back a little bit is if you had this whole system, let's say you've implemented this entire system, there's still obviously A lift from the company's side. There's still places where it gets tricky. Where are those spots that even if you've got this whole system thought through that you still find companies struggling? Is there a trend or does it really depend on industry and organization?

[A]

The places that people struggle tend to be boiling the ocean without first assembling a good win. So we kind of see this barbell strategy play out where you have executive buy in to know and in some ways dream big about all the potential of what's possible when you have visual AI in their business. And you need to pair that with what is a material first thing we're going to do. And so all the way down on a production line, all the way down in an individual facility and that becomes a proof point that the executive team can point to evangelize and expand rapidly. So one end of the barbell is you're working with the CIO to declare what's possible with visual AI transformation. The other end of the barbell is you have a potentially even early career person on the end of the line doing an inspection who notices again, notices a gap. And the value that you create out of that end of line automated inspection is representative of the transformation that the CIO sees is possible. And so what I often recommend is you want to have the executive buy in and vision paired with the true ability to operationalize on the ground what's possible. And so that's we often see, for example, pairing that with forward deployed engineering or what we call field engineers who work directly on customer sites, in customer facilities, in logistics centers, and are rolling up their sleeves, getting their hands dirty, wearing their steel toed boots and ensuring that the technology is delivering the outcomes that it's promised to be able to do. And that's sort of the stuff you were talking about earlier of like there's this last mile stuff, there's gotcha stuff, there's this does it work outside the lab stuff. I challenge you. I mean, look, talk to the folks that we've been able to work with. When you're in the real production environments and you're willing to go to those places of meeting customers where they are, that's where you really get the opportunity to unlock really impressive value. So that gets to some of the threads we were talking around. The executive level can think about the center of excellence, the individual deployment first starting point can pair with the field engineering team to get a first use case live. And then you have the opportunity to benefit from scale where it's like, okay, great, we've got this one use case. We know it's a good proof point. There's confidence, there's excitement, there's buy in. We're ready to scale.

[B]

I'm going to put you in a position you're in. Let's choose an industry. Well, it doesn't have to be a specific industry. You're in the boardroom and, and your, your body's telling you I, you need AI. Everybody's saying it and you are the CIO or the CEO even, and you need to make a decision. And your company or your industry is obviously geared towards computer vision, Physical AI, embodied AI. What are those first steps? What would you say to that executive suite? What do you do on day one to actually start this process? You've, you've described it already. You need to actually pinpoint, you've got to dream big, but then pinpoint the realities. But what is that first step? Do you, do you just brainstorm and get ideas or do you go down to your field workers and ask them what are the frequent questions that are coming up? What are the frequent problems that you see? What would you advise?

[A]

The very first thing that I recommend when I hear folks that are being asked by the board to think about AI is to echo that very loudly with if we don't invest in AI, if we don't become a part of the modern industrial revolution, our business will not exist in as short as a decade. And so the first thing is, I hear you and we couldn't agree more. For the potential, you want to be the only, you want to be the last person riding a horse when everyone's on their car. So the first thing is echoing that first step is recognition of the size, the opportunity in terms of realizing the potential, like operationalizing practically what does it look like commonly we're seeing pairing closely with creating the center of excellence and developing a hub and spokes model where you have someone, the CIO is leading the executive sponsorship of what's possible and you're creating a COE whose mission is to find, solicit and collect the use cases that exist in the ultimate business. Where you ultimately want to land in 5, 10 years is the technology is so diffused that you don't need the COE anymore. But you can't start there because you have to start with assembling the use cases where it's going to be most impactful for your business. You have to identify the champions across those parts of the organization that would be good leaders within the respective business units. And you have to find the Partners in technology that are going to allow you to realize those goals and that potential. Once you establish that kind of hub and spokes model with the center of Excellence, then you do what you mentioned, which is start to solicit use cases and start to do the effort versus value ratio. And if you don't embed from the COE into the business units, then you'll have a group of folks that meet quarterly who talk about all the potential of what's possible without any business change. So you have to embed into the business unit of what the results are going to be. You describe this in the context of a physical AI business. If you are a manufacturer, you have dozens, thousands of facilities across North America and globally, and you realize where things can go. You need your regional managers, you need your plant managers, you need directors to be on that journey. Otherwise you're going to have a boardroom echo without results to show for it. And so the way to realize that is you've got your coe, you've got the individual bus, you've identified the highest promising valuable use cases, and then you rank and execute. You need partners as a part of that, both internal stakeholders and external partners. And you start to assemble those wins. And then what naturally happens is wins beget wins more folks. Pretty soon the COE gets bombarded with more use cases than it can take.

[B]

Everybody wants it.

[A]

And you get this. Yes, and you get this flywheel. And the folks that get involved in the COE realize that their colleagues are seeing vast progression because they're a key part of enabling the modern industrial revolution. And you have a CIO who looks brilliant in hindsight because of the ability to so effectively mobilize a team. So that's kind of like the progress and process. We will commonly be a part of boot camps where we help folks realize because just even education of what are our peers doing, what is possible in our industry, not even about our technology, just about the industry in general, what's taking place and offering that perspective. And like, almost like peer benchmarking, like how do we ensure we don't fall behind is often a really good first step because it helps solidify this question from we need AI to here's what we can actually do with AI, here's what others are doing with AI. We've been fortunate where we've been deploying solutions for five years in visual AI. And so we have a really good grasp of what folks are doing across market segments. I mean, millions of folks build with our technology now. And so you have such a natural sample of That's a good use case. That's a bad use case.

[B]

Exactly. You almost notice it before they notice it because you've seen it before. And when it's unique, I'm sure you get excited as well when there's a, when there's a new horizon of what you can do with your technology. Joseph, this has been really interesting to me. I think computer vision has been ripe for modernization, it's been ripe for scaling. I think a lot of companies have spoken before about how well it works in the lab, but struggling to get that deployment. And I think what I hear from you today is that it's, it's a living system, it's a learning system, and you need to treat it as an ecosystem. Not just, let's say, on the, on the level of where you implement it, from the edge device to the actual models and then the output, where the insights are getting actionable, but also within the organization. It needs to be a complete ecosystem where everybody buys in, where it's clear where the wins can be. And then you win at scale, as often you do in AI. Is there anything I missed?

[A]

I think that's great. I mean, the one thing I would, I think it's a great summary. I would note that visual AI in particular is a stepwise change for vision because as you mentioned, it continuously gets better the more that it sees. And the cost and risk of doing nothing has never been bigger because competitors already are. We're seeing it. And so the opportunity is, is there

[B]

for the taking and, and the time to value. There's no question anymore. It's. Are you saying time to value across? It's not just the leaders, the ones that are adopting early that are seeing the wins. When you step into the ring, you'll start making wins.

[A]

Yeah, I mean, I would describe it in the Gartner cycle of adoption, if it's now the early majority, the innovators, those dominoes fell three years ago with the rise of ChatGPT. Now you're seeing early majority get on board. And the good news is there's still time to be part of the early majority and create differentiation. Maybe the bad news is the window is waning. So I think the good organizations that are operating at high scale can really bring to bear a lot of their data and process in the physical world. But you're seeing the companies that are investing in visual AI capabilities and those that are even aligned to the rise of AI GE commons really see transformative tailwinds to their business.

[B]

Thank you. I think you've given the audience a lot to chew on. Some interesting things. I'm sure they all want to look at the industries and see where they can win. Thanks for spending the time with us today.

[A]

Thanks for having me.

[B]

Wrapping up today's episode, I think there are three key takeaways from our conversation with Joseph Nelson. First, visual AI only works when companies have dependable visibility into into the processes they're trying to improve. Second, the impact shows up when that intelligence is wired into the systems that govern production quality and safety. And finally, the organizations that move fastest are the ones that pick top level ambition with one focused deployment that proves valid and sets the pattern for scale. If you have an AI solution, position your brand alongside the Fortune 500 leaders defining the enterprise AI roadmap for the opportunity to showcase your solution to the executive currently funding and scaling global initiatives. Partner with Emerge. Secure your [e-mail] partners that's go emerj.com bartner for further executive level analysis and to join our network of leaders delivering workflow impact with AI, visit emerge.com on behalf of the team at Image. We'll see you on the next episode.