I hear a version of the "AI will change everything" pitch a dozen times a month. It will result, I’m told, in real-time visibility, predictive insights and faster decisions.
Yet a sophisticated artificial intelligence pitch doesn't make a company worth backing. I want a founder who can tell me where the data came from, and walk me from "we spotted a problem" to "here's what we did about it."
That path from problem to action tracks against the same five pillars I use to evaluate a company: data, integration, portability, real differentiation and vision. Of course, I look at other criteria, including finances, markets, customer pipeline and leadership, but these five are the ones I weigh most heavily. They weren’t created in a whiteboard exercise. I built them by sitting across from founders long enough to see who could back up the pitch and who was simply reciting one.
Data first. The companies that held up under pressure were working from something real: a sensor reading, device log, or actual event in the physical world. The ones that didn't were reconstructing data after the fact from systems that never agreed with each other in the first place.
That distinction becomes even more important as AI moves from identifying problems to acting on them. Venu Gutlapalli, chief executive officer of Tag-N-Trac, believes that as agentic AI matures, the dashboard-watching will disappear. Routine monitoring and execution will happen without anyone babysitting a screen. The control tower will stop reacting and start resolving, creating what Venu calls an autonomous resolution hub.
But such a hub is only as reliable as the data feeding it. That’s why Gutlapalli gives pharma leaders a simple piece of advice: Get the data right before you build the control tower. High-quality sensor data is the foundation. The platform comes platform second. When a founder's vision of the future lines up with how the company operates today, I pay attention.
Integration comes next. Can the technology work within existing enterprise systems and operating processes, or does it create one more isolated source of information?
For me, the answer is straightforward. It must fit into the way a company already operates, and turn information into action where the work is happening. If it creates another dashboard or data stream that someone must reconcile by hand, it isn't integrated.
CargoSense shows what this looks like in practice. CEO Rich Kilmer believes reliable data must do more than identify a pattern. It must confirm that the pattern is happening in the physical world, and connect it to the operational context around it. CargoSense has built a model that can predict with strong accuracy whether a shipment will arrive on time that day.
The system pairs real-time carrier event scans with live location and weather data, then weighs how long the shipment can withstand those conditions before something breaks down. Combining these inputs allows it to separate a routine blip from an event that requires human attention.
Kilmer describes the goal as keeping a "human on the loop." AI handles the routine analysis and response, then pulls a person in exactly when judgment is required. Instead of cutting people out, it's about placing them where they matter. This is where promising technology becomes a real business or doesn't. A strong model isn't enough; the technology must connect data, operational context, existing workflow, and the people responsible for making decisions. If it can't do that, it won't scale, Kilmer believes.
Then I look at portability. Can the technology move into a new industry or operating environment without requiring the company to start over? I believe it must.
Take Controlant. Its platform, built around internet-of-things sensors and real-time visibility, underpins temperature-sensitive monitoring for pharmaceutical and life sciences shipments. It’s the same monitoring approach that extends into food and beverage supply chains. This ability to carry the underlying platform across industries with very different risk profiles and regulatory demands is the portability I look for.
Real differentiation is criteria number four. Having a "moat" is no longer an answer. I must know how deep that moat is.
A patent provides protection until someone designs around it or it expires. Years of validated data is harder to copy. Nobody can manufacture that history overnight. Scale may be the most durable advantage of all: more deployed integrated devices, and richer data. How a founder answers this differentiation question tells me whether that person truly understands the company's advantage.
Vision comes last, but it matters a lot. I want to know what a founder thinks will happen as sensor data and AI converge over the next two or three years. The roadmap slide matters less than the vision behind it. Is this person building toward autonomy, or simply automating what already exists?
The companies mentioned above are solving different problems in different corners of the supply chain. But they arrive at the same conclusion: Visibility isn't the finish line anymore. The bar now is understanding what’s happening, determining what it means, and fixing it.
That direction extends well beyond supply chain visibility. Gartner named physical AI, systems that sense and act in the physical world, as one of its top strategic technology trends for 2026. Recognizing where the market is headed is the easy part. Identifying the companies capable of leading it is harder.
When those five pillars line up, I'm no longer evaluating a pitch. I'm looking at a company I'll fund. Miss even one, though, and let's just call it a good conversation.
Joe Volpe is vice president, managing director and founding partner at Merck Global Health Innovation Fund (MGHIF).
