James Deller explains why machine learning has become a fundamental requirement for running a serious company, and why most founders still treat it like a bonus feature rather than infrastructure.
James Deller has spent enough time developing machine learning products to realize what most founders overlook. It’s that technology stopped being impressive a while ago. No one applauds companies that use electricity. Machine learning is moving in the same direction, but many companies haven’t updated their expectations accordingly.
“The moment something works well and becomes boring, it’s already become a must-have,” says James Deller. “That’s just the current state of machine learning, and most companies still treat it like a party trick.”
From marketing lines to load-bearing infrastructure
James Deller founded his company, 1Touch, around a matching engine that pairs vendors with influencers. This was a job where you had to survive with real privacy restrictions and real money at stake. There was no room for the system to become a slog. If something goes wrong with the match, the issue affects the actual business relationship, not the pitch deck slides.
This kind of pressure teaches founders the gap between dressing up a product with “AI-powered” language and relying on something that actually works. James Deller said it clearly. The functions to be announced and the systems to be evaluated are not the same animal, even if they are described using the same words.
His reading of how this will impact the market is straightforward. Companies that use machine learning as plumbing rather than press releases are likely to still excel years from now. Everyone else will appear to have missed a turn that no one warned them about.
“The features you present and the system by which you are evaluated are two different things with the same name.”
The bar moved but no one sent a note
James Deller says most founders outside of serious technology companies don’t keep track of how much things have changed. Personalization, churn prediction, pattern discovery — five years ago, these were the real edge. Now, they are just roles in a functioning business.
He simply frames the stakes. If a competitor warns you of churn before it happens, and another team is still waiting for a quarterly report to know after the fact, that second team isn’t keeping up with innovation. They are behind in the basics of company management.
James Deller believes the board is not in control on this particular point. “Every meeting the board grills the founders about the runway,” he says. “They should ask about model adoption with the same seriousness. It’s no longer a happy question. It’s a survival question.”
“It’s not that innovation is slow; it’s that operations are slow. And that’s much harder to fix in a hurry.”
This is rarely a problem with the model. It’s a data discipline issue.
When founder James Deller talks about the stalemate in machine learning adoption, he assumes that the stagnation is technical: a lack of engineering talent or a lack of model sophistication. According to his experience advising companies that have overcome this very barrier, the real obstacles lie even further upstream. Either the data is messy and unstructured, or no one trusts it enough to act on it.
You can’t implement a sharp model into a business that doesn’t operate on evidence rather than opinion. This model doesn’t fix bad inputs, it just repeats them faster and more confidently.
As James Deller observes, the founders who actually derive value from machine learning are those who treat clean, reliable data as a prerequisite for their work, rather than a side project to be modified later while the model is running.
“Giving smart models messy business doesn’t give them intelligence. They only get confident nonsense.”
Models are not a substitute for judgment. It rearranges it.
James Deller carefully refutes a common misconception about technology: that machine learning does not eliminate the need for human judgment. It simply moves judgment elsewhere in the process.
The most difficult discussion at 1Touch wasn’t deciding which algorithm to use, he says. They were about what the model should actually be optimizing for, what trade-offs are acceptable, and — this is the part that people underestimate — recognizing when confident outputs were silently wrong in ways only humans would notice.
Founders who treat models as stand-ins for their own judgment, rather than as tools that still require judgment, tend to become more confident long before the market does.
“This model doesn’t take judgment out of the room; it just changes where judgment should appear.”
Privacy isn’t about slowing down — it’s part of the blueprint
By building a matching system that worked while respecting user privacy, James Deller learned something counterintuitive. That said, privacy and performance are not actually at odds with each other as people assume.
In his experience, constraints produce better designs. Building with privacy built in from day one requires more honest feature engineering and a more rigorous look at what data is actually needed versus data collected out of habit or laziness.
His warning to founders is direct. If you treat privacy as a legal checkbox, you’re building something that will require a complete rebuild the moment regulators catch up – and they always do.
“Privacy isn’t a fence around good machine learning; it’s part of the blueprint for building it right the first time.”
James Deller’s advice for founders starting today
In James Deller’s view, if machine learning isn’t already part of your core operating assumptions, you’re not paying attention, and you’re accepting a structural disadvantage over competitors who have already made the transition. That doesn’t mean bolting AI onto every screen for a press release that no one asked for.
That means going feature by feature and asking honest questions. Where should large-scale pattern recognition be done where the tasks currently performed by humans or spreadsheets are slow and unreliable?
“Machine learning stopped being an interesting part of the pitch a long time ago,” says James Deller. “That’s what serious companies are like now. Founders with such accolades are building companies that will still be relevant long after the novelty wears off.”
“The novelty has already worn off. All that remains is whether we’ve built something real underneath it.”
