Daniel Brady

The through-line

Daniel Brady's recurring argument is that machine learning has to be the product, not decoration on top of one. He and co-founder Zach had both watched data science work end up as prototypes shown in board meetings, the kind of thing companies "would raise $50 million based on" while the models never reached production, "because businesses change, because it's complex, because actually serving those models is really, really difficult" [1]. Orita was founded as the corrective: "we wanted to start a company where machine learning was the product that was delivered. It wasn't a nice to have on top of some other primary value" [1]. He is candid that this sounds obvious now and did not then: "back when we started it was like no, you know, you have a same day delivery company, you can use machine learning to make it better, but you can still just have a delivery company" [1].

The second thread is speed, and it runs from his exit from research through to how he reads a market. He left academia "mainly because it was like too too slow," summarising his PhD as a joke on himself: "in my second year of my PhD, I figured out what it was that I was going to work on, and it took me four years to keep proving that that was true" [1]. The same impatience shows up in how he evaluates opportunities, which he judges almost entirely by how fast people move toward him.

On why he left the lab, and what science gave him

The ambition was fixed early and never had a fallback: "I didn't even think about an alternative to it. I mean, I I wanted to be a neuroscientist since I was in middle school" [1]. The trigger was reading The Terminal Man, in which surgeons stimulate motor and sensory cortex in awake patients, and the seventh-grade conclusion that "this is who we are as people is is like this thing right now" [1]. He ruled out the clinical path because the science was more interesting to him than the medicine, given that "we don't know how brains code information" [1]. His academic work centred on how early life experience shapes the way the brain processes information, across neurobiology and postdoctoral research on sensory systems in the developing brain [6][7], and he has since spoken publicly about critical periods, neuroplasticity, synesthesia and childhood development alongside customer segmentation as parts of the same intellectual story [7], and about carrying brain science into e-commerce practice [5].

The transition into tech was accidental and instructive. A friend becoming CTO of a same day delivery startup asked for help on routing; Brady applied "a little bit of math and probability theory" and did not think it was special: "This is how we would analyze this data, you know, as a scientist." He was made their first employee immediately [1]. The lesson he drew is that scientific method is scarcer in startups than scientists assume: "that kind of skills of like science stuff like that were were not super common in that sort of space. And so there was just like this real need and real hunger" [1]. Before studying artificial neural networks, he had studied biological ones [6].

On choosing a market, and why healthcare was abandoned

Commerce was not preordained. Orita's early work was "extremely technical," including applying for NSF grants for research considered too speculative for conventional funding, and getting well down the path of a grant with a hospital to apply machine learning to helping nurses do their work [1]. That died on access to data, not on the idea: "the red tape in order to get access to that information was so long we were like we can't exist like that... two guys can't wait 9 months for the legal team to clear that sort of thing" [1]. Commerce won because the pull was already there, through knowledge of logistics and brand problems and a network of friends and investors at commerce tech companies who kept routing work to them: "we just kept getting work in that in that space very quickly" [1]. The implicit rule is that a two-person company should go where the data and the meetings are immediately available.

On collapsing a broad capability into a one-sentence pitch

The consulting era gave Orita too much surface area. Working as something like a fractional CTO or "chief algorithmic officer," they built CDPs, reverse CDPs, business intelligence reporting, direct mail routing, at a time when only a company like Nike would have such things in-house [1]. Brady treats that breadth as the problem: "we had helped solve so many issues for these brands. It was kind of hard to understand and to focus on which one to do" [1]. The unlock was advice from adviser Matt Finen of fairing, who pointed at a small section of their customer behaviour report on marketing waste and told them to "pretend that that is all you do" [1]. Nothing under the hood changed: "even though the algorithm the analytics none of it changed under the hood we just rapidly collapsed the messaging just to be like hey this is what we do it's a simple one sentence answer" [1]. The resulting pitch was to use ML to make targeting much smarter, expressed to brands as sending 400,000 emails instead of a million for the same revenue [1].

The seed of that pitch came from an adjacent channel. They had already done it in direct mail, where postage made waste expensive and the list was full of duplicates and people who had moved, so a brand could get the same revenue from 100,000 or 75,000 postcards, "or maybe even a little bit more because there's new pockets that you weren't targeting originally" [1]. Email was the same hypothesis in a faster channel with better APIs and a common tool, Claio [1]. The productised version handles junk and spam accounts, duplicate IDs, suppression of unengaged contacts and reactivation of people when they are ready to buy, accounting for how customers interact with a specific brand rather than in general, and is positioned as saving brands more than 25% on email marketing [4][6][5].

On what product-market fit pull feels like

Brady's test for a real market is not survey data, it is response latency. A friend running social media forwarded the pitch to six brands; five booked meetings within thirty minutes and the sixth only declined because they did not use Claio. All five became pilot customers with instant access to their email platform [1]. That was enough to rewrite the company: "literally that day we changed our website to say that that was what we do. Even though we hadn't actually done anything in Clavio yet" [1]. He frames this as the one thing he was confident he would recognise: "I didn't know a lot of things but I think that I would know intuitively what a product market fit pull would feel like and now you could mess it up in a million ways but like when you feel it it's like very obvious" [1]. The tell is specific, that busy people are "immediately responding to us immediately hopping on calls immediately giving us access to everything" [1].

The willingness of customers to bear risk was the other signal. As Aaron Schwartz described it from the outside, Careway let Orita switch hundreds of thousands of contacts on or off in a week, and the point was not the fee but that a wrong call would have cost Careway a great deal of revenue [1]. Pricing followed the proof rather than preceding it: "We'll save you this money. We'll take a quarter of it monthly." And they're like, "Done." [1]. Distribution arrived the same way, with Jason from D Lashes having already told everyone he knew, before Orita had any go-to-market motion or vocabulary for one [1].

On saving money turning out to be finding money

The AB test that validated the pitch overshot the expectation. They had hoped for a 10 or 15 percent improvement and got 35 percent, and with it an unplanned second effect: brands sent less, saved money, and got more engagement and more clicks, which is deliverability. "We didn't even know that we were stumbling into that. Like that was just like a secondary effect to us" [1]. That accident reframed the company, from a cost savings tool into a growth engine, a shift Brady and Schwartz note customers are still catching up with two years later [1]. The general principle he draws is that a product's benefits are discovered rather than designed: "you can play with the data and you can go build a product and then you keep unpeeling like what the benefits are" [1].

On when machine learning is actually worth using

Brady separates convenience tools from ML that changes outcomes. For removing menial or moderately complicated tasks, a chat tool is fine at any size and worth using if it helps [1]. The kind of work Orita does starts to matter for a different reason, and he is precise that scale is a proxy rather than the cause: it "is correlated with becoming larger... correlated when you have more and more data through your system but what it what is actually the driving factor is that you have a lot of variability in your customer base" [1]. His model of a brand's maturity is a sequence of ever finer cuts: a small brand does the same thing to every customer, then two things, then keeps slicing to gain performance until manual segmentation runs out of road [1]. This connects to the founding technical premise of Orita, that existing tools and language models cannot give data-backed answers about customer targeting without hallucinating from blog posts rather than the brand's own data [1].

On retention, brand voice and what marketers overlook

Brady has explored retention beyond the model layer with practitioners, covering what great retention looks like from the first email to the tenth purchase, the case against the email-every-day approach, using AI to write faster without losing brand voice, and the argument that the customer service inbox is the most important retention signal most marketers ignore, alongside how to balance brand marketing against performance and when to stop testing and simply call the customer [8].

Takeaways

• Build companies where machine learning is the delivered product, because ML layered on another primary value tends to stall as prototypes and board slides that never reach production [1].
• Judge a market by response speed: five of six cold-forwarded brands booking meetings within thirty minutes was enough to change the website the same day, before any results existed [1].
• Narrow the message even when the technology is broad; Orita kept the same algorithms and collapsed the pitch to one sentence about marketing waste on Matt Finen's advice [1].
• Machine learning pays off when a customer base has high variability, not simply when a company gets large; size and data volume are correlates, variability is the driver [1].
• Cost savings and growth can be the same product: cutting email volume raised engagement and clicks through deliverability, an effect Orita did not anticipate and later built the business around [1].
• Price against proven savings, taking a quarter of the money saved monthly, which removed the need for a conventional sales process early on [1].
• Avoid markets where data access is gated by legal review; a nine-month clearance cycle is fatal to a two-person company regardless of the quality of the idea [1].
• The retention signal most marketers ignore is the customer service inbox [8].