Exclusive Interview with Ron Dahan, founder of Upspring AI
Which LLM's leading the way? A new creative strategy landscape for e-commerce brands and how to best utilize these tools as a founder.
Flashback to April of 2021, when Apple’s new privacy changes known as ATT (App Tracking Transparency) sent e-commerce brands leveraging Facebook ads on a slow and painful path. By slow and painful, I mean Mel Gibson being hung, drawn and quartered at the end of Braveheart type stuff. So what happened when suddenly brands realised they couldn’t target users as effectively?
Advertising actually went back to what advertising is all about. The creative had to do the work, meaning you had to find a way to convince cold users driving along a highway of ads all day-long to click yours and buy your product. Being able to target users to such a refined level of accuracy temporarily died. This birthed a whole new era and emergence of software tailored towards a concept we call ‘creative strategy.’ This is the brain behind the assets that you’re building.
Back to the beginning - Tel Aviv
Ron’s journey started at the conventional ‘needing to learn how to code so I can build it myself’ crossroads that a lot of successful founders encounter. Attracted to high-tech, despite claiming he was ‘the worst student’ (which I doubt), Ron always knew he wanted to build something of his own.
Fortunately, he was doing this in what’s arguably one of the best start-up communities in the world, probably shy of Silicon Valley: Tel Aviv. Ron describes it as the kind of place where you can sit in a coffee shop next to a billionaire founder on one side and a young entrepreneur building their first company on the other. It’s small, concentrated and incredibly collaborative.
“Tel Aviv’s a small place,” Ron explained, before putting it more bluntly: “Israelis aren’t that good with privacy. I get calls or bump into people asking for introductions, saying, ‘Hey, can I have an introduction to the CEO of ___?’”
That led us into a conversation about attitudes towards risk, and whether start-up ecosystems simply become stronger when risk is easier to access. If it’s easier to move money, raise capital, meet founders and get introductions, you inevitably create an environment where more people are willing to build, regardless of whether every investment works out.
It reminded me of conversations with founders back in 2022 about Israel’s ability to deploy capital into innovative technology businesses, including those whose founders or technology originated in defence and the military before being repurposed commercially. While Ron didn’t describe Tel Aviv as having the same almost obsession with risk as San Francisco, the density of talent, capital and access creates a similar flywheel.
Against that backdrop, Ron built his career as an engineer across several tech verticals, from payments to founding his own booking system for events, which is still used domestically today. But it was his tenure at Minta, a Tel Aviv-based creative tech company, that ultimately gave him the foundational knowledge for what would become Upspring.
Not all LLMs are equal (this week, at least)
Ron and I first chatted when Igloo was beta testing Upspring several months ago, and he explained to me how he’d built a benchmarking solution that enables Upspring to fetch its insights from multiple LLMs at once and (to put it simply) pick which machine will give the best output for each fetch. I asked Ron which model Upspring leans on today. His honest answer: "If you asked me two weeks ago, I'd have told you something else. This week it's Gemini again." For video analysis at scale, Gemini still wins on cost and accuracy. For reasoning-heavy tasks and creative suggestions, Sonnet and Opus are more flexible. Fable is the sharpest for creative and directional thinking. And then there's the Chinese cohort (Sora alternatives, Kling, Wan), which he calls "insanely good" and basically free by comparison, in part because the Chinese government is happily subsidising the race (shock).
The more interesting shift he flagged is the rise of purpose-built models. Base 44 is his example. It's a vibe coding tool that spun up its own single-purpose model for generating websites. Not smarter than Sonnet 5. Just faster and cheaper at that one job, because it isn't wasting cycles reasoning about your existential dread on the side. His view: the big foundation models will still own the general ground, but expect a wave of vertical models where the company has enough proprietary data (Base 44 has 10 million generated sites, Harvey has legal, AllBirds has shoes lol).
Upspring's real bet on infrastructure and solving the agency dilemma
I’m going to be frank, and Ron appreciates my pragmatism when discussing what’s actually useful and what’s less tangible when it comes to SaaS offerings as an agency. Unfortunately, if we DON’T offer software solutions to clients, we’re ‘behind’, whether the clients use them or not. If we DON’T offer ‘creative reporting’ or insights for their designers (even if it would cut the middlemen out if the designers learned the strategy piece), we’re ‘not supportive enough.’ As an agency, our role has had to shift in the last 12-18 months from primarily media buyers to basically ‘paid marketing consultants’ who need to do lots to compete. This is because clients' expectations increase as AI makes ‘anything possible’ from their vendors, resulting in us doing things here and there that aren’t typically built into the scope.
So I talked about this shift with Ron and Upspring’s role in supporting agencies. The key we’ve realised is positioning Upspring as a tool that feeds the infrastructure. It’s practically a waste of time trying to convince brand designers to ‘go look at Upspring’ or ‘take a dive into the creative intelligence tool.’ We find ourselves hopping on calls going, ‘Did you look at Upspring for the next set of briefs?’ Response: ‘Umm, yeah, a couple of times, but do you mind providing recommendations for our next sprint?’ Our media buyers then inherit the role of strategists as well.
People want speed and outcomes handed to them, so the future isn’t about what the software can do. It is about embedding it within the system in-house to send instructions.
Whether it’s the structured breakdown of every video and static, what's actually working, what your competitors are running, or where you sit in the market, you bring your own AI (Claude, ChatGPT, whatever internal workflow you've already built), and Upspring feeds it the fuel and automates a workflow to push tasks based on that intelligence layer so it becomes actionable. Ron put it well: "I don't want to keep chasing the AI context you already have in your own tools. I want to fuel it."
MCPs are helping, but does automation improve account performance?
As a semi-pro vibe coder myself, I can confirm that access to MCPs and APIs is foundational to building an AI-enabled infrastructure, both for clients (if you’re building apps or portals for them) and internal teams. You can automate instructions around budgets, creative flags, reporting and much more, at the very least keeping team members and stakeholders accountable for changes. From an ops point of view, that’s a big step in the right direction.
The way I see the future of this structure is, to some degree, definitely automation. People want to do less of the ‘boring’ stuff and more of the ‘fun stuff.’ However, the greater question I find myself asking is: is all of this actually improving performance? I wrote an article recently called ‘CAC is rising, what to do about it?’, so I’ll be the first to say that clawing back a basis point across the P&L can have a valuable compounding effect. If that means enabling or pausing rogue assets 10% faster because AI has made that possible, then great.
However, is the broader narrative still the same? Auctions are still getting more expensive (broadly, CPMs are rising), and the best ‘creative concepts’ and long-standing businesses, bootstrapped ones at least, are those with great ideas, great products or the best CAC:LTV. So much of a great ‘out-of-the-box’ idea is still human, because humans can reason live, harnessing experiences and context that AI can’t see (at least until AI robots start living and breathing like we do).
It’s like a trader having access to a formula that allows them to predict markets, until they realise every other trader also has tools that help predict markets. Suddenly, your competitive advantage vanishes and you have an ‘efficient market’.
So, is robust AI enablement software for ads specifically beneficial? Yes. Does a human reading Upspring insights thoroughly and spending an hour or two thinking of exciting concepts outperform an LLM pushing ‘instructions’ based on its intelligence to designers? I think so.
I think there’s a faster and bigger lift today from AI helping users conquer landing page improvements, deploy code to Shopify and improve the mid-to-lower-funnel experience than there is at the top of the funnel. AI can build incredible things, there’s no doubt. The intelligence stack will help, but AI’s ideation skills still need work.