OpenAI is breaking records again in 2025. But behind the buzz, nearly half of all enterprise AI projects are being abandoned, according to a recent report.
Getting from proof of concept to broad adoption is the hard part. However, companies in SaaS, fintech, communications and logistics are pushing through in creating AI startup successes — with industry-specific tools built for a purpose.
In cybersecurity last December, Flare raised $30 million to fight credential theft. The SaaS startup tracks stolen logins on illicit dark web channels in real time and resets passwords before hackers can cash in. The fintech industry saw embedded lending platform, Jifitti launch Tap Now, Pay Later™, a tool that drops approved funds directly into digital wallets, eliminating friction.
Meanwhile, over in logistics, Transmetrics is helping smaller freight operators — the 96% that operate with 10 trucks or less and lack enterprise tech — plan routes and optimize fleets using AI. Founder Asparuh Koev is now taking it further, using data models to tackle the carbon cost of road freight.
The most effective AI startups aren’t reinventing workflows; they’re disappearing into them. They feed off proprietary data and deliver value by solving friction in targeted spaces.
When it comes to leveraging AI, here are four reminders for startup founders to keep in mind in 2025.
Invisible UX is king
Burnout is rising fast—69% of workers report feeling it. Leaders can barely plan ahead when each day brings a fresh set of curveballs. Tipping points for technology maturity also keep shrinking: mobile phones took 18 years to go mainstream, the web took 7, WhatsApp 3.5, TikTok 9 months—then ChatGPT did it in just 2. As soon as the public adjusts to one system, a new disruptor takes its place. For broad AI adoption, tools must fit seamlessly into users’ existing day-to-day.
SaaS leads with UX-first AI. Nate MacLeitch, CEO of communications SaaS provider, QuickBlox, expresses, “The best AI isn’t something users notice — it’s something they feel. A smoother handoff, one less screen to toggle, a patient question answered before it hits the inbox. In healthcare, where attention is stretched thin, adoption happens when the tech gets out of the way and lets people focus on care.”
Fintech leaders also grapple with highly sensitive, highly regulated data, and frictionless transactions can be the difference in whether a user completes a purchase, builds trust in the brand, or drops out entirely.
“In lending, the moment of truth is seconds long. If a customer hits friction, you lose them,” states Yaacov Martin, CEO of Jifiti. “The challenge isn’t just approving credit, it’s embedding that decision invisibly into the buying journey. When AI works behind the scenes to assess, approve, and disburse — without ever interrupting the experience — that’s when adoption skyrockets.”
Real-time feedback loops matter
In the first half of 2025 alone, over 29 million individuals had their data compromised in healthcare breaches. This is a stark reminder that when AI systems can’t get timely feedback in critical settings like cybersecurity, healthcare, or fintech, they expose brands to financial and reputational harm.
“In cybersecurity, real-time has to be the baseline,” says Mathieu Lavoie, CEO of Flare. “By the time an alert is sent, an attacker might already be moving laterally. If we detect leaked credentials but wait hours to act, the damage is done. The value comes from what happens next — confirming the threat, taking action, and learning from it. That feedback loop is where AI earns its keep.”
It’s a similar story in healthcare, where delayed communication can lead to missed care, duplicate work, or patient frustration.
“AI can generate insights, but if they’re trapped in a silo or delivered too late, they lose impact,” adds MacLeitch. “AI only works when it stays in sync with the humans using it. If a system flags a patient risk but no one sees it in time, you’ve just automated a delay. Real-time feedback isn’t just about speed, it’s about relevance. Teams need to see, respond, and adapt while the moment still matters.”
Concerns around security also mean that AI adoption is limited in corporate and business use cases. AI productivity tools like ChatGPT are often banned due to concerns around IP and sensitive corporate data. Rajat Mishra, CEO of Prezent, believes we need a move towards specialized models for these business users.
“Contextually intelligent models can also be trained on proprietary datasets without compromising security. Given the wave of copyright lawsuits against OpenAI, Microsoft, and others, this is crucial for companies that handle sensitive data,” he explained.
Proprietary data builds moats
In 2025, only 12% of organizations say their data is of sufficient quality and accessibility to support effective AI use. Meanwhile, 64% cite data quality as their biggest barrier to using data-driven systems at scale, and 67% admit they don’t fully trust the data powering their decisions.
“Off-the-shelf AI can only go so far. What separates the AI startups that create impact from generic output is the quality and uniqueness of the data feeding the system,” comments Martin. “Across industries, companies with access to deep, domain-specific data, along with the ability to structure it, are able to train models that actually reflect the messy, high-stakes decisions their companies make.”
That distinction becomes especially clear in sectors like logistics, where operational knowledge runs deep but is rarely captured in structured form.
“In logistics, we don’t just need information, we need the right information, cleaned and shaped by context,” says Asparuh Koev, CEO of Transmetrics. “A lot of small fleet operators know their business inside out. They can load a truck faster than any algorithm; they know every turn on the route, every client’s quirks. That knowledge stays in people’s heads, but it’s actually in the data too — buried in dispatch histories, fuel logs, loading times. When you surface that through structured training reports or predictive planning, you’re able to create best practices that everyone can benefit from.”
The latest AI adoptions aim even wider, which comes perfectly in conjunction with technologies making this technology more accessible to people. For JD Raimondi, the head of Data Science at Making Sense, a Silicon Valley software development company, says, “This will hopefully lower the bar for technology adoption and usage. One of the great successes of AI is that almost no skills are required to use it. While this carries risks, for the general population, it is likely to open doors to a new flood of information and possibilities, helping address social concerns.”
AI economics must align with value
For AI to stick, it has to earn its place in daily operations and in the minds of its users, a battle that is getting tougher to win. Nearly 47% of organizations globally are currently piloting AI agents or exploring new use cases, yet only 2% have fully scaled deployments. Contributing to the gap is trust in autonomous AI, which has tumbled from 43% to 27% globally, underscoring the need to prove value early and earn user buy-in.
“Leaders need to solve problems teams already care about, and fit training into daily routines without disrupting them. In sectors like logistics, where margins are thin and every liter of fuel counts, there’s little patience for experiments,” adds Koev.
“People don’t adopt AI because it’s exciting, they adopt it because it saves time, money, or hassle in a way they can feel right away,” Koev continues. “You don’t start with a grand vision of transformation. You start with a quick win. That could be optimizing capacity, predicting delays, reducing waste, depending on your business’s existing data quality, and you build trust from there. If teams don’t see value early, the tech won’t make it past month one.”
The approach to this technology should be straightforward: AI should never replace; it should be adopted to enhance operations, and must continuously function under human supervision and judgment, Christian Struve, CEO and Cofounder at Fracttal, the Europe-based AI integrated maintenance platform, says “I believe that artificial intelligence doesn’t redefine what intelligence is, but it does force us to rethink our relationship with it. The important thing isn’t that a model evolves, but that it does so in alignment with the goals we define as humans.”
What we are seeing is that the real edge isn’t in the algorithm, it’s in how quietly and effectively it fits into the work that already matters. Startups that win won’t be the loudest; AI startup successes will be the ones that solve real problems with the right data and in the right places.
Article co-authored by Emily Singleton