Devstars
By Stuart Watkins, Founder of Devstars
There’s no shortage of articles telling you to “start an AI business today.” Build a chatbot. Launch a content agency. Create a personal shopping assistant. The internet is full of these listicles, complete with optimistic revenue projections and reassuringly low startup costs.
The reality is a bit different.
Around 90% of AI startups fail within their first year. An MIT study found that 95% of enterprise AI projects deliver zero measurable return. And of the 14,000+ AI startups launched globally in 2024, roughly 40% had already shut down by early 2026.
So before you remortgage the house to build an “AI-powered” anything, let’s talk about what’s actually going wrong — and what the survivors are doing differently.
The single biggest reason AI businesses fail is also the simplest: they’re not really businesses. They’re wrappers.
A wrapper takes someone else’s AI model — usually OpenAI’s — puts a nice interface on it, and charges a monthly fee. The problem is that your entire product is one API update away from being obsolete. OpenAI adds your feature to ChatGPT Plus, and overnight your paying customers have no reason to stay.
We’ve already seen this play out. Jasper AI, once valued at £125 million, was acquired for parts. Copy.ai merged with a competitor. Character.AI’s talent got hoovered up by Google. These weren’t unfunded side projects. They had real teams, real traction, and real venture capital behind them.
The pattern is always the same: startup builds a vertical AI tool, gets traction, then the platform underneath releases the same capability for free. Game over.
Here’s a quick test. Take your business idea and remove the words “AI-powered” from the description. Does it still make sense? Does it still solve a problem someone would pay for?
If the answer is no, you don’t have a business. You have a feature.
The businesses that survive aren’t built on AI as the product. They’re built on AI as the method. There’s a massive difference.
We use AI across everything we do at Devstars — from content production to competitive intelligence to data analysis. But we’re not selling “AI.” We’re selling outcomes: more visibility, better conversions, stronger digital infrastructure. AI is how we deliver those outcomes faster and at a quality level that wasn’t possible five years ago.
Those “12 AI Business Ideas You Can Start Today” articles love to quote startup costs of £1,000 to £5,000 and monthly revenues of £10,000+. Let me be straight with you: those numbers are fantasy for most people.
What they don’t mention:
At least 30% of AI projects get abandoned after proof of concept because the underlying data is rubbish. You can have the cleverest model in the world, but if you’re feeding it messy, incomplete, or irrelevant data, you’ll get messy, incomplete, irrelevant output. We’ve been doing this for 25 years, and data quality is still the thing that makes or breaks projects.
No-code tools have made it easier to build things, but “easier to build” doesn’t mean “easier to build well.” The difference between a demo that impresses your mates and a product that retains paying customers is enormous. It’s the difference between knowing how to use a power drill and knowing how to build a house.
The AI market is moving so fast that what looks like a gap today might be filled by Google, Microsoft, or OpenAI tomorrow. You need to be solving problems that big platforms can’t easily replicate — either because they’re too niche, too regulated, or require deep domain expertise.
After 25 years of building digital solutions and watching three major technology shifts play out, I’ve noticed that the businesses that survive share some common traits.
The AI businesses making real money aren’t the ones with the flashiest demos. They’re the ones automating tedious, expensive processes in industries that move slowly. Think legal document review, inventory forecasting for mid-size retailers, or compliance monitoring in financial services.
Boring is beautiful when it comes to sustainable AI businesses. The more specific and unglamorous the problem, the less likely a big platform is to bother solving it.
If your business generates proprietary data through its operations, or if you’ve built deep expertise in a specific domain, you have something that can’t be replicated by a teenager with a ChatGPT subscription. That’s your moat.
We’ve spent nearly a decade building and optimising property portals for Investors in Property. That depth of understanding — the Salesforce integration, the multilingual expansion, the CRO research — none of that comes from a prompt. It comes from years of working inside a specific domain.
The MIT research found something interesting: companies that purchased AI tools from specialist vendors succeeded about 67% of the time, whilst those that tried to build everything internally failed twice as often. The sweet spot is AI augmenting human expertise, not replacing it.
At Devstars, we call this the Expert-AI Partnership model. AI handles the volume and production. Humans provide the judgement, domain expertise, and quality control. When we helped South Coast Powersports achieve 100+ high-intent keyword rankings in six months, it wasn’t because we pointed an AI tool at the problem and walked away. It was structured data mastery, technical GEO expertise, and strategic thinking — amplified by AI.
Nobody wakes up in the morning wanting to buy AI. They want to sell more products, reduce costs, reach new markets, or stop losing customers to competitors.
The businesses that thrive frame everything in terms of outcomes. Not “we use GPT-4 with retrieval-augmented generation” but “we help you show up when your customers are researching their next purchase.”
Here’s where it gets interesting for anyone thinking about where to invest their time and energy.
The biggest AI-driven business opportunity right now isn’t building AI products. It’s adapting to what AI is doing to how people find and choose businesses.
Zero-click searches now account for roughly 65% of all Google queries. When AI Overviews appear on a search result, click-through rates drop by around 60%. Nearly 7 out of 10 Google searches end without anyone clicking on anything.
This isn’t a blip. It’s a structural shift.
What’s happening is that AI tools — Google AI Overviews, ChatGPT, Perplexity, Gemini — are becoming the research layer. People ask AI for recommendations, comparisons, and advice before they ever visit a website. By the time they do click through to your site, they’ve already made most of their decision.
We call this the Bifurcated Search Model. There’s the research phase, which increasingly happens in AI tools, and the purchase phase, which still happens on websites. If you’re only optimising for the second phase, you’re invisible during the first.
This is what Generative Engine Optimisation (GEO) is about. It’s not a product you can wrap in a nice UI and sell as a subscription. It’s a strategic discipline that requires deep understanding of how AI tools select, cite, and recommend content.
Most agencies haven’t caught up with this shift yet. They’re still selling keyword rankings and traditional SEO whilst the ground moves beneath them. We’re seeing it with our own clients: impressions rising, clicks falling. We call it the Jaw Effect — and it’s the clearest signal that the old playbook isn’t working.
There’s a genuine first-mover advantage here, but the window is closing. Within 12 to 18 months, the bigger agencies will wake up to GEO and start offering it. Right now, businesses that move early have a real competitive edge.
When we rebuilt Headmasters’ digital presence across 56 salon locations, we didn’t just chase vanity SEO metrics. We focused on high-intent local search — the “hair salon near me” queries that actually drive bookings. Eight years later, they’re still dominating those searches whilst competitors are scrambling to work out why their traffic is disappearing.
That’s the kind of thinking that survives technology shifts. Not building another ChatGPT wrapper, but understanding how people actually find and choose businesses — and making sure your clients show up in the right places at the right time. That’s what a good digital marketing agency in London should be doing right now.
If you’re thinking about starting an AI business, ask yourself five questions:
The AI revolution is real. The opportunity is genuine. But the opportunity isn’t in building the 10,000th chatbot or content generation tool. It’s in applying AI strategically to solve specific, valuable problems — and in helping businesses adapt to the fundamental changes AI is making to how customers find them.
That’s not as catchy as “12 AI Business Ideas You Can Start Today.” But it’s a lot more likely to still be relevant in 12 months’ time.
Stuart Watkins is the founder of Devstars, a digital growth consultancy specialising in Generative Engine Optimisation (GEO), bespoke software development, and fractional CMO services.
With 25+ years of digital experience and long-standing partnerships with clients including Headmasters, Investors in Property, Heathrow Airport, and the Ministry of Defence, Stuart helps ambitious businesses navigate the shift from traditional search to AI-driven discovery.
Based in Jersey, Channel Islands, working with clients across the UK and internationally.
Tell me what you’re trying to fix. Half an hour, no pitch, no slide deck.
If we’re the right fit we’ll talk about what’s next. If we’re not, I’ll point you to someone who is.
