AI: from pilot to production

The AI scaling problem is hiding in plain sight. Most enterprises have approved budgets and launched pilots, yet very few are seeing meaningful results. Using generative AI is great — but most users and companies are just scratching the surface when it comes to the potential that can be unlocked. 

 

McKinsey’s 2025 State of AI survey backs this up: 88% of organisations now use AI in at least one business function, but only 6% qualify as “AI high performers” generating meaningful impact. The gap between running a successful pilot and scaling AI across an organisation is the Death Valley where most initiatives fail.

 

This is not a technology problem. It is a commercial and organisational one.

 

 

A note on perspective

For the last 20 years I have worked across venture capital and private equity as a commercial and business leader. I invest in and assess dozens of companies, from early-stage AI-native to later-stage traditional technology and advise portfolio companies on accelerating revenue growth. This gives me a view from both sides: I see the AI vendors trying to win enterprise customers and I see the enterprises trying to make AI work at scale.

 

 

Why most AI pilots fail to scale

When I look at AI initiatives that stall, the same issues appear again and again.

 

The first is lack of business clarity. Too many initiatives start with the question “what can we do with AI?” rather than “what business problem are we trying to solve?” This sounds obvious but it is remarkably common. In my experience, more than half of generative AI initiatives never make it past proof of concept. Teams build impressive demos that demonstrate technical capability but have no clear connection to revenue, cost or customer outcomes. When budget decisions come around these initiatives struggle to justify continued investment.

 

The second is that workflows are left unchanged. Enterprises bolt AI onto existing processes rather than rethinking how work should be done. The real value comes from redesigning work around what AI makes possible. Most companies never get there.

 

The third is underinvestment in change management and skills. Companies budget for technology but not for the human side of transformation. For every dollar spent on technology, you may need to spend three dollars on helping people adapt. McKinsey has identified a similar ratio as critical to successful AI transformation. Yet most companies ignore it. They assume people will adopt new tools because the tools are better. This rarely happens. Resistance, confusion and old habits slow adoption.

 

Meanwhile, most employees have received no training on how to use AI effectively. The technology is ready, but the organisation is not. In fact, one of the most telling patterns I see is individuals quietly using AI tools on their own because getting corporate buy-in takes too long.

 

When your employees are solving problems with AI faster than your procurement process can approve it, that is not a technology gap. It is an organisational one.

 

 

What the best enterprise adopters do differently

The companies that successfully scale AI share a different set of behaviours.

 

They pursue fewer higher-impact use cases. Rather than spreading investment across dozens of experiments they focus on a small number of applications with clear business value where the potential impact is measurable. Depth over breadth.

 

They redesign work, not just automate it. Instead of asking “where can we add AI?” they ask “how should this function work in an AI-native world?” The productivity gains from redesign are multiples of what automation alone delivers.

 

Klarna illustrates both the opportunity and the nuance. Rather than bolting AI onto its customer service operation, the company redesigned the entire model. Its AI assistant now handles two-thirds of all conversations, doing the equivalent work of over 850 agents, and saved the company $60m. But in mid-2025 the CEO acknowledged they had “overpivoted” on AI at the expense of service quality and began rehiring human agents. Redesigning work around AI delivers transformative results, but the human element remains essential — at least until now.

 

They measure business outcomes not traditional adoption metrics. Tracking how many people are using AI is a good sign, but ultimately it boils down to measuring impact: has customer satisfaction improved, have processing times fallen, have error rates dropped. This distinction matters because adoption without impact is just cost.

 

They treat people and training as the primary investment not an afterthought. Change management is the real cost of AI transformation and the companies that get it right scale faster.

 

 

What AI tech companies are doing to win

The smartest AI vendors have recognised that technical capability alone does not close deals or drive adoption. Enterprise buyers have been burned by software that never scaled and end up sitting on the shelf. Winning in this market requires a different approach.

 

The best vendors are shifting from selling features to solving those painful points — the specific, persistent problems that keep customers up at night and drive a compelling event to act now rather than later. They lead with outcomes rather than capabilities. They can articulate exactly how their product reduces cost, increases revenue or mitigates risk. The ones gaining real traction go fu

 

rther. They root their solutions in industry-specific painful points that customers recognise immediately. When a vendor can speak to the particular challenges of a sector rather than offering a generic horizontal tool, the conversation shifts from “why AI?” to “how quickly can we deploy?” That specificity is what drives value — not just for the enterprise but for their customers too. That creates a powerful flywheel.

 

I see this first-hand through companies we have invested in. Two examples illustrate the pattern.

 

The first is an AI platform that is saving pharmaceutical companies up to 18 months in drug discovery by automating clinical trials. The company built the first truly AI-native platform in the sector, automating key trial processes and patient recruitment. Major pharmaceutical companies have onboarded as customers, with the platform delivering 21x ARR growth in a single year. It works because the founders understood a specific, high-value problem: clinical trials are historically manual and painfully slow. They created a purpose-built solution rather than a horizontal tool looking for a use case.

 

The second is an AI platform automating clinical administration for healthcare providers: generating notes, handling patient calls, managing emails and processing insurance claims. Clinicians using it save over two hours a day and clinics report a 30% reduction in administrative overhead. Clinical compliance, data security and integration with existing health record systems were built into the product from inception, not retrofitted. That is what gave healthcare organisations the confidence to embed it into their daily operations.

 

None of these companies set out to build a generic AI tool and then go looking for a problem to solve. They started with a specific industry, a specific pain and built purpose-built solutions around it. In most cases the founders have lived industry experience. They have felt the pain themselves. When you have personally sat with the problem, building the right solution becomes significantly easier. That is the pattern that wins.

 

Particularly in regulated, complex sectors like healthcare and financial services, the barriers to entry are high and the tolerance for failure is low. That complexity is precisely what creates opportunity. The vendors that do the hard work of building for these environments create defensible positions that horizontal AI tools simply cannot replicate. It is also where I focus much of my investment attention — because when a company solves a painful, industry-specific problem in a regulated market, the moat is real and the value compounds.

 

Beyond solving the right problems, the best vendors are getting the commercial model right too. They are building in governance and compliance from day one rather than bolting it on later. They are designing for workflow integration so the product fits how people already work. They are moving to usage and outcome-based pricing that aligns cost with value. And they are investing in customer success because if it works in the demo but not in the real world, you lose credibility.

 

The gap between AI experimentation and AI at scale is where competitive advantage will be won or lost over the next few years — and for AI businesses, we are seeing an exceptionally strong vintage. The companies that close it will not be the ones with the most advanced models or the biggest budgets. They will be the ones that treat AI scaling as a business transformation rather than a technology project.

 

The technology is ready. The question is whether organisations are ready to change how they work. If they fail to act, they do not just miss the financial upside — competition and market shifts will leave them standing behind.

 

---

 

Mitul Ruparelia is General Partner - Global Fund at Arãya Ventures

 

Main image courtesy of iStockPhoto.com and da-kuk