You’ve Built the Model. Now What? The Hardest Part of Productising AI
12 Nov 2025
Any founder knows that building an AI model is only the beginning. The real test is turning technical innovation into a usable product that scales.
Across the Faces of AI series, founders shared the toughest parts of that journey - from market readiness and regulation to maintaining alignment as their teams and products grow. Their stories show that success depends as much on people and process as it does on the model itself.
1. From Proof of Concept to Product
Katya Lait, nettle
Scalability isn't top of mind yet. We’re still in our first year, building bespoke versions of nettle for clients. Year Two is when we’ll step back, consider the ecosystem of solutions we’ve built, and focus on scalability.
Some obvious steps we’re taking now are distributing requests across specialised agents, reducing unnecessary API calls, and building generic components we’ll reuse across clients. As much as we can, we cache everything and have the system learn from itself - a crucial feature for our mission to close the knowledge gap in risk engineering.
Beyond data sovereignty and open-source innovation, the biggest challenge has been translation. We’re working in a country where English isn’t spoken, so our AI reasons in English but delivers results in the user’s language. It’s complex, but it means we can support virtually any market moving forward.
George Parry & Charles Cross, Emma AI
George: Our hardest technical challenge was building a system where any user can ask any question, about any part of their data, at any time, and get the right output. Interoperability in health and social care was another major hurdle. Too many companies treat patient data as their moat - even though it belongs to the user. We’ve worked to make Emma the opposite: an open platform anyone can connect to and from.
Charles: Another challenge has been bridging the gap between initial excitement around AI and real-world impact. In a regulated, human-centred sector like care, plug-and-play tools don’t work. Our partnership model means working closely with providers, iterating fast, and being open to tough conversations.
Eric Marcuson, ClinBI
For most founders, one of the biggest challenges is figuring out what customers actually want. Product–market fit isn’t something you check off and move on from. It evolves - as channels, customer value, and priorities shift.
The other big challenge is people. Whether it’s your team or your customers, understanding how people think, communicate, and collaborate is critical. That means always talking to the customer and those around them.
Patrick Sharpe, Artificial Societies
What we’re building is unique - there aren’t many prosumer tools that let people run real experiments. One of the biggest challenges has been making that experience accessible without dumbing it down.
Early on, there’s a temptation to overbuild. But we’ve had to practise real restraint. Our mantra is that simplicity is the ultimate sophistication.
2. Bridging the Gap Between Tech and Adoption
Eduardo Mandela, Maihem
The biggest challenge (and opportunity) is the gap between cutting-edge AI and real-world adoption. Most companies don’t have the talent, understanding, or confidence to use these tools effectively.
I call this the ‘chasm of adoption’. That’s where we come in - bridging the gap through education, helping people understand how AI systems work and how to use them safely and effectively.
George Hancock & Eric Topham, Octaipipe
George: We’ve always been slightly ahead - building solutions 18 months before the market’s really ready for them. The challenge isn’t whether AI can help; it’s whether the market recognises the problem it’s solving right now.
Eric: 99% of people talking about AI don’t really know what they’re talking about. That makes scaling difficult because most organisations don’t know what they’re buying or how to be ready for it. You can build a brilliant churn prediction model, but if the business doesn’t know what to do with that prediction, it’s useless.
Roshan Tamil Sellvan, AdvisoryAI
It’s important to set clear ethical boundaries, especially in finance. We’ve built safeguards into our model architecture - training the AI to filter inappropriate language and analyse outputs against FCA frameworks before deployment. Multiple specialised models validate accuracy and compliance at each step. If a report doesn’t meet quality thresholds, the system prompts a rewrite.
We started out building an assistant, but realised we’d created a powerful report-writing tool. That adaptability has opened new markets - from finance to law. We’re now expanding into the US, Australia, and Canada, each with its own regulations and nuances, but that’s part of the challenge we’re tackling as we grow.
3. Scaling Without Losing What Works
Jonathan Low, Scope
We've taken a modular approach to development because most inspection workflows are unique. As we scale, the challenge is keeping the data layer consistent so we can onboard customers faster and make it easier for them to adjust their reports and workflows.
As a small team, we've stayed very customer-focused. The goal is to keep that mindset - ensuring everyone on the team maintains the same level of customer involvement as we grow.
Scott Wilson, Covecta
Our biggest challenge - and it’s ongoing - is scaling while maintaining impact and quality. As we work with more customers, the volume and complexity of product requirements increase.
We’re focused on prioritising the features that deliver the greatest value to users and building an engineering engine that consistently ships high-quality AI capabilities at scale.
Once the model works, the real work begins.
Productising AI means navigating regulation, readiness, and human behaviour - all while proving value at scale. These founders show that lasting impact comes not just from technical excellence, but also from from clarity, adaptability, and a focus on real-world use.
Generative connects top AI talent with the startups defining the future.