How Art Fund is using AI to help people discover art and culture they'll love

Art Fund is the national fundraising charity for art, supporting museums and galleries across the UK through grants, professional development and fundraising. They are also responsible for the National Art Pass, which gives members discounted and free entry to hundreds of museums, galleries and exhibitions across the country.

A key role of the Art Fund website is helping members discover places to visit and make the most of their membership. Through its Explore section, users can browse more than 1,000 museums, 350 exhibitions and 300 events at any given time.

Art Fund already does a lot to help people discover relevant content, from curated guides and themed recommendations to editorial features and regional round-ups. But with such a large and constantly changing catalogue of museums, exhibitions and events, helping the right people find the right experiences remains an ongoing challenge.

Recently, we worked with Art Fund to explore whether AI-powered recommendations could help National Art Pass holders and non-Pass holders alike discover museums, exhibitions and events in a more personalised way. What started as a workshop to explore opportunities quickly evolved into a working prototype that is now in public beta.

Before exploring potential solutions, we ran an AI discovery workshop with six people from across the Art Fund team. Participants brought a mix of digital experience, confidence levels and perspectives on AI, helping ensure the conversation reflected a broad range of views.

Rather than starting with the question, "Where can we use AI?", we focused on a more practical challenge: what problems are we trying to solve?

Together, we explored the needs of two priority audiences: under-30s and families. We looked at what they were trying to achieve, what was getting in their way, and where small improvements could make a meaningful difference.

By grounding the conversation in audience needs, we were able to identify opportunities that felt both useful and achievable. As ideas were explored further, they were also tested against the reality of Art Fund's existing systems and data.

Some concepts sounded promising, but relied on data and information that didn't exist. Others quickly proved more complicated than the problem justified.

One idea consistently rose to the top: a conversational recommendation tool that could help people discover museums, exhibitions and events in a more natural, personalised way.

Once the idea had been prioritised, the focus shifted to learning as quickly as possible. Rather than spending months producing specifications and business cases, we built a quick first prototype and shared it with the Art Fund team. The goal wasn't to create a finished product, but a working proof of concept - something tangible that people could react to and help shape.

Over the following weeks, the project followed a simple cycle:

Prototype → Show and tell → Gather feedback → Improve → Repeat

Three rounds of iteration helped refine both the experience and the underlying approach while keeping momentum high. Using existing website content as the foundation for recommendations reduced technical complexity and allowed the team to move from workshop discussions to a working prototype within weeks.

One of the themes that emerged throughout the project was the importance of involving colleagues early. By bringing a wider group into the workshop process, including people who might not normally be involved in digital innovation work, the resulting idea felt more shared and collaborative.

This was particularly important given the subject matter. AI continues to generate both excitement and uncertainty, and creating space for people to contribute ideas, ask questions and challenge assumptions helped build confidence in the direction of travel before any technology was put in front of users.

It also brought fresh perspectives into the conversation, helping ensure the solution reflected a wider range of audience needs.

Although it is too early to have consolidated feedback from real users yet, we already have learnings from the project for our combined Torchbox/Art Fund team.

One lesson has been around expectations. When people use tools like ChatGPT or Claude every day, it's easy to assume that a specialist recommendation tool should behave in exactly the same way. In reality, recommendation tools operate within much narrower datasets and require different rules, behaviours and guardrails.

The project also highlighted how quickly you can move an interesting prototype to an operational product. As the tool progressed towards launch, new considerations emerged around hosting, privacy policies, data handling and ongoing support.

None of these challenges were insurmountable, but they reinforced the importance of involving legal, data and governance stakeholders earlier than you might expect.

Following extensive internal testing and a small project to get the tool ready for public use, the tool has now been soft launched by Art Fund as a pilot, and is being promoted to key segments of their audience to start with. You can view the tool at recommendations.artfund.org/

This pilot phase will gather feedback from both existing National Art Pass holders and potential new members to understand how people use the tool, where it adds value and what improvements should come next.

As with the rest of the project, the focus remains on learning, iterating and understanding how AI can support people to discover more of the art and culture available to them.

Through our Innovation Fund, we're helping charities explore ideas like these by co-funding AI proofs of concept with the potential to create real impact.

If you're curious about how AI could help your organisation reach more people, improve services or work more effectively, we'd love to hear from you.