Today we’re back in the world of AI, for a startup helping teams design and build better digital products with the help of ‘stunt double’ user testers. Read on for full details!

But first, a few things you need to know about:

• Baltic Ventures has announced the second cohort of its Liverpool-based (but nationally focused) accelerator programme. There were some great teams in the first batch, so these are well worth a look

• Edinburgh Innovations, part of the University of Edinburgh, has announced the startups picked for the sixth run of its AI accelerator.

• A co-founder of Italian unicorn Scalapay, Raffaele Terrone, has announced his new London-based AI startup Desia today, alongside a $3.3 million pre-seed round led by Dig Ventures.

– Martin

Help us understand pre-seed in 2024

Hey, VCs! Angels! Founders! Startup support teams!

Take PreSeed Now’s two-minute survey to help us understand the early-stage startup landscape as it stands in late 2024.

Stunt Double uses AI agents for smarter, scalable product testing

When building a product, you’re obviously going to want it to be the best possible fit for your end users. Achieving that can involve a lot of tests, checks, and human effort.

Looking to provide an alternative approach is the appropriately named London-based Stunt Double, which looks to create what co-founder and CTO Paul Slattery describes as “a digital twin of your users, or prospective users.” 

The software runs through your product and performs various functions for you through the ‘eyes’ of those simulated users.” Using generative AI, the simulated users can impersonate specific personas that developers or designers want to test for.

In practice, the MVP version of the software can, for example, assess early-stage designs in Figma to see if they’re going to resonate with a specific target audience. 

It can also assess user journeys, to make sure something like a purchase flow works as expected. And it can assess the links within a website or app, to make sure they go to the places a user would expect them to. 

The benefit of this approach, Slattery says, is that it offers a scalable version of a type of human that is difficult to find:

“One of the things that's hard with qualitative testing is to find someone who has some knowledge of the domain that you’re in, but is also articulate enough to be able to give you feedback that actually is useful for you.

“What we've been able to do via prompt engineering is ensure that we can get extremely useful feedback that is based on a particular archetype for an actual experience.”

How it works

As Slattery explains it, Stunt Double combines computer vision with an LLM. This allows it to look at an app or website from the perspective of a specific type of target user. It then takes actions (such as clicking a link) as if it was that type of user.

Yes, we’re firmly in the realm of AI agents here.

Slattery’s fellow co-founder, and the startup's CEO, Michael Parker, gives the example of the ‘Audition’ plugin they’ve built to let designers ‘audition’ their designs in Figma with multiple ‘stunt double’ AI-generated users.

The software offers a range of preset ‘stunt doubles’ for users to work with when they begin a new project. These help test products against things like the needs of senior leaders, brand voice compliance, and data protection law compliance.

The designer can then click on an artboard and the ‘stunt doubles’ get to work. They’re asked questions like ‘what do you see?’, ‘what do you understand?’, ‘where would you go next?’ and ‘what questions do you have?’

“They’re very basic questions for now, but they're the sort of ones that you would do in a qualitative interview,” says Parker.

“The framework is a very safe methodology, and we use this as a way of guardrailing the outputs that come from the model, because we don't want it to go too wild. And we also don't know what the inputs are, necessarily.

“We're working on a way to then suggest insight on the original design and what you can do in terms of a refined design and iteration.”

The story so far

Parker and Slattery met when they worked together at Boston Consulting Group’s tech build and design unit, BCG Digital Ventures (since renamed BCG X).

Slattery comes from a software engineering background, building tech in the agency and consultancy worlds. Parker started his career as a designer and then spent time in product management. Together they identified the need for better early feedback and qualitative testing for product builders and designers.

Stunt Double is around 10 months old, and the pair have developed the product to the point where the first users have come on board the MVP version.

“We want to work with the first three design partners to really make sure that we're focusing on true problem-solution fit and product-market fit and not get distracted,” says Slattery.

“There are different archetypes; everything from small software companies that are struggling with scale, to bigger companies who need earlier feedback. And we just want to be careful we don't get trapped into enterprise requirements too early. 

“So we want to work with these design partners in the first few months, and scale up what we're doing from there. And then engage with more enterprise-driven customers as we're scaling. 

“We can see just even from our own networking, that there's a lot of interest in tooling like this.”

Subscribe to Premium to read the rest.