Sandlab

testing all the time

Testing all the time

Our not-so-secret sandbox to ask questions, try things out, and sometimes get it wrong.

Testing approaches or tools, creating solutions, bending methods to better suit them… With the rise of AI it's become even easier to quickly try things that puzzle us, to enjoy the luxury of getting it wrong, and to refine it all into something that works.

And when something works, why not share it with everyone!

Why Sandlab? We wanted a name that fits our identity: a scientific foundation that doesn't get in the way of creativity. Our sandbox. But also our lab.

Tools we've built ourselves

Our studies

Can an AI embody a persona?

When people talk about UX research and AI, the reflex of plugging personas into data to bring them to life comes up (very) quickly. So we plugged seven (real) personas into their customers' notes and had an LLM make them talk. We tested various setups, of course. The conclusion, maybe an obvious one, is that there's no magic way to embody both the customers' average and their standard deviation. Put another way: either a persona renders diversity poorly, or it does, but then it's not really a persona. Either way, the quality of what comes out is not great.

Forgetting without warning: can we rely on AI to report research insights?

For an AI to answer from user interviews, is it better to give it excerpts found by a search, or the interviews in full? An article that talks RAG and context window, and that realises the main reliability problem isn't hallucinations anymore, but undocumented omissions.

To make an image lighter, should you lower the resolution or push the compression?

On 10 RAW photos, 4 formats and 6 resolutions, we measured perceived quality. JPEG XL gives the best trade-off. At acceptable or high quality, lowering the resolution cuts file size as much as changing format.

Synthetic panels or silicon users: what if we no longer needed to interview real customers?

The AI tsunami is coming, and the whole digital industry is changing radically. AI replacing designers, researchers and project managers. And of course, the question of replacing the users themselves in studies came up very quickly. AI used to simulate users taking tests. But does it work if we try to replace real people with AI, when we put them in front of a real questionnaire? Spoiler: “meh, not really”.

Dark mode: why is consensus so hard to reach?

Negative polarity, eye strain, performance, perception, eco-design: everything recent scientific literature teaches us (and where it contradicts us) about Dark mode, through the lens of the S-O-R model.

Measuring workload in UX research: lessons from the NASA-TLX

Lessons learned with the NASA-TLX, a simple, fast and proven questionnaire for characterising the workload associated with a product, an interface or a service.

Benefit from the French Innovation Tax Credit (CII)

Thanks to our official approval you can claim a 20% tax credit by investing in innovation, on the part of the project covered by our services: improving usability.

Learn more about the process

Asking questions for our clients too

While this space mainly serves to test things and advance our approaches, above all it lets us put those results to work to improve the way we work with our clients.

See a selection of our case studies

R&D, a long-standing story

Research and innovation: we grew up in it and we love it. Nearly half the team chose to extend the pleasure of studying all the way to a PhD. So we opened our own R&D division: Flowtide.

Testing new methods, discovering new tools or opening up new markets… bringing you the best solution to your needs by drawing on our research capacity, our knowledge in neuroscience, in neuroergonomics, and more broadly in scientific experimentation.

Our main focus is the assessment of cognition through a psychophysiological approach: neuroergonomics. For instance, we measure cardiac activity through ECG, perspiration through electrodermal activity, or visual attention through eye tracking, to better understand human cognition.

We've worked on advancing our user tests, improving the performance of top-level athletes, and better analysing user feedback (and all of it in no particular order).

The French Ministry of Industry also recognises our ability to carry out research and innovation work, by certifying us as a private innovation body, allowing our clients to benefit (where applicable) from the tax credit linked to innovation spending.

Akiani research library

We publish in the leading international conferences and journals

Yannick Daviaux, Dan Dutartre, Sami Lini, Fabien Lotte. 2021. 4th Dataquitaine Day: AI, Operations Research, and Data Science.

This presentation details our collaboration with the POTIOC team at the Inria research institute. We focus on the analysis of physiological signals to characterize the cognitive state of top-level e-sports athletes, in order to suggest training paths and performance improvements.