From deploying a new AI product to more than doubling our polling volumes, 2023 has been a busy year at Focaldata.
2023 has been a big year at Focaldata. We ran over 2,000 polls on our platform in 2023 (+137% compared to 2022), answering over 91,000 questions in the process and serving over 100 clients. We shipped a beta version our AI qual interview tool FD_Chat. Australian Prime Minister Anthony Albanese spent the best part of a TV interview in October casting shade on an (ultimately very accurate) MRP forecast of ours he didn't like. And we've also been busy hiring: onboarding over 20 researchers, engineers, data scientists, advisors, and operations specialists over the past 12 months.
We've seen unprecedented growth in the past 12 months, tripling the number of responses and more than doubling the number of polls run in the year. We've also conducted research in more markets than ever before. We have more tracking studies live than at any point, including delivering daily and weekly multi-market tracking data for the likes of Little Moons and New Balance. You can find more information about our work this past year with Mediahub and New Balance here.
Data quality removals have skyrocketed in this period. We believe this is linked to two trends, one internal to Focaldata and one external to the market more broadly. We've been doing more polling in markets where fraud and poor quality data are more common, such as India, so this has meant more removals. But panel quality has been a problem for a long time and trending down in almost all markets. The recent progress for generative AI and large language models (LLM) has accelerated this: without smart checks in place bots can be more effective and profitable than ever before.
We've been investing heavily in our automated data quality defences to counter this, leading to almost doubling the % of respondents who are removed on quality grounds. We expect that the traditional 10% industry-wide quality removal benchmark (i.e. researchers expect to throw away that amount of data on quality grounds) will quickly start looking like 20-30% as everyone catches up with the pace of change.
We've been hiring at pace across the year, expanding the team with over 20 new researchers, engineers, advisors, designers and operations specialists. Mousom Gupta joined as Head of Engineering from fintech Plaid. Mousom's prior experience includes with Stripe, Amazon, AirBnB, Facebook and Lyft. He's since spearheaded the recruitment of several world-class engineers to join our existing team helping to make our platform faster and more efficient plus work on new product development. We welcomed our first people lead Mariya Hristova who has been instrumental in the rapid growth in the team. We've brought on several new researchers and data scientists to partner with clients on their research and analytics projects. We also made our first product and design hires.
It's exciting to see what such a talented gang can get up to next year. You can find more information about the whole team here. If you know anyone who might want to Focaldata, then please check out our careers page here: we're hiring.
Our engineering team has shipped hundreds of new features, fixes and improvements over the past year, plus a new product. A couple of highlights below.
This month, we launched the public beta of our AI qual product FD_Chat. With FD_Chat, our clients can collect 1,000s of in-depth, AI-moderated interviews at 10x the speed of traditional qual methods. They can then get back unique and actionable insights that you can explore in a conversational style, asking questions, digging down into key findings.
We believe this is the future of research. Why? Because every decision maker wants to understand why numbers are moving. Right now, they rarely do.
The reason why so little research contains these insights is that qualitative interviews and focus groups are expensive and slow — and there's rarely enough data to derive real insights. Open-ended survey questions are valuable in many contexts but intelligent follow-up questions are how you uncover why people think, act, behave the way they do.
Recent academic studies of this topic also show that AI-facilitated interviews outperform traditional qual methods as well as open-ended survey questions (see paper). We've also observed a positive experience for respondents: 95% of them said they’d like to participate in another AI interview.
At Focaldata, we want to give clients the most representative, accurate data at the fastest speed possible. We've always wanted to reduce the time taken to produce weighted data (previously a manual step). This year, we were the first company to automate it. This means that if you run a poll on our platform, then you can see weighted results when live, and you don't need to wait around after fieldwork for weights to be attached. You can download weighted data at any point, for free.
Earlier this year, we started working with the FT's chief data reporter John Burn-Murdoch to provide regular polling for his weekly column Data Points. He uses statistics and graphics to dig into the most pressing issues of the day, covering everything from the economy to climate change, social issues and healthcare. The first FT/Focaldata study investigated the challenges posed by one-dimensional political polls. We polled 1,009 British adults and 1,027 US adults, focusing specifically on their attitudes towards immigration. The majority believed that immigration was too high. This concealed more nuanced (and positive) attitudes when framed with further context, such as why people were immigrating or the jobs they were coming to do. More to come next year.
In October, we ran a 4,600-person poll in Australia combined with local-level MRP estimates to understand where the Voice referendum in Australia was heading. We we forecast 39% Yes vs 61% No. Our prediction was correct within 0.25% — the most accurate forecast by any pollster. We'll be running more public MRP forecasts next year in the run up to the elections in UK and US.
Sky News interviewed Australian Prime Minister Albanese just before the vote and questioned him about the Focaldata forecast. PM Albanese was leading the Yes camp. His response: “I have seen reports, I'm amazed that they actually get written, that seems to be based on some UK company based upon modelling without talking to anyone...You know, we'll wait and see what happens on Saturday." Our prediction (based on survey responses from 4,600 Australians) was also picked up by The Economist and most of the Australian broadsheets in the run up to the vote.
Thank you for any and all opportunities we've had to work together over the past year. If you're a client today, thank you for your continued partnership. If we're not there yet, we hope we can get you over the line next year with the right project.
Subscribe to get the Bi_Focal newsletter delivered directly to your inbox.