ESOMAR’ 20 questions for AI-based research and insights products
PART ONE OF THREE
1. What experience and know-how does your company have in providing AI-based solutions for research?
Fathom is founded by a team of researchers who have spent the last decade specializing in open-ended research at scale, leveraging AI.
As consultants, we worked in complex political topics, on tight timelines, with ever changing context - and needed a solution that could accurately identify and map themes from data sets with multiple open-ended questions, fielded to 10,000 respondents, in 72 hours or less.
We experimented with many methodologies, leveraging many different AI approaches, integrated into many iterations of what has now become the core of Fathom.
Since then, we’ve processed more than 30M+ open-ended responses, our Director of Engineering has literally writen the book on AI Dev Ops, and we’ve delivered the highest quality text analytics for a rapidly growing list of the most discerning customers who we are grateful to call subscribers and thought partners at Fathom.
2. Where do you think AI-based services can have a positive impact for research? What features and benefits does AI bring, and what problems does it address?
This could be an essay (or a few!) all on its, and we recently published one that you can check out for a deep dive!
The short version, we are deeply committed to a human-centric approach that leverages AI to assist researchers in achieving their goals quickly, confidently and accurately - and that always keeps the researcher in the driver seat.
3. What practical problems and issues have you encountered in the use and deployment of AI? What has worked well and how, and what has worked less well and why?
With regards to data analysis and text analytics specifically, we’ve learned through a ton of trial and error, that AI is very good at well defined tasks with clear parameters for success, at pattern detection, and at summarizing results.
It’s much less good at deciding what the parameters for success should be, which patterns matter within the context of the work or project, and with staying clear of embellishments that sound good but take away from accuracy in summarization.
To that end, AI is super powerful when it’s:
Guided by researchers who understand their goals and context
Leveraged to take on mundane tasks (like coding open-ended themes) and pattern detection (like identifying meaningful signal in those themes)
Easily verified, validated, edited and adjusted by researchers who are in full control of the end results
Under these conditions, well deployed AI solutions can not only increase the quality of research outcomes, but can also increase the quality of life for researchers.
4. Can you explain the role of AI in your service offer in simple, non-technical terms in a way that can be easily understood by researchers and stakeholders? What are the key functionalities?
Fathom is designed for the AI and human researcher to collaborate at every step of the process. Here’s what that means.
When you import your data to Fathom, a simple interface will say “It looks like this is open-ended data (you probably want themes and insights form this)” and “It looks like this is closed ended data (and you probably want it for segmentation)” and “It looks like this is meta-data (and you want to map back to this in export)”.
AI did that for you. And now you can verify that it got it right. Recategorize if needed. Turn anything on or off, according to your preferences.
Next, you’ll be prompted to provide some strategic guidance. This is going to be fed into the generation of the code frame. It will help the AI to interpret and code data to support your research goals.
Now you’re ready to process. The AI is going to take a first pass at building the code frame and coding your data. This is actually a bunch of software, and a few different models, working together to accomplish all the steps necessary to do this well. For you, it’s just a button.
Once the AI has taken that first pass, a human reviews and refines the AI’s work. The human expert (could be someone on your team or someone on our team) validates, revises and refines the coding, adding additional strategic focus and nuance.
And now you're ready for analysis of your fully coded data. AI will assist in summarizing the themes, and assist with identifying meaningful signal and variation in your data. You can validate and verify these findings in the coded data, which you can move to seamlessly, and can update the summaries to help ensure they perfectly meet your reporting needs.
5. What is the AI model used? Are your company’s AI solutions primarily developed internally or do they integrate an existing AI system and/or involve a third party and if so, which?
Fathom was developed over 5 years, partnering with OpenAI, Microsoft, and Anthropic, and combining with our own models, built on tens of millions of proprietary data points.
Our terms with our partners give us the ability to leverage all of their models, but do not give them any access, rights or training permission to Fathom data, or the data of any of our subscribers.
Be sure to sign up for to get notified when release Part 2. Next up, Part 2: Ethics, Transparency and Human-in-the-loop and Part 3: Data Governance are published!