Discover powerful insights from smaller panel sizes
When commissioning market research have you ever had to trade off accuracy for cost ? Or granularity for speed? For the world’s top companies, it’s an ever present problem. Focaldata squares this circle by combining your survey data with information from other sources, and then using machine learning (hierarchical bayesian modelling to be exact) to extract new, deeper insights that are more accurate.
Despite the scary name our modelling is quite intuitive. Our ability to go deeper and still be more accurate than traditional methods, is that traditional methods throw away information,
Consider estimating what young women in Scotland think from a representative UK sample. First, traditional agencies filter out respondents in England, then filter out respondents who were men or old. Finally, then would look at the remaining subset and say “there you have it, this is the answer”.
But if you take a step back for a second, you realise that this might not be optimal. If you just had a dataset of English people, would that tell you nothing about what young women in Scotland thought? Of course it wouldn’t. It would give you insight into what the young thought, into what women thought, and indeed what young women thought.
So why ignore so much of information contained in the data? Multilevel regression extracts information from all respondents, pooling data, and thus increasing the accuracy of our predictions (e.g. using information from women in England, to understand what women in Scotland think).
To do this requires a lot of complicated statistics, but at its heart it is rooted in common sense.
HOW IT WORKS
Upload existing data or commission a new survey through our online application
Our proprietary database of the UK population contains millions of different demographic types of people
We use a range of data science techniques, including multilevel regression (MRP), naive Bayes, SVM, neural nets, ensembles and decision trees
With better-informed strategies and campaigns, discover new revenue streams - and cut costs as our engine leverages previous research