Made in-house at MIT
We tried to do this the easy way. LDA would do the trick, we thought, and if not another off-the-shelf approach would analyze employee feedback well. A few days later, we found out that doesn’t work. A few years later, we finished building a solution that does.
Designed for your use case
Everything about our platform is designed to better understand employee feedback, specifically. Trained on millions of real employee feedback comments, our platform knows that “SVP” should not autocorrect to “RSVP” and that “too many chefs in the kitchen” rarely refers to cooking.
Analyze like a Pro-fessor
If you are trying to measure strategic agility, it can help to ask someone who has taught a class about it at Harvard Business School and written four books on the subject. Our platform is human-supervised by subject matter experts, ensuring the AI knows its stuff.
Three Benefits
ACCURACY: 90% OF THE TIME, IT WORKS ALL OF THE TIME
CultureX uses “human in the loop” supervision to ensure significantly higher rates of accuracy than most NLP solutions, around 90%. Each term in our proprietary dictionary of over 20,000 terms has been human-checked by subject matter experts, and only the most accurate terms make it into our platform.
RECALL: WE PICK UP ON “KAFKAESQUE”
High-frequency, information-rich terms are relatively easy to catch. It’s the long tail of low-frequency, information-rich terms that are difficult to identify, but cumulatively they are cited so frequently that you cannot afford to miss them. Our platform is designed to pick up on information-rich terms even if they are only cited a few times in a million.
DETERMINISM: AT EVERY STEP, WE KNOW HOW IT WORKS
Call us old fashioned, but we prefer a deterministic model, where we know what is causing everything to happen. When our platform says a company’s innovation has improved since the last quarter, we can drill back down to the exact phrases that resulted in that assessment. And since we always know what is driving results, we can always give actionable advice.
WHAT’S IN A TERM?
Terms are the building blocks of the CultureX platform. Terms are not keywords, they are more like semantically coherent phrases or ideas. A term specifies the distance between different words constituting the term, exclusion phrases that would prevent the term from being triggered (e.g. “vision” might refer to strategic vision, but not if the employee writes “vision and dental”, in which case they are likely talking about their benefits), stemming and lemmatization (e.g. “the management” means something very different than “my manager”), as well as other information. Using this approach, CultureX can capture a wide variety of semantically rich phrases in a flexible and accurate way. Some of our favorite terms are “dilbertesque” (buzzwords), “kumbaya” (collaboration), and “let flowers bloom” (creativity).
In a milllion employee reviews, employees will mispell the word “bureaucracy” an average of 15 different ways. Our favorite way is “burocrazy”.
22,000
Terms in the CultureX Platform
200+
Cultural topics measured
Want to learn more?
Get in touch