How Culture Amp is tapping generative AI

Australian employee engagement platform Culture Amp had been using its own artificial intelligence (AI) models to analyse employee sentiment and comments for its customers over the past few years, but it was only recently when it started to take its AI efforts to the next level.

In April 2023, the Melbourne-based company with a global customer base started a three-month trial of Vertex AI, Google Cloud’s machine learning and generative AI platform. Following the trial, during which Culture Amp qualified the security architecture of Vertex AI, among other things, it was ready to leverage the large language models (LLMs) available through the platform to build a new content summarisation capability.

Development of the new capability, which started over a month ago and is slated to enter production later in the year, would help organisations summarise employee survey comments into topics and actionable insight, automating a process that typically takes HR administrators up to hundreds of hours to complete.

“Even if every employee is only putting a handful of comments into their survey, HR administrators are still potentially faced with tens of thousands of comments, which they would spend days to read and categorise,” said Doug English, co-founder and chief technology officer of Culture Amp.

“The ability to bring the actual comments upfront and being able to synthesise the major themes very quickly means that HR administrators can take action much quicker than they would have otherwise been able to,” he said.

Additionally, HR administrators retain full oversight of which employee feedback traces back to the summarised insights, allowing them to correct the results for potential bias and organisational context while maintaining the anonymity of employees’ responses.

“The most important thing for us is keeping our data private,” English said. “With Vertex AI, we can use Google’s models, add our own models on top of that, or tweak the parameters using our data while keeping our data isolated and separate,” he said.

The company is already planning to build its own models with data from its organisational psychologists, as well as anonymised customer data, which it already uses when benchmarking how companies are engaging their people.

As Culture Amp doubles down on AI, it has developed an AI decision-making framework that centres around building trust with users, keeping control and expanding value.

“Building trust is about making sure that we keep the data safe and secure, and have good data governance,” English said. “It’s also about building confidence in the results that were presented to the user and being able to see where the comments were used in different aspects of the summarisation, because ultimately, the output of generative AI is that is there is no one right answer. But if the user can see how you got to that outcome, you will build trust even if they’re not 100% happy with the output.”

As for keeping control, English said it is about empowering users with the ability to expand on insights or add to the summary of a certain group within their organisation: “We want them to be still in control – they need to own the summary and any actions that have been taken off the back of that summary.”

In implementing AI capabilities, English said Culture Amp is focused on use cases where AI can save significant time in a bid to expand the value of the technology. “Something like the summarisation of comments is a really great example where we’re potentially saving hundreds of hours of work,” he added.