Sense is Empirisys’ powerful, intelligent survey product designed specifically to dig deep into company culture.
This post, is going to explain how Sense’s AI Sentiment Analysis can be used to deliver rich cultural insight using free-text survey responses. This feature was built using OpenAI’s GPT- 3 model, the biggest language- based AI software commercially available.
In a previous blog post the process of training OpenAI’s GPT-3 model has been described so that it can complete a specific task: Sentiment Analysis. We explained how we created the AI Survey Analysis tool, and now we’d like to share with you how we are using it within Sense.
What does this tool do, and why is it useful?
Sense’s AI Survey Analysis analyses the sentiment of written answers and comments from surveys. The raw written responses are analysed utilising a bespoke AI model, which can understand the context and tone of the response to determine the sentiment in a similar way to a human. In fact, this tool performed almost identically to a person assigning sentiment during testing.
Written responses can offer fantastic insight into personal perception of culture within a company. Historically, survey text data has been left under-utilised due to the sheer amount of time required for manual analysis. Sense seeks to change this narrative: every survey response can be automatically analysed for sentiment, mining valuable metadata in a matter of seconds.
How can this metadata be used to deliver insight and visualise company culture?
The typical survey will ask questions about a range of different themes and is conducted across a company, which typically will have different groups, departments or teams. The culture of a company is an intricate web, and surveys can become rather complex to account for this: at Empirisys we specialise in developing intelligent surveys that can create a deep understanding of company culture. We also know that a great way to surface complex insight is through data visualisation. With that in mind, we need a means to visualise and explore the sentiment metadata from survey responses. Something that displays the interconnected nature of a company and reflects true feelings in an understandable way, to allow the data to be investigated and interrogated. For this, we have used a network diagram.
A network diagram displays naturally inter-connected data, with nodes that are connected by relationships. In our sentiment network diagram, there are two different types of node: the source node and the target node. The source node is typically a group within a company, for example a department, team or location. The target nodes represent different partitions of the survey, for example different question themes. The relationships between these node types can then intuitively represent the links between each department and each survey theme.
Here is a description of all the components within the network diagram and an explanation of the information that each component contains:
Let’s have a look at the Network Diagram in action. This example was created using real survey data that has been anonymised and is permitted to be used for demonstrations. It was conducted across a large industrial company from within the high- hazard industry, and contains 235 written responses from 6 Departments, answering questions about 15 Themes. The square nodes represent the Departments within the company, whilst the circular nodes represent the different question Themes that were answered.
This network diagram shows the links between every Department and every Theme. Whilst containing a huge amount of information and showing some useful synoptic patterns, it is a lot of information to take in at first glance. The Network Diagram is presented in Power BI using Zoomcharts’ fantastic Drill Down Graph Pro, which allows for incredible interactivity to explore patterns. Check out Fred’s blog post to see how we implement network diagrams in Google Data Studio with custom visuals.
Synoptic information obtained from the Overall Network
• The most written answers came from the department: Production (Largest square node)
• The most answers were written about the Theme: Clarity (Largest circular node)
• IT had the most negative average sentiment (Orange node: 20-40% of responses were positive)
• Teamwork and Work Culture were the most positively answered themes overall (Dark green nodes: 80-100% positive responses)
What can the network diagram show?
By filtering the network diagram upon the Departments, Themes, Engagement and Sentiment, the Network Diagram can break down the information to show myriad permutations of cultural patterns and directed insight. For example, what if a specific department manager is interested in what their business area think about different topics?
Each department can be viewed individually, then the links show the sentiment of answers from each department about each topic. Additionally, the colour of the link can be compared to the colour of the topic nodes to compare how the sentiment of that specific department compares to the overall sentiment about that topic. For example, Logistics has a negative outlook upon Clarity: perhaps this is an area to focus upon and investigate further. Questions about work culture, Accessibility and Teamwork were the most positively answered. Of course, this information can then be linked back to the raw survey responses to see the exact answers, but the synoptic view that this visualisation shows allows for valuable exploration.
As another example, the overall network diagram can be filtered to show positive links with high engagement (high proportion of answers about a topic) to show highlights in the company’s performance in the eyes of the employees:
There is an overwhelmingly positive view throughout many departments that Teamwork is a positive aspect of the company: there were many written responses that were analysed using AI and identified to be positive. Additionally, two departments had high, positive engagement with the Theme: Accessibility. This kind of pattern be clearly visually represented with ease using filters upon the Network Diagram.
How can you utilise this powerful tool?
These are just two examples of how the Network Diagram can be used. The amount of information available at the fingertips of the analyst allows for incredible insight to be uncovered with ease. Use cases can be both exploratory or confirmatory: gut instinct and feelings can be ratified, and investigatory analysis can be conducted to uncover latent cultural patterns. Other examples of use cases include:
• Identify siloes: groups of departments with similar feelings towards a particular topic
• Survey breakdown by department/ theme
• Specific areas of strength or weakness within the company
• Identify popular response topics
• Synoptic sentiment view over the entire organisation
Utilising the power of AI and sophisticated network visualisations, tasks that were previously intensely time consuming can now be completed with ease to deliver rich and valuable insight into how a company perceives different topics. This is an incredible leap forward in the exploration and visualisation of text- based survey responses.
We’ve shown the tip of the iceberg of possibility of Network Diagrams to visualise and explore survey responses: to find out more about how Empirisys can aid in the analysis of your company culture, please get in touch at: info-sense@empirisys.io
