Scaling Corporate Culture Assessment with Artificial Intelligence

Company culture is a vital part of management strategy, and yet, it often proves elusive to manage. In this article, I propose a data-driven combination of a classic culture framework, Voice of Employee (VoE) methods, and the right text analytics technology that will yield scalable results, even for the largest organizations.

Importance of Culture Assessment

If the culture is healthy, employees are happy, motivation is high and productivity sky-rockets. “Companies with strong cultures saw a 4x increase in revenue growth (Source: Forbes, Found on Culture IQ)”. On the other hand, neglecting to measure and manage culture can have disastrous effects. A decaying culture is like a disease that spreads slowly, eating away at main functions of the company, weighing down motivated employees, leading to low productivity and ultimately complete failure. “When employees are undermined at work they begin to undermine their colleagues. These contagious negative behaviours cast a dark cloud overwork, and are estimated to cost organizations about $6 billion each year in health problems, employee turnover and productivity loss” (Found on bonfyre).

Executives cannot directly dictate company culture, but it is possible to measure and manage culture, just like any other part of a company’s functional area.

How to Assess Culture

There are several frameworks defining culture that can be followed for guidance. I find the classic framework by Johnson & Scholes, the Culture Web, to be the most pragmatic.

The Culture Web is a tool that brings to light an organization’s Cultural Paradigm, a set of assumptions about the organization which is held in common and taken for granted by most employees. There are six components to the web, which overlap and work together. In the below image you have a breakdown and explanation of each dimension.

(Diagram adapted from “Fundamentals of Strategy” by G. Johnson, R. Whittington, and K. Scholes. Published by Pearson Education, 2012, found on Management Centre)

For example, the organizational structure is known to all employees through the official company hierarchy which is the visible to all. In addition, many companies have “phantom” structures, fuelled by actual influence: “I know that if I go through the normal channels this task will take 2 months to complete, so instead I will talk to X who knows how to get things done.”

Using the Culture Web In a VoE program

Step one

Use the framework to audit the current state of the company culture. As any other Voice of Employee project, you can use an employee survey. You can enrich that data with more spontaneous or ad-hoc channels such as the public chat channels or employee walls. If you are planning to integrate a culture survey in your assessment, check out the breakdown of questions for each cultural dimension in this great article from MindTools. If you are planning to use text analytics to analyze the results, use mainly open-ended questions. For more on designing surveys for text analytics, click here.

Step two

Working back from the company goals, figure out what your company’s ideal culture would look like in each dimension. Here are a few examples: For a company with ambitious growth goals, you’ll need to design a nimble, results-driven culture: promote stories of how big deals were struck, install visible dashboards with revenue performance. On the other hand, if the company is focusing on stability and quality make sure this mindset is promoted: encourage stories that promote achieving excellence through attention to detail, routines of maintenance and regular check-ups.

Step three

Compare the existing Culture Web with the ideal one. Plan how to migrate and encourage the transition towards the ideal web. Aspects of the current culture should be reinforced or discouraged depending on the role they play in leading the company towards its goals. Introduce new cultural elements where needed. Modify ingrained organic culture structures, knowing that it will likely take time.

Challenges of Scaling

Building a survey inspired by your chosen cultural framework and sending it out to hundreds or thousands of employees is accessible to most companies. However, once the answers come back, dealing with the high volume of replies can use up a lot of resources.

I once had the opportunity to talk to the VP Strategy of a company with over 9000 employees. He mentioned that all the leadership team would take three days every year to read employee feedback from the annual employee survey. He would mention how painful it was to read about the things they weren’t getting right, but at the same time, he found these comments useful because they would provide the blueprints for improvement. As I was listening I was astounded at the level of commitment towards their VoE program.

In order to get to the bottom of the real corporate culture, the surveyor will need to use open-ended questions, which produce results in unstructured data, also called free-form text. These are the kind of questions that will reveal the best insights.

How text analytics can help

Advancements in text analytics technology have made it possible to analyze thousands of comments from employees while maintaining the same level of insightfulness as manual methods. Since there are several approaches to text analytics currently on the market, I will use the performance and features of our platform, Keatext, to discuss how one would go about doing the analysis.

Speed & accuracy

There are several benefits to using text analytics. First of all, the biggest value of text analytics is speed. You want your strategy to be based on fresh data. If you’re sending a survey to a company with 9000 employees, with 6 open-ended questions (one for each cultural dimension), even at a 70% response rate, it would take months to read everything, let alone analyze or annotate. On the other hand, our text analytics platform would finish processing everything in about 30 minutes. This frees up a lot of time to start exploring the data.

Very often the main concern when it comes to text analytics is accuracy. Natural language is very complex so it’s obvious that a machine can’t pick up subtle nuances like humor or sarcasm (yet!). Sometimes the system can even make obvious mistakes, with a general accuracy of around 90% (measured with precision and recall). It’s important to note that even human annotators will have 10%-20% differences in the way they interpret data, so perfection is out of reach for humans and AI alike. However, at scale, these issues are less important, because they don’t impact the overall analysis results. The analysis can also be tweaked by the user if needed. As our algorithms ingest increasing amounts of data, they also learn to optimize results.

Unbiased, credible results

Qualitative data is about to go through a renaissance with text analytics because the results are statistically significant once more than a few thousand datasets are uploaded. This allows for a very accurate view of complex systems such as company culture. Any assertion can be backed up by data, safeguarding the strategy against human biases.

Revealing blind spots

Choosing the right text analytics tool for your VoE program is vital because technologies differ a lot. The Keatext platform uses machine learning algorithms which extract information from context, meaning everything relevant will be reported on, without the need for human intervention. For more information on technical differences, read my previous article where I discuss the major differences between keyword-based text analytics and the next-generation technology. The bottom line is, when it comes to assessing culture, a system that can find unexpected insights is very important in order to be effective at exposing cultural assumptions management have been making. Systems relying on pre-defined keywords will likely find the obvious and miss topics that were not programmed before. Make sure your system is capable of discovering the unexpected.


    Leave a Comment