Why we’re launching a new automated service for employee surveys
In 2014 we started building our capability in using advanced text analytics to make sense of large volumes of multi-language feedback from employees.
From the outset we decided on using a human-in-the loop, inductive approach as the way of reaching the highest quality results. An inductive approach means that we learn the topics from the data itself ensuring that the classification and coding model is the best possible fit to what our clients’ employees are talking about.
Our clients have always been large, very sophisticated organizations. We work for some of the world’s most prominent employers and the survey and technology businesses that support them. An inductive approach works well for these organizations as they typically have large volumes of data to enable us to build highly-accurate models.
Over the last 6 years we’ve coded many millions of sentences of feedback. As analysts we’ve searched for patterns and similarity in the results, which topics are common and how employees describe common topics.
Whilst an inductive approach lets us build a model quickly for any question, typically with employee engagement surveys companies typically use variants of two questions:
What are the good things about working for this company?
What could we change to make working here better?
Our new service provides organizations a quick, very cost-effective way of analysing these and similar questions. Our model identifies about 125 of the most common themes we’ve seen mentioned in answers to these questions.
When the stars align
Building good text analysis models requires two key components; first you need sophisticated algorithms and secondly you need large volumes of ‘training data’ - examples of sentences from which the algorithms can learn from.
Over the last few years there has been significant improvements in the sophistication of text analysis approaches. New techniques have enabled data scientists to build incredibly powerful text models relatively quickly.
At the same time we at OrganizationView have been able to learn from the millions of answers we’ve seen to build detailed, accurate models of similar quality to those previously created by skilled qualitative researchers. As well as having large volumes of training data we’ve learnt and incorporated hundreds of small ‘hacks’ that we’ve learnt to ensure our models can perform well with real-life employee feedback.
This combination of factors mean we’re now in a position where we can deliver very high quality results for clients without needing a data scientist to build a custom, inductive model.
Our new automated ‘standard’ service
We want highly accurate analysis of employee feedback to be the new normal. We want to make it so affordable and simple that organizations don’t think twice about coding their employees’ feedback. We want to put the sophisticated tools and analyses that our large enterprise clients have been using in the hands of medium sized organizations.
We’ve therefore decided to make the starting price for what we think will still be the most accurate coding of employee text €200. This will enable you to code 500 answers of feedback. There are no long-term commitments and in the near future you’ll be able to pay with a credit-card. We believe that this will be cheaper than doing the job yourself manually, or even getting a junior member of staff to do it and yet you’ll benefit from our deep expertise. Anyone who has experience of manually coding text feedback will realise how painful this job is.
For those with larger volumes of feedback the price will depend on the quantity of feedback that you have. Needless to say we’re pricing it to encourage as many organizations as possible to have affordable access.
What you’ll get
On completing the coding you’ll receive an email with a link to your results. This will include:
Two datasets (spreadsheets) with each answer and the codes that we’ve assigned. One will be a ‘long’ format needed if you want to analyse the results in a Business Information tool such as Tableau, the other being a ‘wide’ dataset which will be ideal if you want to find relevant quotes by filtering the data in a tool like excel.
This dataset will include your original ID so you can join it back to the other survey results and if the answers were provided in multiple languages, our English translation.
A data visualisation of the distribution of themes that we’ve identified with some key statistics
A short guide on how to use the results and best-practice advice on how to integrate them in your communication to executives and managers.
Additional options
Coding your feedback is just the start of how we can reveal insight into your data, especially if you have over 1,000 responses. These include:
Thorough reports on your text answers including analysis of your feedback in combination with other pieces of information such as employee groups and answers to other survey questions.
Coaching on how to communicate this data to your executive teams
Data visualisations that bring your core messages to life.
Getting started
If you have a dataset that you want us to analyse please use the form below to get in touch. We’ll guide you through the process.