Over the last few years, the need to centralise control and treat data as an asset has become a source of competitive advantage for some. This trend will continue and this space will become much more competitive than it is today through the emergence of easy to use tools to diagnose, business transformation use cases and artificial intelligence breakthroughs.
2016 has been a really big year and no doubt this year will be even bigger for data science. We have recently seen 'Just walk out shopping' by Amazon - not the first revolution in the retail space by this giant, Transport revolution by Uber and Telsa and no doubt we will see more breakthroughs in health that accelerate cures, slow the ageing process and allow faster recovery times.
Supporting these achievements are passionate individuals and organisations that have very clear goals and a design mentality, allowing for flexibility and the freedom to explore. Under this model, we will continue to see innovation within artificial intelligence through frameworks like 'TensorFlow' by Deepmind that will give individuals who are passionate about data science, open-source tools and ways to collaborate to constantly improve their process and innovate on business problems.
For the techie or enthusiast, to install Tensorflow and to build your own artificial intelligence solution is ridiculously easy. First, install anaconda python and enter this command in anaconda command line:
pip install tensorflow
As a data and HR professional, I will also expect to see many new changes within the workplace. For any consultancy or software company, this space represents an amazing opportunity and we have seen accelerated investment in many things like recruitment where companies are looking to reduce unconscious bias and replacement costs through predictive candidate scoring.
Here are some interesting applications that can be done now if you have a data scientist and the support from senior leadership at a low cost (20-60k), depending on the problem.
Use Case on Employee Experience
We need to be more data-driven in how we understand our employees. For most, we know what our attrition rates are both for high performer and new starter but we lack the right information to do something about this. This is usually difficult as we tend to look at traditional employee list data and make decisions based on a cost centre hierarchy and an infrequent employee engagement survey.
Employee experience is about many little things adding up and one important view is allowing employees to contribute in a meaningful way with the right trade-off to ensure it is also done in a cost-effective way.
Recommendation - Imagine freelancer for work. Start by building an internal capability framework that moves away from a position view to a skill view of the organisation. With this information, you can then feed current state from various data sources and with machine learning and natural language processing build a tool that can be used to leverage talent within the organisation to complete work. An outcome is a tool that can be built with existing organisational data using open source tools to create more engaging work and save on resourcing costs by levering internal capability.
Customer Service - The tutorial above will help you to create a chatbot. Using deep learning through TensorFlow improves the reliability of chatbots considerably, offering near human-level experiences for customers. Data for this may be available in your CRM. What moves this from good to great is an online repository that is properly labelled and categorised to ensure the bot answers the customer problem without escalation.
Employee wellbeing is a top priority for most organisations and from what I have seen a largely untapped area of opportunity. This can be a sensitive area where intervention may uncover many issues that will be difficult to deal with however if you start with a small intervention group you can build predictive models to deal with core organisational issues in a productive way and see tangible improvements through regular reporting.
The data for this is an outcome i.e. the employee has raised a stress incident yes or no, representing a classification problem. Using this outcome and by combining other data such as manager-employee attrition, satisfaction scores, types of work performed, early warning data such as employee one-day absences, sentiment survey data and behavioural email metadata (if available), you should be able to create a reasonably accurate model to predict employee stress with an aim to improve employee wellbeing and allow for early intervention.
The key is to start with a small group by creating a high probability threshold, reinforce that this is not a witch hunt on managers and the aim is to address negative organisational behaviours and provide the right leadership education to managers to improve. As you start to see improvement the model will likely need to be refreshed and depending on how successful you can reduce the probability threshold to target more and different areas of the business.
Finally, as you become more proficient in utilising data science to improve the workplace you can leverage the same techniques and datasets to target other problems such as why people leave the organisation, how can I reach the right candidates in the market, how can I tap into the right talent at interview, reduce selection biases to create greater diversity in the workplace., the possibilities are endless.