Six Key Tips for Impactful Data Science Applications

Data science has the huge potential to completely transform businesses, and how people can utilize information and data to drive decision-making and change. However, too often the ability to utilize data science and analytics is left to specialists creating a slower turnaround on decisions, and insights and being less reactive or efficient. To tackle this I have spent a large part of my career trying to bring the power of data science to the fingertips of everyone, whether technical or not. This blog article captures my story on this. 

For the past 20 years my passion has been for data analytics, informatics, and modeling, now better known as data science. My particular interest in this field was in combining well-established, and emerging techniques in data science with what was then, a rapidly advancing technology: the internet. My question was: how to increase the usage and impact of any efforts by data scientists, as well as make their own jobs more focussed on data science, rather than often slow, error-prone programming? 

To this end, my focus has always been on utilizing the web whether it be a company intranet, a customer tool, or a public-facing website. In my previous role, we entered a trial with TIBCO Spotfire to enhance our data and modeling capabilities. TIBCO Spotfire had three key abilities that caught my interest:

1. The ability to call external data science tools such as 

   , TERR, Matlab, and now even Python.

2. Any analyses saved into your Spotfire library were instantly available as a web-based version of that tool, without any additional coding.

3. Spotfire also came with a Javascript API allowing you to easily integrate Spotfire with any other web-based tool.

Example of Spotfire integrated into another webpage. Source: https://www.environment.gov.scot/our-environment/people-and-the-environment/waste-and-resources/

This meant not only could data scientists access incredibly powerful data science tools such as R while utilizing a GUI-driven development of data science analyses but they could even publish it to a limitless size of audience without any web coding experience or skills. However, with some basic javascript skills, it would even be possible to integrate into other corporate tools for users to benefit from. 

Over the next 8 years, I began working to exploit these aspects as best as I could to bring the power of data science to as many people as possible, technical or not. To achieve this I developed six key philosophies to ensure the greatest impact possible for data science in my organization. These first two can be described as business based, the next four as design based:

1. Business-led, question-driven focus

For any data science exercise to be a success and have a great impact, it must have a clear and strong business focus. Therefore seek out the key question or questions your data science application is going to answer for your users and business, and maintain a focus on answering these needs clearly and effectively. This is not to say your application can?t go beyond these key questions however, be careful to maintain a focus on these and not dilute how well your application answers the key question over less relevant information.

2. Focus on collaboration and your customers

Data science should be an iterative and collaborative process with your customer. Your customer may be a manager, an external customer of your company, or even your own team member. In this context, consider everyone a customer and seek regular feedback, discussion, and interaction with however you want to benefit from your application. In my experience, I utilized an Agile informatics development process to ensure that collaboration was at the heart of any work, and the early release of applications was a founding principle.

3. Focus on simplicity, consistency, and usability

If you want your colleagues and/or customers to use your application, focus on making your data science application as approachable and simple for them. Remember, this will be one of many tools they likely have to use and learn, as well as many other priorities they have in their work. So make the tool as familiar to them as possible. This is why I find web-based applications so effective, as everyone already is comfortable with browsers and web pages, so they are much more likely to try using it over another new desktop application. However, people?s expectations of usability is now much greater than before thanks to the ease of use of phone apps, and websites like Google and Amazon. As data scientists, we must react and adjust to this.

Here are three simple design philosophies to help meet these expectations:

4. Least clicks, maximum gain

A customer should have to interact the least possible but yet get the most gain. In other words, question every input, does your average customer really need it, and does it add to answering their key question i.e. the reason they are using your tool? Is that extra piece of functionality really required, or is it only for a niche set of users? For niche functionality, hide it away. Don?t spoil the majority of users' experience for the minority?s benefit.

5. Use standards and guides

Using an agreed set of standards on aspects such as layout, design, even colors and functionality will bring consistency to your tools, while setting a high standard, and ensuring a consistent experience for your customers. This again helps increase uptake by users of new applications as they already know how to use them, and how they will work and look, removing or reducing any learning curve to a minimum. However, the use of standards should be done in a way that does not stifle creativity for your data scientists. 

Example: Using simple question-based navigation links, and search for data - Source: https://www.environment.gov.scot/data/data-analysis/ecosystem-health-indicators/

6. Design for 5 minutes of use

Design your tools and information in such a way that, if your customer only has 5 minutes of time, ask yourself: could they answer their key question, or get what they need? Is your tool intuitive and familiar enough, they will instinctively know how to use it? I was once given 5 minutes to present a Spotfire application to a board. This application had taken months of effort, had several technical innovations in it as well as over 50 datasets. Surely my efforts warranted more time and attention?! The reality is, for your application to have an impact, you must work on your customers terms, not yours. A board member as well as most staff are very busy and need fast answers.

Example: Using techniques such as auto-completes can make your tools much quicker, more familiar and intuitive for your user - Source: https://www.sepa.org.uk/data-visualisation/water-environment-hub/

Final thoughts...

Following these philosophies, I have had great success in taking data science beyond the desktop of data scientists and technical staff, to all staff in a business. In my last experience, this resulted in building over 140 Spotfire drive tools which were used by 80% of the staff in the business, covering all aspects of the business. Spotfire became an essential and recognized data science and reporting tool for the business. I hope reading this can help you with data science applications and the impact they have also.

Many of the tips and examples shown here can be achieved by using the techniques in this article: