Most organizations are now working on data analysis, but not all projects are yet delivering the desired business value. Productivity often proves to be a major hindrance when carrying out certain tasks. What are the biggest stumbling blocks and how can we avoid them?

The fact that every company is looking for solutions to achieve productivity improvements was recently made clear at the annual Gartner Data & Analytics Summit. As part of the conference, SAS organized a panel that provided customers with a platform to discuss obstacles and solutions when executing data projects. Below we discuss the six key findings.

1. Integrate your data strategy into the company culture

Data literacy (Data literacy) remains a major barrier to productive work with data in most organizations. The term refers to the ability to examine and use data meaningfully. Compared to more specialized profiles in the company, business users in particular have difficulty clearly formulating a question for a data project.

Instead of investing all of their energy in training programs on how to use specific tools, companies can better focus iteratively on communicating about the importance of data in company culture. What data is available? What can you do with it and where does the company want to go? Combined with accessible and automated technology (we’ll come back to this), this will significantly lower the barrier for many users.

2. Define what you want to achieve

Of course, in order to deal with data quickly, you also need to be able to identify information quickly. The lack of data governance and a good data catalog is therefore a second obstacle that many companies struggle with. On the one hand, you have to prepare the data to a certain extent, as a data scientist cannot achieve much with raw data. On the other hand, editing all the data in advance would also be detrimental to productivity.

A data set is just an intermediate step and not an end goal. Therefore, try to find a middle ground when preparing data and focus on information that is sure to add value to the company. Ultimately, no data set is 100% perfect. Then move away from the idea that you have to capture all the data correctly and determine what is sufficient in the context of your project. For regulatory reports the bar can be very high, while for an average business case 80% may be sufficient. You can read more about this in the next article in this three-part series.

3. Storytelling: Share success stories with your users

If you want people to adopt data analysis best practices and tools, you need to communicate their importance to the business. Instead of communicating dryly about it, you’ll have more impact by highlighting success stories. Make results tangible and show how analytical models create added value for the company. Once they understand the bigger picture, they will be more motivated to use data to support their work.

Make results tangible and show how analytical models create added value for the company.

An important footnote: Don’t view a successful pilot as a measure of productivity. Some organizations are stuck in a vicious cycle of pilot project after pilot project. There is always another case that deserves their attention, which is why most pilots never put it into practice. A model is therefore only a success story if it actually delivers business value.

4. Design thinking makes your projects concrete more quickly

Design thinking is a popular approach that also offers many advantages for analytics. Essentially, it means that every project begins with a clear description of a problem before thinking about a solution. Of course that helps too Data literacyThreshold because every stakeholder needs to understand the problem. Too often, for example, we see data scientists and executives sitting around a table together and not knowing what they are talking about. The design thinking approach ensures everyone is on the same page.

5. Top-down is more effective than bottom-up

Goals that come from the top are often better formulated and produce results more quickly – as opposed to a bottom-up approach, where goals come together more randomly and are more difficult to achieve. Therefore, try to support and communicate the goals of your data projects from the top down as much as possible.

6. Focus on automation and self-service

Automation goes hand in hand with productivity. A modern analytics platform therefore handles many tasks automatically. Consider installing and commissioning test environments, integrating and testing developments, retraining existing models, etc. Automated workflows are increasingly taking over tasks that previously required manual steps. And this not only benefits productivity, but also improves the consistency and quality of a model.

In addition, tools offer more self-service functionality, allowing users to work more autonomously and reducing waiting times within a project. For example, looking up data sources in a catalog or setting up simple pipelines using GUI tools. But also for building a cloud infrastructure via specially integrated apps that use the APIs of public cloud providers. Both automation and self-service can become successful and scalable if you apply them in a governance framework and all responsibilities are clear.

From 80/20 to 20/80

Measuring is knowing… You can only improve your productivity if you know where things are going wrong. This also applies to developers of data analytics solutions. For example, SAS works with a specialized team that analyzes how much time users gain or lose when building models. They rely on a tool that brings together all information and insights from the SAS environment on a dashboard. Among other things, this approach has ensured that users now work with uniform interfaces and are no longer dependent on separate tools that each support one task. This improves collaboration in the organization and of course productivity.

On average, data scientists today still spend 80% of their time preparing data and only 20% effectively developing models. Gavin Day, Executive VP at SAS, therefore expressed the ambition to reverse these numbers in the long term. As we made clear above, there is still potential for increasing efficiency in many areas. For example, if the data scientist no longer has to deal with raw data or anonymizing data for privacy and security reasons, we can significantly reduce this 80%. Since data models have the ultimate goal of making faster and better decisions, business operations will certainly benefit.

This article is part of a three-part series. In the next part, you’ll learn how new initiatives around data lead to greater productivity. Among other things, we delve deeper into the importance of self-service and a data catalog for users.

This article is a contribution from SAS.