I agree We use cookies on this website to enhance your user experience. By clicking any link on this page you are giving your consent for us to set cookies. More info
Patrycja Kujawa, Vice President, Information Management, Auto Canada [TSE: ACQ]
It can be tempting to believe that business intelligence has had its day. The move into interactive data visualizations left companies with a rash of busy, colorful dashboards that combined maximum data with endless flexibility and often failed to adequately inform business decisions. “But this is what they wanted!” became the exasperated cry of dashboard developers as this exploratory format taxed employee productivity and unearthed myriad data quality issues. Users increasingly asked to export underlying data into Excel where it was triangulated with other sources of truth and manipulated to meet business needs.
While some version of this played out across many organizations, AI and ML captured the collective imagination of executives. Companies rushed to form data science teams and the mess of dashboards from the golden age of BI faded into the background.
Fast-forward to today and many of the hoped-for gains of AI and ML have not materialized at scale. Over 80 percent of advanced analytics projects are said to fail as the same data quality issues that plague the dashboards are compounded by a disconnect between data scientists and the business. A focus on solving technical and statistical problems often eclipses the effort taken to understand the right problems to solve, operationalizing models or getting corporate buy-in.
Sometimes you need to go backwards to move ahead. Rethinking BI is far less sexy but if done properly, can add considerable value and provide a business-centric initiative from which to launch into other and more advanced data realms.
Tackling a collection of messy, colorful and often inaccurate dashboards is a daunting exercise. At our peak “death by dashboards” stage, senior leaders were frustrated by the paradox of dashboard abundance and information scarcity. The use of these tools by frontline workers, whose collective decisions determine the company’s bottom line, was spotty at best. Data quality problems, design issues and a lack of understanding of the decision points informed by this data emerged as top culprits.
The journey to better data quality is beyond the scope of this article but thinking differently about design and keeping focused on the company’s net profit goal was a crucial factor in our journey to BI 2.0.
“A focus on solving technical and statistical problems often eclipses the effort taken to understand the right problems to solve, operationalizing models or getting corporate buy-in.”
Adapting the design sprint approach pioneered by Google Ventures, we set out to develop an ideal state: a series of “diagnostic” dashboards that would enable a user to move from problem to action item in less than three clicks. But before we could think about dashboards, we had to think about what the business is ultimately trying to accomplish.
Starting with net profit, business leaders for each profit centre developed a pyramid of lagging and leading indicators. This pyramid gave us the framework to tap into the brains of various front-line subject matter experts and map out their decision-making process. We would begin by assuming a lagging indicator was below target and subsequently diagnose the problem through a series of yes or no questions. For example, if net profit was below target we would ask if it was because sales volume was down, if yes we would launch into a flow of “was it x?” yes or no, if no we would ask if it was a margin issue and build out that branch. The yes and no branches were built out until we had a tree of all possibilities mapped in order of causal probability with the most likely cause first. Activities that affect the indicator (that is, outgoing calls affect sales volume) would flesh out this sequence of leading indicators and we would then identify the information or intuition decision-makers used to answer this currently. In the case of intuition, we drilled down into its constituent cues or observations and sought a proxy data point. Once the mapping of business logic and data was completed, we focused on a dashboard design aimed at reducing cognitive burden through simple navigation, a minimalistic color scheme and a strategic leveraging of pre-attentive attributes.
A focus on ideal business knowledge state versus data constraints led to some grueling data consolidation and extraction efforts when these dashboards went into production. It also accelerated data quality and management efforts, led to a push to optimize systems and identified several use cases for modeling, algorithms and machine learning models. We even discovered where we could build in our first little bit of applied AI. All in all, BI 2.0 was an amazing catalyst for the next level of data maturity and monetization within our company.