Two of the biggest obstacles standing in the way of optimal data analytics are nearly impossible to avoid: not everyone uses the same methods and not everyone thinks the same way. Sometimes we get too involved with our own data, or the potential impact our analysis can make, before understanding some foundational requirements that are needed.

Therefore, MIPS team within Target created an outline of a five-step model to provide analytics. MIPS stands for Metrics Ingestion and Prediction System. Using this tool, a team or organization possesses the ability to uncover underlying trends and take automated action based on these trends. The levels consist of storage, visualization, predictive analytics, prescriptive analytics, and automation. These levels are categorized as best practices for teams or organizations to conduct analyses on valued data used for analytics.

Level 0 – Value

We added this section to make sure any work being done won’t be in vain. It’s imperative to define the value, desired outcome, or goal. This is a prerequisite to continuing any technical work or moving on to the initial level, data storage integrity.

Level 1 – Data Storage

Infrastructure needs to be set in place to consistently deliver and store data. Data should be consistent, accurate,and valid. Having incorrect data at level 1 will cause a compounding cascade of harm at every subsequent level resulting in misleading, or just plain wrong forecasts, anomalies, and recommendations. Data storage integrity paves the way for visualization

Level 2 – Data Visualization

Dashboards are created to provide insight into underlying trends and gaps, and to inform decision making. If there is a noticeable gap of missing data, then a pipeline may need to be fixed before continuing. Some cases can also stop at this level. If the data seems to be increasing or decreasing at a steady rate and machine learning or more sophisticated methods are not needed, this can be relatively easy and quick to implement. This is also a good time to validate that you’ll be able to provide the value you expected.

Level 3 – Predictive Analytics

Forecasts illustrate the norm. With quality data, forecasts can help teams plan for what’s to come or what should be currently happening. If data seem to be outside the norm, then alerts and any plans for action should be put in place. Combining forecasts with anomaly detection enhances quality, stability, and reduces the need for manual observation.

Level 4 – Prescriptive Analytics

Recommended actions based on the data analysis allows for proactive action to be taken. At this step, recommendations can either be accepted or ignored. When trust is established between the team providing the recommendations and the team accepting them, automation can speed things up.

Level 5 – Automation

Automated action reduces manual intervention, for a self-adjusting system. When reaching this level, systems should be fluid, streamlined, and reliable.


Note: Not every use case will need to follow all the levels in this template. However, for use cases that will eventually need to have automated recommendations from predictive systems, it’s best to have a foundation set. With stability in each of these levels, value will be added faster than ever before.

About the Authors:

Francisco Cancino is an Engineer, Tom Anderson is a Product Owner, and Michael Yee is a Lead Scrum Master for the Metrics Ingestion and Prediction System (MIPS) team within the Infrastructure and Operations organization at Target.