From a ten-thousand foot view of data and analytics, there are 4 key areas to focus in:
- Making visualizations
- Structuring & modeling data
- Doing analysis
- Building and operating machine learning models
1. Making visualizations
This involves being able to communicate data to others, stakeholders, your boss, your team, yourself. How well can you create visualizations that effectively communicate meaning?
Examples:
- Line charts or time-series charts
- Bar charts
- Scatter plots
- Scorecards
- Tables
- Comparison numbers (both absolute and relative (%))
- Compare vs. previous time metric (ie. vs previous period, vs previous week, vs previous month/year/etc.)
- Combination charts (ie. line chart and bar chart on same axis)
- Filters (Dropdown, Single Select, Custom, Regex)
Knowing what to use in different scenarios:
Do you know which chart/visual is best to use for different scenarios?
Examples:
- Time-series charts (line charts) for showing performance over time
- Bar charts for showing volume per a dimension
- Scorecards for showing key metrics
- Comparison numbers for showing different in performance since the last time period
2. Structuring & modeling data
This involves the data engineering, analytical engineering, and data wrangling side of things. How well can you manipulate and structure your raw data in order to work with it better for analysis and visualizations?
- What is the schema of data you will need?
- What dimensions will you need?
- What metrics will you need?
In bigger enterprises, more powerful tools is most likely present, ie. programming languages, ETL software, orchestration and scheduling tools, cloud solutions.
In startup companies, analysts work mostly with Excel and Google Sheets due to budgeting and time constraints to try and get small, quick wins up as soon as possible.
3. Doing analysis
Ad-hoc analysis involves taking a large problem question and diving into the data to understand 1.) what is the root of the issue and where is it coming from and 2.) ways to solve the problem.
Since a recurring report / dashboard is not needed for this, the analyst will not need to worry about structure as much. The main goal here is to identify root cause of issue and potential solutions.
If the beginning question is not so much of a problem as it is just trying to understand performance and where the company is currently at, then main goal will be to identify current performance, past performance, expected performance, and goals (if not already set).
Saving and storing these analysis is important too for referencing in the future, so structured organization of project files will be helpful.
4. Building and operating machine learning models
Artificial intelligence, neural networks, NLP, recommendation systems.
Before you get to this stage, you should have already identified and built:
- Reports or dashboards understanding and communicating current, past, and expected performance (with goals set)
- Monitoring systems to spot anomalies in data to see when issues come up
- A couple of scripts or processes that save time when digging into data to identify cause of issues
Once you’ve gotten the basics and fundamentals down, you’re ready to tackle machine learning.
In this step, you’re essentially taking data you already have and are currently collecting, and using it to make predictions and even take action in real-time. This step has the tightest feedback loop so far…
Without machine learning, what we’re doing with data is digging into it to understand our current state and behavior, determining what action steps to do next, and then implementing, all manually. With machine learning, all of this can be done automatically and in real-time, tightening up the feedback loop.
