With the intricate steps, loops, and reasoning required for robust data analysis, automating an analyst’s work could be as challenging as achieving true Artificial General Intelligence (AGI)!
I can outline at least six distinct steps in an analysis workflow, with steps 2-5 often occurring in iterative loops:
1) Develop hypotheses.
2) Retrieve the right datasets.
3) Apply the appropriate logic to transform these datasets.
4) Rotate, pivot, slice, and dice the data as needed.
5) Execute calculations and algorithms to compare and attribute results.
6) Interpret the outputs.
Upon examining this workflow, two striking observations emerge:
- The number of decisions required at each stage is substantial.
- The process exhibits high variance, with numerous possibilities to consider.
These factors lead me to believe that automating this entire workflow from start to finish is akin to reaching AGI!
However, it is striking how little automation exists today in this field. Analysts often navigate this workflow manually, facing significant complexity and tedium, particularly in steps 2-5.
By creating the right abstractions, we can greatly reduce this burden. Providing direct access to metrics and metric trees, and utilizing established algorithms can save analysts considerable time while enabling them to explore hypotheses that might otherwise be too complex.
I believe that the journey to automation should prioritize streamlining steps 2-5. Meanwhile, the crucial tasks of developing hypotheses and making interpretations are likely to remain inherently human and manual, despite the potential for biases involved.
