To address this question, let's first define what we mean by analytics work.
We’re not talking about open-ended research where analysts explore data for insights—this type of work is inherently difficult to streamline.
Instead, we focus on the structured workflow where decisions are made, actions are taken, and results are reviewed to inform new decisions. This feedback loop forms the core of most organizational analytics efforts.
Let’s examine this further.
An organization makes a lot of decisions - big and small - across multiple functions. Vast majority of these are not executed with controls so that it is trivial to see the impact. In fact, for some of the most impactful decisions, it is impossible to cleanly test.
As a direct consequence, the work of analytics is to tease out what is working, and what is not among a morass of decisions across points in time that affect an entire ecosystem of connected metrics. Stated this way, analysts sound like miracle workers, and many are!
Let's explore whether this workflow can be broken down into more manageable, assembly-line components.
1) Modeling Metrics:The first step is to map out the universe of observations and codify the logic behind them. This involves defining and organizing the metrics.
2) Modeling Metric Relationships:Although not widely practiced yet, it’s possible to codify the network of metric relationships using frameworks like metric trees.
3) Modeling Segmentation:Similarly, mapping each metric node to various segmentation cuts can provide deeper insights. This approach allows for a more granular analysis of metrics based on different segments.
By addressing these three components—modeling metrics, relationships, and segmentation—you create a comprehensive representation of the business process directly linked to data.
With this framework in place,
- Could we consider common, recurring questions to be a “workflow”?
- Could we establish standards and practices around these workflows?
- Could these standards be modularized or templated?
As an example, let’s take a sales funnel process where the goal is to maximize the key output metric of contracts closed.
If this funnel process were modeled via a metric tree framework, it’s now a lot more tractable for software to traverse the tree, search through segments and provide valuable information - which the user needs to interpret for sure - but the grunt work is effectively automated away.
Maybe there is another module you can use to extract a list of accounts that are stuck in a specific step in the funnel.
While we may never fully achieve assembly-line efficiency in analytics, breaking down the process into these distinct components and codifying them can streamline the work.
This allows for better-defined operations, ultimately making the analytical process more efficient and effective.
