A Thought Experiment: What If Analytics Models Were Semantic, Not Structural?
I’ve been thinking a lot about why analytics and forecasting platforms feel harder to use than they should.
Not harder to build, harder to think with.
Most modern data stacks are incredibly capable:
Yet the questions users struggle to answer haven’t changed much:
-
“Compared to what?”
-
“At what level?”
-
“Is this rolled up correctly?”
-
“Why does this number look wrong?”
This feels less like a compute problem and more like a meaning problem.
Where Meaning Lives Today
In most systems I’ve worked on, meaning is scattered across:
None of this is explicit.
If someone asks:
“Can we compare forecast vs actual by category this year?”
The system doesn’t reason about that question.
It executes SQL and hopes the result makes sense.
A Different Framing
What if we treated analytics as a semantic problem first?
Instead of asking:
“How do I query this data?”
We ask:
“What does this question mean?”
That leads to a different model:
-
Measures are concepts, not columns
-
Dimensions describe ways of thinking, not tables
-
Levels define where meaning changes, not just rollups
A Minimal Semantic Model
At its simplest, the model might describe:
-
Measures:
-
Actual Sales
-
Forecast
-
-
Dimensions:
-
Time (Day → Month → Year)
-
Product (SKU → Category)
-
This isn’t new, OLAP has done this for decades.
The difference is where this model lives.
Semantics as Data
Instead of encoding this in code or cubes, imagine:
-
The model is configuration
-
It’s versioned
-
It’s query able
-
It’s understandable by both humans and machines
Once meaning is explicit, everything downstream changes.
In the next post, I want to explore how this model could be used as a reasoning layer, especially when natural language enters the picture.
Comments
Post a Comment