Forum Discussion
Doman Driven Design for Business Intelligence
Further thinking on this..I have ended up going instead down a 'Behaviour Driven Development' route using 'behave' behave 1.2.7.dev6 documentation
Scenario: Fit Pareto/NBD model with PyMC
Given I have a dataset of customer transactions
When I fit the Pareto/NBD model using PyMC
Then I should obtain the model parameters
That could be replaced and enhanced with an ever-more nuanced description of self-evident customer value such as presenting churn predictions in the form of number of churning/non-churning customers, their likelihood of churn and their dollar values. This seems to be going down a similar route: Tutorial: create, evaluate, and score a churn prediction model - Microsoft Fabric | Microsoft Learn
once again..I will update when i get further.
Ha, good old von Moltke! Yes..have looked at Purview but its very clunky. Interesting you say Power BI has such an option..to what are you referring?
Ah...again a great memory jogger...i remember seeing that but it slipped my mind. Thanks.
It occurs to me that certain predictive analytic artefacts could be measured as inherently valuable. I am working in the SaaS world an have been looking at Blog Posts - PyMC Labs (pymc-labs.com)
-If a visualisation identifies future customer lifetime value, broken down by customer.
-If it then identifies which of those customers is likely to churn by descending order of value.
-If it then identifies day by day what the changes to that are...has the customer churned or been retained. This is shown in a Confusion matrix - Wikipedia which is a pretty standard ML artefact.
It seems to me that the the 'value' of that analysis is self evident and independent of any subjective assessment. It simply reports that the company has retained or churned $x per day against expectation. That might be a straw man starting point against which to judge other BI artefacts...how close to that repeatable, measurable standard can you get to with other visualisations that require more subjective validation. Let me know if you have any appetite for a quick conversation.this conversation will be anything but quick. Just for context - how big or small is your company? What's your estimate on the number of data subjects ("objects") in your business ? How many reality tunnels do you think you have?
Good point about the short conversation!
Its a startup SaaS company with the standard set of SaaS metrics you would expect - lets say 10-20 to start with or even just the 5 or 6 mentioned below.
Your use of the term 'reality tunnel' was unfamiliar to me (Robert Wilson???) .
It is certainly an interesting concept when thinking about self-service BI but to keep it simple I would start with simply one perspective...that of the person asking for the analytical product. The acid test is what is the expected and actual incremental business value of that product and how would you measure it.
If I can propose a reference point of my CLV/churn example that assesses the entire CLV by customer at varying risks of churn each month and then measures performance against that then that would seem to be a good stake in the ground to assess a new feature or use case. This is my idea and I intend to start experimenting but I thought I would explore any existing work first.