How to build a Model
Last updated
Last updated
Advizor has contracted algorithms, from Rapid Insights, that allow the application to compare a target set of data, against the rest of the dataset. The algorithms quantify what makes a subset, of the dataset; distinct from the rest of the dataset. For example, Advizor has created a Customer’s.advm project, that is available in the Advizor AnalystX Demo. Within this Customer’s list, a user can define which customers are more valuable than others. I.E. Do the highest prospect Customers live in specific Regions or work within specific Industries, etc. Also, for example, the Customer’s.advm project contains a chart titled ‘Margin % v. Revenue’. Given that most Users are interested in ‘High Margin’ and ‘High Revenue’ customers, the user can select an area, from the Scatter Chart, of prospects that are ‘High Margin’ and ‘High Revenue’. The user can then identify common attributes that will be utilized to target certain groups.
Launch the Advizor AnalystX v7.3 client and select the ‘Customers’ project, located from the ‘Demo Projects’ list.
The following Customer.advm project loads.
Select the 'High Margin' 'High Revenue' Customers, from the following splatter chart.
Select ‘Model’, located in the upper right-hand corner of the screen. Then select the ‘Predictive Modeling’ button.
Select the ‘New Model’ button, from the ‘Predictive Modeling’ panel that slides in from the right.
Enter the text ‘High Margin High Rev’ into the ‘Model Name:’ field. Select the ‘Target from Selected…’ button. (This maps the graphics selection to the table.)(Maps selection state into the data.)
Select ‘Yes’ button, to save selection state into the table.
From the ‘Explanatory Fields’ list, select the variables (factors) that may contribute to ‘High Margin’ and ‘High Revenue’ customers. (Note that a user will normally want to avoid ‘Explanatory fields’ (variables) with a large number of unique values.) (Also Note that the user will need to un-select the fields that the user is attempting to predict. In this example, the ‘Margin %’ and ‘Revenue’ Explanatory Fields, because those are being predicted.)
Set the ‘PValue’, from the ‘PValue’ dropdown list. (The ‘PValue’ setting defines how statistically strict the model is (threshold). “How different the target data needs to be, versus the rest of the data, in order for the model to flag the distinctions.”
Select the 'Train Model' button, to start the Predictive Modeling.
The 'Predictive Model' processed all the different variables and predicted that only two attributes were predictive of 'high Margin' and 'High Revenue'. Those two were 'Industry' and 'Region', represented in the following '% Contribution to Model' Chart.
Note the '% Concordance' number field. This is key because it identifies how much of the outcome can be explained by the variables in the model. (In this example, the '% Concordance' is 78.9%, so this means that there is a 21.1% possibility that there are variables; of which, could make a Customer 'High Margin' 'High Revenue' but were not selected for the Model.
Another important datasheet is the 'Model Terms' chart. Note specifically the 'Coefficient' column. This identifies which Coefficient is most likely to result in a 'High Margin' and 'High Revenue' Customer. In the following example, the 'Industry' with the highest coefficient is 'Industry.Government' and second is 'Industry.Construction'.
If a user hovers over the 'Industry' and 'Region' graphs, located on the '% Contribution to Model' panel, the user will see how the graphs compate to each other, based on the variables of the prediction. In this example, Industry is .6 (60%) of the model impact and Region is .4 (40%) of the model impact, as to why someone would be 'High Margin' and 'High Revenue'. (So in this example, Industry is 60% and Region is 40%, of the 78.9% 'Concordance%' that makes a customer, via this dataselection, 'High Margin' and 'High Revenue'.
Additionally, the 'Predictive Modeler' creates a new 'Score', which is located within the 'Project Workshop' tree and under the Table; of which, the model is executed.
If the user selects the new 'Score' node, that is created beneath the Table, the (Predictive Score) data is posted. The 'Sample' column shows the 'Scores';l which identify the liklihood that a customer will be 'High Margin' and 'High Revenue'.
Since the Predictive Model score is added to the Table, if new customers are then added to the table, the score is then updated. If the user goes back to the 'Best Customers' toolbar. The 'Score' is added to the dataset, so that each time the .advm file is opened and/or when the nightly job runs, the new data is included in the model. The user can also then use this new score field to graph by, etc. A user can also use that formula, within the expression, to automatically flag to a chart, when a new 'High Margin' 'High Revenue' potential customers are identified.
