You've just deployed new or edited computations. This means their definition is now stored in the library of computations. 

Now, to use them in a Customer Fit model, there is one more step to do before you can visualize it or use it in a draft model. This step is called "reloading your computations".

Why? A draft model is based on a dataset, loaded at some point in time. If the dataset was loaded before new computations were deployed, then this dataset doesn't contain the new computation. Hence the need for reloading the computations calculated for the dataset. 

When to reload computations

There are several cases:

• You updated a computation marked “Live” on the computation page: The changes are immediately taken into account in your Live customer fit model, but not in draft customer fit models.

• You updated a computation not marked “Live” on the computation page: The computation is not used in your Live Customer fit model, so the changes do not affect it. If the computation is used in a draft model, you need to “Reload computations” for each model you want to use this computation in. Read below for the How-To.

• You created a new computation: The new computation is not used in your Live Customer fit model yet, so this does not affect it. If you want to use the new computation in a draft model, you need to “Reload computations” for each model you want to use this computation in. Read below for the How-To.

How to reload computations

To reload your computation(s), go to the Overview section and click on Reload computations so that the datasets can be enriched with the new computation(s) you created.

Once the dataset is enriched and loaded, you should receive a confirmation email.

What reloading computations does

Reloading computations does 3 things: 

1. includes any new computation in the model where you reloaded computations

2. updates any computation whose definition has been updated 

3. refreshes any existing computations

For example, on Dec 2021 you loaded a dataset, and the enrichment (computations) is as follows:

A year later, you

• create a new computation called is competitor,

• change the definition of employees combined to prioritize Salesforce enrichment over HG Insights enrichment

• and you reload the computations

The dataset enrichment therefore gets updated with the new computation and the new values of the existing computation "employees combined", according to its new definition. 

The computation "hiring position" also changes as a company frequently opens and closes job opening and our provider PredictLeads sends us monthly fresh data. 

Impact of reloading computations on a model

How does reloading computation impact a model? 

1. You can now use new computations in the model: visualize the computation insights and use the computations for overrides, signals, tree nodes...
   To visualize this computation on the training dataset for a model, head to Insights and search for the label of your computation to check out the result (learn more on How to read the Customer Fit insights). 

2. The distribution of scores in the model can change. When deploying your model the scores in your CRM can change even if you have not changed the model.  Confused? Hang on you'll see

   •  Since the enrichment can change, the decision tree values can change and therefore the distribution of scores can change.

   • Let's take an extreme example
     Let's say you have a split in your decision tree separating companies with less than 500 employees. On Sept 2021, there were more companies with employees combined <= 500 than with > 500  

A year later, you update the computation employees combined to prioritize your Salesforce enrichment over HG Insights enrichment, which has higher company size estimations than HG Insights enrichment. Therefore some companies that were in node 2 move to node 3 and now node 3 converts better than node 2.  

and now larger companies get scored higher than smaller companies. 

By just changing and reloading the computation, now your model has changed without you touching the configuration of the model itself. Therefore when you'll deploy the model the scores in your CRM may change.