Clusters
Unlike views, clusters are groups of feedback that you define. You can easily train clusters to automatically find similar feedback for you.
How to create a cluster
There's two main ways to create a cluster:
From within a space, create a cluster from the top of the page.
Or hover over any highlight or trend and click save.
You can choose to create a new cluster or add it to an existing one.
Name your cluster and choose where you would like it saved.
If you've just created a space, you will see this cluster the sidebar.
How training works
Highlights added to a cluster can be used to train it – to find more like it. Each piece of feedback marked as training works like a seed. And when you add multiple seeds, Sauce AI will triangulate highlights based on what's common between each.
As you start to build you cluster with a few pieces of feedback, Sauce will start recommending similar feedback for you to review — helping automate manual work. Each recommended feedback is presented with a similarity score. E.g. Below we see the first recommendation with a similarity score of 81%. This is a similarity score is based on that highlight's relevancy to the cluster as a whole.
As you build and add to your cluster, toggle between viewing all highlights or just to view training to better understand and control how it behaves. Both tabs present similarity scores to the cluster as a whole.
Adding feedback
There's three primary methods to add feedback to a cluster. 1) Adding similar feedback recommended to you based on training. 2) Using search and filters to add feedback you're looking for. 3) Feedback that is automatically added.
1) Adding similar feedback recommended to you
Click the + Add button and drawer will open listing out all recommended feedback organised by similarity score.
2) Using search and filters
Click the + Add button and drawer. Click on the second tab "Search all feedback" and you can search feedback just the same as anywhere else in Sauce AI. Using all the same search mechanics.
It's important to note that each method will come with logic as to whether that feedback is also added as training. Use the below table as a reference for this logic.
| Added feedback... | Is feedback added to training? |
|---|---|
When creating a new cluster from a trend or an individual highlight. | Yes |
When adding similar feedback found (80%, 70% ,60%) | Yes |
When adding highly related feedback | No |
When adding highlights from elsewhere around the App, to an existing cluster. | No |
When the cluster automatically adds feedback based on training | No |
This logic makes it easy to add to clusters while training silently operates in the background. However, you can always control what is added to training and what is not. To do this, hover over feedback in a cluster and choose to remove it entirely, or just remove it from training.
3) Feedback that is automatically added.
Clusters are on auto-pilot, reviewing each new piece of customer feedback as it enters Sauce. Any highlight that has a similarity score of 70% or higher to a training seed is automatically added. Anything below that is presented for you to review as a recommendation.
And this rule applies across all training seeds. Meaning they all work collectively to automate your workflow.
Avoiding feedback
Customers often use different terminology when referring to features. If you're building a cluster about "Templates" but seeing feedback about "Rich text editors" hover over that feedback, open the training options and click "Avoid feedback like this".
This will add a training seed that avoids any feedback like this being suggested.
Let's breakdown the example above. We now know, if a highlight is close enough to one of the positive seeds (I.e. 70%> similarity) it is automatically added. However, if it is closer to one of the negative seeds than any positive one, it is rejected. For example:
If a highlight scores 90% with a positive, and 82% with a negative it will still be added
But if it scores 90% with a negative, and only 82% with a positive, it will be rejected.
Last updated
