Creating a synthetic score

A synthetic score is a score Thematic works out from comment text, for something your customers care about but were never asked to rate, such as cleanliness, ease of onboarding, or value for money. You describe what you want to measure in plain language, check Thematic's scoring on a sample of comments, and save. Thematic then scores every comment in the dataset.

Synthetic scores work on individual survey datasets.

You need permission to manage the dataset. If you can't see the Synthetic scores tab, contact your organization's Thematic admin or your Thematic CSM.

Open the Synthetic scores tab

Open Manage Dataset for the dataset you want to score, then select Synthetic scores. The tab lists any scores already on the dataset.

The Synthetic scores tab in Manage Dataset, listing four scores and the Add a synthetic score button

Select Add a synthetic score and choose how to start:

  • Custom starts from a blank description that you write.
  • Synthetic NPS is a ready-made starting point for how likely a customer is to recommend you. It comes with a description and four categories already filled in: Promoter, Passive, Detractor, and Not relevant. You can edit all of it.

If you see the message You cannot add another synthetic score, your account has used all of its included scores. See Managing synthetic scores.

Step 1: Define score

  • Write one to three sentences describing what you want to measure. Be specific about the topic and what good and bad look like, for example: How satisfied customers are with trolley availability and parking at our stores.
  • Choose the column that contains the comments to score.
  • Set the score range. The default is 0 to 100.

Thematic uses your description to draft the score categories.

Step 2: Categories

A category is one possible result for a comment. Each has a name, a short description, and a score. Review the drafted categories and change anything that doesn't fit:

  • Use the edit and delete icons to change or remove a category, and Add Category to add one.
  • Leave a category's score empty to make it No score. Use this for comments that say nothing about what you're measuring, or that are too mixed to call, so they don't affect the average.

Under Apply score to, choose what each comment is scored on:

  • Whole comment suits overall measures, such as NPS.
  • Relevant sentence(s) suits specific topics, such as seat comfort, where only part of a comment is about the topic.

The Categories step, showing four score categories including two set to No score, and the Apply score to options

Step 3: Review accuracy

Thematic scores a sample of 15 comments. For each one, select Correct or Incorrect. When you select Incorrect, choose the right category and, if you like, explain why. The explanation helps Thematic learn what it missed.

The Review accuracy step, with one comment marked Incorrect and the correct category being chosen

The accuracy figure at the top adds up across every comment you review. When you've finished a batch, choose one of:

  • Improve instructions & re-test uses your corrections to improve the scoring, then scores the same comments again.
  • Score next 15 comments checks a fresh batch with the current scoring.
  • Continue moves on to save.

Repeat until you're happy with the accuracy. Reviewing more comments and re-testing is free and doesn't count towards your account's synthetic scores.

You never need to write scoring instructions yourself. Thematic writes them behind the scenes from your description, categories, and corrections.

Step 4: Save

Confirm the score's name and save. Thematic applies the score to every comment in the dataset. This runs in the background and can take a while on a large dataset.

Using your score

Once it's saved, a synthetic score works like any other score. Select it from the score dropdown in Analysis, and use it in Dashboards and Answers. Combine it with your themes to see what's driving it, using impact and score change.

To change, improve, or delete a score later, see Managing synthetic scores.



Was this article helpful?

Your feedback helps us improve our documentation.