Two new ways to answer why, faster: Text and segment analysis

By Berit Hoffmann - July 7, 2020

At Sisu, we accelerate and supercharge the most time-consuming parts of an analyst’s workflow. Today, I’m thrilled to announce two new ways for analysts to diagnose more, faster: tools to deliver far more comprehensive segment analysis and faster queries for deep text analysis.

Together, these new capabilities make it easier to turn all the rich context you’re capturing in your data warehouse into clear, actionable business decisions.

Go from the what to the why with segmentation analysis and group comparisons

First, we’re taking experiment and segment analysis to the next level, unlocking new levels of detail for group comparisons. Current testing methodologies are effective when the question is, “Did my test outperform the control?” But even with a positive result, the inevitable follow up is, “Why did the test group do better?”

Instead of manually digging through hundreds of hypotheses to try to answer why, Sisu uncovers it automatically. You define two groups based on any criteria in your data, and Sisu returns the specific subpopulations or factors that performed better or worse for one group versus the other.

As an example, let’s look at audience engagement for a media streaming company. In this scenario, the team is investigating a new “auto-skip” feature for ads to see if automatically skipping them improves session duration. Typical segment analysis can provide incomplete answers because even if you do discover one group has a longer session duration, you don’t know why or whether it was uniform across all segments. What if you could not only understand how much a group outperforms but comprehensively determine why you’re seeing differences in behavior – is it a small effect across everyone in the population, or a significant impact on just a few?

In the example below, the overall segment analysis was inconclusive: viewers who experienced auto-skip showed a marginal decrease in session duration. But, looking deeper, Sisu shows that specific subgroups did act very differently. In one case, lower-income households who watched The Handmaid’s Tale spent almost 10% more time watching during the test. That’s a compelling result, found in seconds. Now the team can confidently target “auto-skip” for only the groups that it helps.

Group comparison example - The Handmaid's Tale


Beyond these standard segment analysis use cases, there are dozens of applications for detailed group comparisons. Media companies can quickly diagnose audience changes between the first and second episodes in a new show. Game publishers can precisely target high-interest players with fine-tuned offers. And retailers can finally understand customer preferences across hundreds of potential segments.

Lightning-fast categorization and contextual text analysis

Second, you can now use Sisu to quickly incorporate rich text analysis into your toolkit. Traditionally, teams spend countless hours converting unstructured text into a normalized and structured format. They often end up only looking at a small subset of the data or excluding key context in favor of simplicity. This can lead to faulty or incomplete conclusions from the analysis.

Sisu’s new text analysis capabilities allow you to aggregate all your text data into a single field and automate the process of finding common terms, categories, and correlations. Looking to understand which products are purchased together? Let Sisu find the connection. Trying to diagnose which sales and marketing activities combine to improve conversion rates? Sisu can find the most effective combinations.

Here’s an example: an eCommerce company with hundreds of different product SKUs doesn’t want to create a unique column for every product in an order. Instead, we recommend aggregating all that rich data into a single, set-valued field, and Sisu automatically parses the text for analysis. This enables you to quickly and easily uncover whether certain product names or combinations are impacting key metrics, such as AOV or order count.


The speed and comprehensiveness of Sisu’s automated text analysis cannot be matched using traditional visualization and dashboard tools. As an analyst, you don’t have the time to (re-)create new views, custom queries, and fresh reports every time stock changes. With Sisu, you can now automate these critical inquiries, and stop spending hours designing queries to normalize, structure, and parse this kind of data. For a few tricks on building datasets for this feature, check out Charles’s related post here.

Whether you’re testing new product features, rolling out new user acquisition campaigns, or simply trying to diagnose customer feedback, we’d love to show you how Sisu can accelerate your segment analysis and expand your understanding of why your metrics are changing.


Read more

Proactive Analytics: Root-Cause Analysis vs. Outlier Detection

A look at the differences between root-cause analysis and outlier detection and how these tools can break the cycle of reactive analytics caused by dashboards and legacy BI tools.

Read more

SQL Tips and Tricks: Using LISTAGG for session data

If you collect all your sessions data and no one is around to use it, is it really collected at all? A guide to using LISTAGG in SQL.

Read more