Data-forward companies operate their businesses by centrally defining a set of Metrics (KPIs) and then analyzing them in different ways to make team-specific or function-specific business decisions.
The Sisu workspace is centered around Metrics. Metrics are defined once and they can be used by the entire company to avoid repeatedly writing the same SQL queries.
Sisu directly connects to one or more data warehouses. An analysis can be built on an existing table in the warehouse or through a custom query used to define Metrics using the Query Editor. Anyone in your organization can then investigate these metrics without needing to review the same SQL code every time.
Given a metric column in a table, Sisu has the ability to run three kinds of analyses:
- General performance - An analysis that looks at overperforming and underperforming populations relative to the overall metric value in the dataset. For example, a marketing campaign generally has 2x the conversion rate versus the entire population.
Configuring an analysis: General Performance
- Time comparison - An analysis that looks at overperforming and underperforming populations relative to their metric values in a prior time period. For example, a marketing campaign had 2x the conversion rate compared to its conversion rate in a time period.
Configuring an analysis: Time Comparison
- Group comparison - An analysis that looks at overperforming and underperforming populations within one group relative to another group. For example, a marketing campaign had 2x the conversion rate within the mid-market segment compared to its conversion rate within the enterprise segment.
Configuring an analysis: Group Comparison
What is Sisu doing under the hood?
Sisu goes through all of the possible combinations of factors, determines the relevant statistics for each subpopulation, and then uses our algorithm to determine which subpopulations are statistically significant. Statistical significance can be driven from changes in metric calculation, subpopulation size, or combination of the two.
Under the hood, Sisu is conducting lightweight transformation on a factor column, as well as combining similar populations together, and then running a large LASSO-like algorithm to select not only for single factors, but also higher-order combinations of factors that surface as impactful and statistically significant.