Traditionally, the contribution of analytics experts to business decisions has often been delivered in the form of static outputs: tables, charts, reports, and some explanation of how the results should be read. While that can create valuable insight, it also creates a bottleneck. As soon as new questions emerge, analysts need to step in again to run additional analyses, interpret the output, and translate it back into decision language.
Recent advances in AI do not only expand the range of analytical methods. They also make it much easier to build more interactive and decision-oriented forms of communication around data. Instead of handing over study results only as Excel files or slide decks, it is increasingly possible to provide lightweight applications that allow decision-makers and stakeholders to explore findings more dynamically, compare scenarios, and move more directly from evidence to action.
This app is intended as an encouragement to further explore this idea. At the same time, it serves as a pragmatic prototype for a typical driver analysis workflow. The reference use case is a common brand tracking setup in which a KPI such as brand strength is measured alongside ratings of different brand attributes. The goal is to understand which attributes are most strongly linked to the KPI and where action is most likely to matter. While this is the main example behind the app, the overall structure is generic enough to support many other use cases with a similar objective.
This app supports a structured workflow to move from survey data to action priorities. It helps you identify which variables are most strongly linked to a selected target KPI, understand how topics relate to each other, and compare intervention scenarios before deciding where to act.
Methodically, the app moves from description to explanation and then to prioritization: from how the KPI and its potential drivers are currently rated, to how variables are statistically related, to which levers are most action-relevant under realistic business constraints.
If you don't have a suitable data set at hand, press the button to generate random data
The first step should be to look at how the relevant target group perceives the KPI and its drivers. Which ones are perceived well or poorly? Are there signs of polarization?
This section shows how respondents distribute their ratings across each variable and where the average score currently stands. It helps a decision-maker quickly see which topics are already strong, which are weak, and which may be polarized. Example: if a variable has many top-box ratings and a high mean, it is likely a current strength; if it has many low ratings, it is a clearer area for improvement.
To prioritize fields of action, you need to understand how strongly the different drivers are connected to the target KPI. But it is also important to understand how the drivers are connected to one another. Which drivers reinforce each other? Which drivers might counteract each other? Which drivers might have stronger indirect relationships with the KPI despite showing no or only weak direct relationships?
This section shows how variables move together in the data and helps distinguish between broad co-movement, more unique relationships, and larger thematic clusters. You find the following perspectives:
This heatmap shows simple pairwise relationships between variables. Red means two variables tend to go up together, blue means one tends to go down when the other tends to go up. Example: if the target KPI and a driver have a strong positive correlation, respondents with higher driver ratings also tend to give higher target ratings.
This network turns the strongest simple relationships into a map. Each node is a variable and each line is a stronger pairwise link. Example: if a variable sits close to the target and is connected by a thick green line, it is strongly linked to the target at the simple relationship level.
This heatmap shows more unique relationships after statistically accounting for the overlap with other variables. It helps separate broad association from more distinctive links. Example: a driver may correlate with the target overall, but show only a weak unique relationship once similar drivers are taken into account.
This network focuses on the more unique links that remain after overlap between variables is reduced. Example: if a connection remains visible here, it is more likely to reflect a distinctive relationship rather than one that is only carried by several similar variables.
This view groups variables into broader themes that tend to be rated similarly. It helps simplify a longer questionnaire into a smaller number of dimensions and also understand the latent topics that lie underneath respondents' perceptions. If several service-related items load highly (the loading indicates to what degree a variable correlates with the factor) on the same factor, they may represent one common service experience dimension. One simple heuristic when deciding on the appropriate number of factors is to look at the so called 'Eigenvalue' of a factor. If the 'Eigenvalue' is > 1 then it explains more than one single variable. To optimize for this criterion choose a facor solution where all factors haven an 'Eigenvalue' of > 1. You see them added to the x-axis labels in bracktes (EV: ...).
The drivers with the strongest direct connections to the target KPI are not necessarily the most relevant or effective areas for improvement. This also depends on the current performance of those drivers (i.e. how they are perceived). Furthermore, changes in different drivers come at different costs, which should also be taken into consideration.
This section combines driver impact with current performance to support action setting. It helps identify where improvement is most likely to matter, simulate expected KPI shifts, add implementation assumptions, and compare saved scenarios.
Use the plot by combining driver importance with current performance. The most useful reading logic is:
This section estimates how direct changes in non-target drivers may shift the selected target KPI in a ridge-stabilized regression model. The control panel on the left lets you adjust all non-target variables, while the results on the right summarize the expected target movement and the contribution of the changed drivers.
The model is designed to be more stable when drivers are correlated with each other. It is still a decision-support tool and should be read as an expected directional effect within the observed data structure, not as proof of causality.
The network views on the 'Driver Connections' tab remain descriptive. They help you understand how variables are interconnected, while the simulator here translates selected driver changes directly into an expected target movement.
Use + and - to change any non-target variable in 0.1 steps. Saved scenarios store the selected driver changes and the resulting expected delta in the target KPI.
This section lets you add basic implementation assumptions for each non-target driver. Use it to capture how expensive improvement is expected to be and how much organizational friction is likely to occur when acting on the driver.
This chart visualizes all saved scenarios by simulated impact on the target variable and total cost. Bubble color reflects the average organizational friction across the drivers changed in each scenario.
Save a local snapshot of the current app state including the active data, selected target variable, saved scenarios, simulator state and intervention inputs.