Properly created dashboards provide the mechanism to drive effective management and resource allocation decisions (Y. Wind)
Managers often track metrics they believe can potentially predict performance outcomes and help them improve decisions. However, it is unclear how to best select such predictive metrics out of a wide range of candidate metrics.
At a snack food company, marketing managers complained about being overwhelmed by this large amount of candidate metrics, which are sold to them as “key performance indicators” and expressed strong interest in retaining a smaller set of metrics, which truly are key leading indicators of performance. This would both render the metrics more actionable and allow the company to reduce the cost of regularly (in our case weekly) collecting these metrics. The decision makers responded to this information overload by picking those metrics that showed them in a positive light (looking backwards). This created friction at meetings because different managers cherry-picked different metrics and had few ‘objective’ arguments for their favorites. This situation is common in cases across industries and continents provided in ‘It’s Not the Size of the Data, It’s How You Use It ‘.
We developed an analytical approach to metric selection in 5 steps
- Delete metrics that show too little or too much variation in univariate tests;
- Reveal leading performance indicators (LPIs) with pairwise tests;
- Quantify how much each LPI explains performance with different models;
- Select the best set of LPIs by assessing their predictive validity in a holdout sample;
- Use the selected set of metrics to perform what-if analyses for proposed courses of action.
In the case of the snack food company, the first 2 steps reduced 96 candidate metrics to just 17 leading performance indicators, while the third step brought it down to the 10 most important. The next steps allowed managers to show the metrics by performance impact versus time to work.
Our managers at the data provider agreed the final metrics were (1) not known to them (i.e. the analytical approach made a difference) but (2) had face validity (to the best of their knowledge). Focusing on the metrics managers felt they could influence, they performed several what-if analyses revealing a strong sales impact of
- within the same week: from promoting product usage as an afternoon lift;
- after 1 week: from increasing brand liking and the quality difference with the store brand;
- after 2 weeks: from promoting product usage to entertain friends and increasing brand satisfaction;
- after 3 weeks: from increasing unaided awareness for the brand.
The managers proceeded to brainstorm about marketing tactics to achieve these objectives, and managed to turn around declining brand sales and prove the usefulness of the analytic dashboard.
Dashboards are living things; you need to regularly re-evaluate the metrics; adding new ones suggested by your employees, and pruning those that do not prove useful in decision making. Check out Smarter Marketing with Dashboards and Analytics to find out how!