This site uses different types of cookies, including analytics and functional cookies (its own and from other sites). To change your cookie settings or find out more, click here. If you continue browsing our website, you accept these cookies.
Announcing Alteryx + Snowflake | Alteryx and Snowflake make analytics and data science fundamentally easier. With the new integrated starter kit, you can push down data prep transformations and more into Snowflake for faster data quality and analytics output. Learn More
Gradus, an international consulting firm based in Brazil, used Alteryx Designer and Server to build a model that predicts whether settling or fully litigating lawsuits will lead to a more favorable outcome. Their client, a multinational manufacturer of home appliances who wanted to automate this predictive process, estimates saving almost BRL 1 million/year (U$ 200,000/year), or 3% of its current liability expenses, as a result of the model’s increased accuracy. Also, they are now better able to address the root issues giving rise to such lawsuits: problems with its warranty terms, poor technical service, and recurring service orders.
Describe the business challenge or problem you needed to solve
Our client, a multinational manufacturer of home appliances, has a long-term relationship with its customers. In this relationship, they incur some expenses to support the post-sale process and ensure customer satisfaction. One of the biggest expenses they face is legal liability arising from customer complaints about product defects, service problems, or warranty issues. Those cases can have either a positive outcome, where the company is not accountable for the problems, or a negative one, where they are considered guilty and must pay the customer damages. In the latter case, it is in the interest of the company to offer a pre-settlement agreement, which represents, on average, half the cost of a judgement, considering both legal fees and the cost of the judgement itself. On the other hand, in comparison with a positive outcome, an agreement usually corresponds to about 3.5 times the cost of fully litigating.
For deciding when to propose an agreement and when to proceed with litigation, the company had established a manual classification process where a third-party followed a heuristic based on criteria empirically defined by the quality intelligence team to decide which cases were worth settling. Beyond the cost to pay for this service, the process itself was not sufficiently accurate, as only 51% of the cases classified as worth litigating actually had a positive outcome, representing a huge opportunity cost (considering that a settlement could have been offered). Moreover, it is possible that some settlements had been unnecessarily offered in cases where they could have won and could have saved about 85% of the settlement value. Also, our client’s heuristics had not changed for a long time and were potentially out of date.
Conceptual illustration of the predictive modelling process to predict the outcome of liability processes
Describe your working solution
In the context described above, we identified an opportunity to implement a predictive model that classifies the cases as either worth settling or worth litigating. With this model, we expected to reduce expenses in two different dimensions: the need for third-party personnel to do the job manually, and the average cost per case by increasing the accuracy of the classification.
To do so, our first step was to clean the historical data available from the previous five years. There were dozens of categorical and numerical variables that could potentially be used to predict the outcome of the legal process. All the data preparation was done with Alteryx Designer, identifying missing information, removing outliers, and ensuring standardization among fields. The workflow was built in a way that new cases could easily be merged into the database with no set-up time.
After that, we divided the data into training, validation and test sets using Alteryx Sample and Oversample tools. Then, we tested different predictive models present on the R-Predictive Suite in Alteryx Designer to try to find the best solution and ended up by using a logistics regression followed by a stepwise tool to choose the best predicting variables. Our choice was based both on the quality of the model results and its interpretability. The latter is crucial because identifying the main causes of incurring liability enables the quality intelligence team to mitigate future occurrences by addressing the root causes.
Final prototype of the workflow used to validate the predictive model
The model was also uploaded to Alteryx Server, where the quality intelligence team can easily access the model and its results. Also, we are able to schedule the model to both run on a desired frequency to classify new cases and to recalibrate on a regular basis by running the workflow that trains it with fresh data, capturing the most recent trends.
Model training on Alteryx Server environment. It can be scheduled to run on a specific frequency to keep the model up to date with the most recent tendencies
Describe the benefits you have achieved
The first impact of the model on our client’s daily activities was the reduction of the number of third-party personnel needed to run the process, as legal process classification was no longer done manually. During the transition, part of the team is being kept in order to validate the model in production, by comparing the results from the predictive model with the heuristic assessment. Once the model is fully validated, the algorithm will enable the reduction of most of the workforce allocated to that activity, since the process will be fully automated.
Another impact is the reduction in the average cost per case. The savings is thanks to the greater accuracy achieved by the model in comparison to the former heuristics. The model is more accurate in two ways: by reducing both false positives and false negatives. That is, the model catches cases that the heuristics would have misclassified as worth litigating and suggests that a settlement should be offered, representing a cost reduction of 85% on those cases. The model also catches cases that the heuristics would have misclassified as unwinnable and suggests they are worth litigating, representing a savings of up to 70% on those cases.
With the first version of the model already rolled out, the company estimates savings of almost BRL 1 million/year (U$ 200,000/year), or 3% of the current legal liability expenses as a result of the increased accuracy. Also, our client expects to save even more by addressing the main causes giving rise to liability: problems with the warranty terms, poor technical service, and recurring service orders, all of which were identified thanks to the predictive model.