The ability of businesses to meet customers’ expectations and deliver products punctually is one of the most critical aspects of the buyers’ experience. An optimized supply chain will:

  • Intensify general satisfaction with the service.
  • Improve the brand´s reputation.
  • Drag new customers.
  • Speed up your cash flow.
  • Avoid breaches of contract.

For all those reasons, it is essential to have a well-optimized supply chain.

However, in some industries, the manufacturing time is quite long, making it challenging to maintain a consistent and fast delivery time. For example, in the case of the automotive sector, one of our clients, Seat, wanted a solution.

Data set

In SEAT’s case, every car has four different configuration levels: model group, trim, TMAIMG and equipment. Thus, counting every possible combination, there are 10,000 total possibilities for each car.

Each Seat car has 4 different levels of configuration

That number of combinations makes it really hard to pre-manufacture cars to shorten the delivery time. That is why Seat contacted us to develop an artificial intelligence project aimed at reducing the delivery time of their cars.

Data analysis

Our team studied how to predict the orders for the next three months based on Seat’s historical sales data (past three years). And we decided to build a software tool personalized for Seat that uses neural networks to predict sales.

Process of the Seat software

Before building the software, it was necessary to pre-process all the data to ensure the data set imported into the software had the best possible quality (remove unused items, remove certifications, update old TMAIMGs, identify and remove fleets, etc.). As a result, we managed to reduce the data set from 100,000 rows to 70,000 rows.

After preparing the data set, it was time to develop the predictive software. First, it applies neural networks with a self-configurable feed. Then, the software uses the Regularized Minkowski loss index and the Quasi-Newton training algorithm to optimize the prepared data set.

Later, the Apriori algorithm is used to optimally obtain the most frequent combinations. And finally, the prediction is validated in three different levels (Trim, TMAIMG, and Combination) by calculating the error and the accuracy.

The total mean accuracy when predicting sales is 88%, indicating a good performance for Seat’s predictive model in their order system.

Conclusions

Before applying artificial intelligence to its logistics, Seat only delivered 50% of its orders in less than one month.

By using Artelnics’ predictive model technology, Seat managed to reduce the delivery time of 50% of its orders to 2-3 weeks, making it the first company to guarantee the delivery time of a car in this timeframe. They called this solution Fast Lane, and granted them a tremendous competitive advantage over the other brands.