Case

Sharper with AI at Nyhléns Hugosons

Do you also think that AI and machine learning are only for big companies and tech giants in Silicon valley? Substorm, together with Nyhléns Hugosons, started an exciting project that proves AI and machine learning work excellently for small and medium-sized companies. Today, the vast majority of companies have data on their operations that can help improve business, but few have taken the step to use it strategically.

image

Funding from Vinnova

– We may not be the company that is at the forefront when it comes to AI. On the other hand, we see ourselves as one of the companies involved in driving development forward here in the north. When we got the chance to start our first AI project in collaboration with Substorm, with help of funding from Vinnova, we could not miss the opportunity, says Anna Ahlqvist, marketing manager at Nyhléns Hugosons.

Data exposes patterns and behaviors

Nyhléns Hugosons is a company with a strong drive, focusing on constant improvement and working more sustainably. For many years, the company has unknowingly accumulated a wealth of data linked to production and sales. This data became the starting point for the AI project.

-It is about being able to read patterns in the data and thus learn more about your own business. Very often it is complex connections that are not always so obvious to detect. Here, AI can help uncover a number of interesting insights and opportunities, says Niklas Karvonen, CTO at Substorm.

Optimizes production and minimizes food waste

And Nyhléns Hugosons sums up a successful AI project with many new lessons:

– A little fun is also that we have invested in a weather app that gives us the weather forecast on our customers/stores location a week ahead. In that way, we can make more accurate recommendations on which products and volumes the customer should buy: sunny and warm weather means higher demand on meat and sausages to put on the grill. Another plus is that the forecast work is facilitated and becomes more accurate. The better accuracy the sellers get, the greater the opportunity we have to optimize production and minimize food waste, and obtain a high level of customer satisfaction, concludes Johan Sjöblom, sales manager at Nyhléns Hugoson.

/ Facts: Data & tech /

Data:

  • Historical sales data (Nyhléns Hugosons)
  • External “open” data sources

Examples of models from the project:

  • Recommendations: Which products a particular store should sell successfully based on other stores sales
  • Forecasts: How much a certain store will sell of a certain item at a certain time
  • Inference: Which store is most likely to sell a possible surplus of an item

Technics:

  • Matrix factorizations
  • K-nearest neighbor
  • K-means Clustering
  • Long Short-Term Memory (LSTM)
  • Principle Component Analysis
  • Single Value Decomposition
  • Random Forest Regression