Client
Nordward
Tech stack
Google Cloud
Solution
Recommender System
Service
AI + Machine Learning
E-fish-ient Product Recommendations for Seafood Group, Nordward
Nordward Seafood Group AB is a group of four leading companies in Sweden, Denmark and Norway that provides the leading restaurants and stores across the Nordics with the best seafood products on the market. To become the leading service partner in sustainability, product development, and customer relations, Nordward wanted to leverage AI to provide customers with tailored product suggestions via their mobile app. Nordward chose Datatonic to deliver a fully-productionised ML Recommender System based on Vertex AI pipelines, leading to a significant increase in average purchase size and effective shortlist recommendations.
Our impact
- Increased the average purchase size significantly over manually-generated shortlists
- Developed an effective recommender system with approximately 65% of items ordered being products that were recommended by the shortlists
- Reduced sales team’s workload by automatically generating customer-level shortlists without any manual intervention
The challenge
Within its field, Norward’s ambition is to become the leading data-driven service partner in regard to sustainability, product development and customer relations. Nordward and all of its four companies have strong local foundations focusing on high customer satisfaction, knowledge, quality and service.
With this purpose, a long-term objective for Nordward is to further enhance the capabilities of its mobile app to provide more personalisation, specifically concerning improving the effectiveness of its shortlists.
As such, Nordward wanted Datatonic’s help to build and test a new data-driven shortlist capability and recommender system for its consumer-facing app on the iOS and Android platforms, and then put it into production. This is a machine learning model that will automatically select products to be recommended to each customer for Nordward’s app.
“We are so proud to be able to help our customers place orders and stay informed in as easy a way as possible through our App. But serving all our customers with quality takes time, and to focus our sales representatives’ time to help solve our customers’ challenges, we need AI to do customer-tailored real-time and unobtrusive product suggestions.” – Christian Ahlin, CIO & Group Head of IT , Nordward
The solution
In order to provide customised recommendations to customers through Nordward’s mobile app, Datatonic worked with Nordward to develop an effective recommender system. In three weeks, we:
- Segmented customers into meaningful groups with similar historic sales and app interaction behaviours
- Built a recommender system based on cutting-edge Machine Learning methods that produces outputs served via the Nordward mobile app
- Deployed the solution on top of a fully-productionised pipeline in Vertex AI
- Provided presentations of data analysis insights, the architecture of pipeline components and implementation details
- Identified, tested & reassessed features, based on the understanding of the data coupled with Datatonic’s broad experience delivering recommender systems
- Trained the model on a pre-processed dataset, evaluated performance and identified the best-performing model
- Used A/B testing to compare the performance of AI-generated recommendations to manually-recommended options, and to evaluate different shortlist configurations
“Working with Datatonic was a great experience, and I highly recommend them to any business looking to improve their sales and customer satisfaction through the use of AI.” – Christian Ahlin, CIO & Group Head of IT, Nordward
This solution enables Nordward to use AI-generated recommendations on its mobile application, increasing sales and the level of personalisation for its customers. With the AI-powered recommender system in place, Nordward can serve more relevant, personalised recommendations to its customers. Since the solution has been deployed, Nordward has benefited from a 6.3% increase in purchase size, and this will likely improve as the ML model learns.
