Customer description‍

A consultancy company specialized in data projects for enterprise customers. The team regularly builds predictive models (e.g. churn, demand, classification), but struggled to make these models reusable to multiple customers.

Challenge‍

Each customer order required customization and separate deployments, making scalability and maintenance problematic. There was a need for a standardized and manageable way of delivering models as a service to customers, including SLAs, logging, and updates.

Solution‍

A solution was developed that includes a Modelling-as-a-Service (MaaS) platform that exhibits AI models to customers as an API. Models are containerized, scaled automatically and provided with monitoring, version control and access control. Customers easily integrate via REST or SDK.

Approach

1. Inventory of models and use cases
   We selected the most used models and standardized the input/output structure by model type.
2. Containerization and API layer development
   Each model was packaged in a Docker container with REST interface, running on a scalable Kubernetes environment.
3. Set up access control and SLA monitoring
   API tokens, rate limiting, logging, and uptime monitoring were set up per customer and model instance.
4. Develop a self-service dashboard
   Customers got access to a portal where they can test models, monitor and consult documentation.

Results

• Within 2 months, 5 AI models were put live as a service
• Customers integrate models directly into their workflows
• Lower operational pressure on the development team
• New recurring revenue from AI services

Learnings‍

The step from a customized project to product-as-a-service gave the consultancy company scalability and reusability. Customers benefit from rapid implementation, the team from more focus, and the company from a more robust business model.

‍