Operating AI

CHAPTER 4Deployment with AI Operations in Mind

Although AI models are developed and trained in the lab, it's important to remember that's not where they will generate any business value. It's not until you deploy your models in a production setting that the potential value of AI can fully be realized. This is true whether your models are more informally deployed as part of an internal process, or a vital part of an internal decision-making framework or even deployed for commercial purposes. However, moving your models from the lab to production is far from an easy task. Successful model deployment is about a lot more than just running your model in another execution environment.

When deploying AI models in production, you need to consider various areas, from legal rights and data access to managing retraining and redeployment of models in a live production setting. Remember that even when the model is deployed in an internal process, people and decisions will depend on it differently from when you run it in the lab. Depending on what type of solutions or operational settings will depend on your model, you also need to consider how to seamlessly deploy a fallback solution if, for example, the model performance degrades to unacceptable levels, or if the model stops working entirely. This chapter will explain how to be more successful in operating AI by adopting an operational approach in the deployment phase.

Companies in the software industry currently deploying AI are reporting ...