chapter six

6 Architect the AI technology mix

 

This chapter covers

  • Choosing between custom and out-of-the-box AI solutions
  • Selecting the right data model for performance
  • Assessing the tech architecture with data flow mapping
  • Bringing clarity in AI regulation
  • Communicating the proposed solution effectively through the Approach Statement

Having identified the right solution from a business perspective, we now turn to finding the best-fit technology to make it work. AI technology is a mix of in house custom-built and externally sourced components for storage, modelling, analytics, visualization and other functionalities in on-premise or cloud infrastructure that needs to form a well-functioning ecosystem.

We begin with the build-versus-buy dilemma, or how to find the right balance between custom and out-of-the-box solutions. We will complement the classical criteria of strategy, capabilities, economics, and risk with AI-specific principles.

We then move to selecting the data model that best fits our problem and AI branch. We examine relational, vector, graph, column-oriented, and unstructured databases to maximize performance.

All technological choices must be compatible between them and with the existing enterprise architecture. To make sure this happens, we will map data flows as a way to follow the data from capture to impact and assess whether the proposed infrastructure is appropriate for the project and the enterprise.

6.1 Scaling expertise in a world of bespoke

6.1.1 Proposal generation for bespoke products

6.1.2 Augmenting experts rather than automating them away

6.2 Architecting the AI technology mix

6.2.1 Should we build or buy the solution?

6.2.2 Which data model should be used to maximize performance?

6.2.3 Is the current architecture ready to support the envisioned solution?

6.2.4 Are there regulatory or ethical risks that require special attention?

6.3 Build or buy?

6.3.1 Tried-and-true build-or-buy criteria across fields

6.3.2 How AI changes the build-or-buy question

6.3.3 Making the right choice for your opportunity

6.4 Choosing the right data model for your AI use case

6.5 Map the data flow to define architecture requirements

6.5.1 Why architecture matters in AI

6.5.2 What do we mean by architecture?

6.5.3 A practical way to align on architectural needs

6.6 Including AI regulation in AI solution design

6.6.1 A brief introduction to AI regulation

6.6.2 A practical way forward during the AI Road Test

6.7 Choosing the technology mix at Teralvon

6.8 Aligning on the Approach Statement

6.8.1 What makes a good AI approach statement

6.8.2 AI Approach Statement template

6.8.3 The approach statement at Teralvon