Design and evaluation of AI copilots -- case studies of retail copilot templates

TL;DR

This paper presents a systematic approach to designing and evaluating AI copilots, illustrated through a retail case study, emphasizing architecture, testing, and safety to ensure effective human-AI collaboration.

Contribution

It introduces a comprehensive framework for AI copilot design and evaluation, highlighting key technical components and principled testing methods with practical retail case insights.

Findings

Key architectural components of a copilot are identified.

Testing and evaluation improve AI safety and effectiveness.

Case study demonstrates practical application in retail.

Abstract

Building a successful AI copilot requires a systematic approach. This paper is divided into two sections, covering the design and evaluation of a copilot respectively. A case study of developing copilot templates for the retail domain by Microsoft is used to illustrate the role and importance of each aspect. The first section explores the key technical components of a copilot's architecture, including the LLM, plugins for knowledge retrieval and actions, orchestration, system prompts, and responsible AI guardrails. The second section discusses testing and evaluation as a principled way to promote desired outcomes and manage unintended consequences when using AI in a business context. We discuss how to measure and improve its quality and safety, through the lens of an end-to-end human-AI decision loop framework. By providing insights into the anatomy of a copilot and the critical aspects of testing and evaluation, this paper provides concrete evidence of how good design and evaluation practices are essential for building effective, human-centered AI assistants.