Power Automate: Multimodal AI Prompts

Key insights

• Multimodal AI prompts — A May 13, 2025 demo shows how to build prompts in AI Builder that accept text, images and structured data, then run them inside Power Automate flows to automate tasks like expense processing.
• Multimodal input — Combine email text, receipt images and Dataverse records to extract structured data and improve AI understanding for real business scenarios.
• Power Automate integration — Embed prompts directly into flows to trigger automated actions, route approvals, and populate systems without manual steps.
• Power Fx & Copilot — Use Power Fx expressions to shape or calculate inputs and let Copilot help generate and refine prompt text and output format for consistent results.
• Reusability & version control — Create reusable “prompt fragments” and track prompt versions to speed development, roll back changes, and maintain quality across solutions.
• How to deploy — Open AI Builder, design a multimodal prompt, add Power Fx logic, connect to Dataverse for context, then deploy the prompt into a Power Automate flow to start automated processing.

Video summary

The Microsoft presentation, delivered by Gomolemo Mohapi during the 13 May 2025 Microsoft 365 & Power Platform weekly call, demonstrates how to build multimodal AI prompts and put them into production. The short demo focuses on extracting structured data from emails and receipts and then feeding that output into automated workflows. Consequently, the session highlights an end-to-end solution for automating expense reporting that combines several low-code tools. Importantly, the talk shows how these components work together in a practical business scenario.

How the demo works

First, the demo uses AI Builder inside the Power Platform to create prompts that accept both text and images, which the presenter refers to as multimodal inputs. Then the example augments those prompts with Power Fx expressions to perform simple logic and data formatting before sending results into an automated flow. Additionally, Copilot assists the authoring process by suggesting prompt structure and output formats, which speeds up iteration. Finally, the demo grounds the prompts in Dataverse so that AI responses reference enterprise data and feed clean records into a Power Automate flow for downstream processing.

Benefits and tradeoffs

The combined approach delivers clear benefits: richer understanding from mixed inputs, fewer manual steps, and faster processing of routine tasks such as expense reporting. However, there are tradeoffs to consider because richer models require disciplined data design and careful prompt construction to avoid inconsistent results. While prompt fragments increase reusability and reduce duplication, they also demand governance and version control to prevent drift across scenarios. Therefore, organizations should weigh the time saved in automation against the ongoing effort needed to maintain prompt libraries and data quality.

Challenges and implementation considerations

One practical challenge is prompt engineering, since prompts must be precise to yield reliable outputs, especially when handling receipts and personal data. Furthermore, grounding AI in Dataverse improves relevance but increases the need for access controls, auditing, and compliance work. Another concern is model behavior: even with guidance, AI can produce unexpected results, so teams must design review and fallback steps inside Power Automate flows. Consequently, firms should plan for monitoring, human-in-the-loop checks, and clear error handling from the start.

Operational costs also matter because running multimodal processing at scale can increase compute and licensing needs. Additionally, integrating Copilot and advanced features can shorten development time yet create dependencies on evolving platform capabilities. Thus, teams should pilot solutions, measure performance, and adapt governance as the deployment scales. In short, careful planning reduces surprises when moving from proof of concept to production.

Business implications and next steps

For businesses, this demo points to tangible wins: faster expense reimbursement cycles, more consistent data entry, and reduced manual review effort. Yet, organizations must balance speed with control by starting small and iterating on prompt design while tracking accuracy and cost metrics. Moreover, collaboration between citizen developers and IT helps align low-code agility with enterprise governance. Ultimately, the demo shows a pragmatic path to embed AI into automation, provided teams accept the tradeoffs and build robust monitoring and maintenance practices.

Keywords

multimodal AI prompts Power Platform, Power Automate multimodal prompts, prompt engineering Power Platform, using AI in Power Automate flows, generative AI Power Platform prompts, multimodal prompt examples Power Automate, AI prompts for Power Platform developers, integrate multimodal AI with Power Automate