RFI / RFP example for selecting a Document Agent 

1. Presentation Generation Capabilities 

Primary stakeholder: Business users 

1.1 The solution must support creating new slides in existing PowerPoint presentations based on natural language prompts entered by users. 

1.2 The solution must support generating complete multi-slide presentations from a single user prompt, with slide-by-slide structure planned by the agent.

1.3 The solution must support editing existing slides, including rewriting or rephrasing text in specific shapes designated by the user. 

1.4 The solution must support styling slides (colors, fonts, text sizes) to conform to centrally managed brand guidelines. 

1.5 The solution must support generating slide content that is contextually consistent with the rest of the presentation by reading existing slide titles and contents. 

1.6 The solution must support updating slide notes (speaker notes) based on user instructions. 

1.7 The solution must support native PowerPoint: generated slides should be based on slide lay-outs, supporting easy manual creation of new and editing of existing slides.  

2. Agent Intelligence and Automation 

Primary stakeholders: AI/ML teams, Product architecture, Innovation teams

2.1 The system shall allow configuration of system prompts and agent instructions to control behavior, constraints, and communication style. 

2.2 The system shall support configurable specialist or sub-agent workflows for different presentation sections or document tasks where this improves output quality and repeatability. 

2.3 The system shall allow human-in-the-loop plan approval, enabling the agent to present a proposed plan and wait for explicit user approval or rejection before execution. 

2.4 The solution must support persistent conversational sessions where the agent maintains context from previous user messages and tool results. 

3. Design System and Brand Governance 

Primary stakeholders: Brand teams, Marketing operations, Document governance teams

3.1 The solution must support consuming centrally managed slide layout guides (e.g., from a template) as an input to layout selection for new slides. 

3.2 The system shall allow configuration of brand guidelines, including available slide masters and layouts, to be exposed to the agent via backend tools. 

3.3 The solution must support selecting slide master layouts based on user intent and content type, using AI or rules to choose the most relevant layout. 

3.4 The system shall ensure new slides created by the agent are based on approved templates and slide masters defined by the enterprise. 

3.5 The solution must support enforcing the use of brand-aligned fonts, colors, and spacing rules when generating or editing slides. 

3.6 The system shall allow updates to layout guides and brand metadata without requiring changes to the agent logic (i.e., via backend services or configuration). 

3.7 The solution must support evaluation of brand compliance for agent-generated content, including automated checks against brand guidelines. 

3.8 The solution must prioritize approved enterprise content, templates, and slide assets as the default starting point for generation, and only generate net-new content where governed reusable assets do not satisfy the user intent. 

3.9 The solution must support inserting approved slides, approved visuals, approved text components, and structured business data into generated presentations as part of the same workflow.

3.10 The solution must support role-based, profile-based, or jurisdiction-based distribution of templates, content, legal text, layouts, and presentation building blocks, including dynamic personalization based on enterprise identity or metadata. 

4. Platform and Application Integration 

Primary stakeholders: Enterprise architects, IT platform teams, Microsoft 365 administrators

4.1 The solution must integrate with Microsoft PowerPoint as a web add-in or equivalent, running within the PowerPoint client (desktop or web) task pane.

4.2 The solution must operate within Microsoft 365 enterprise environments without requiring end users to manage additional credentials beyond their corporate identity. 

4.3 The vendor must ensure that all interactions with PowerPoint respect Microsoft’s add-in performance and security guidelines. 

4.4 The solution must support coexistence with existing Templafy add-ins in the same PowerPoint environment without conflict. 

4.5 The system shall allow integration with a central content library (e.g., approved slides, templates) where the agent can reference or reuse existing content, subject to access controls. 

4.6 The vendor must provide APIs or adapters to connect the agent to additional enterprise data sources (e.g., CRM, DAM, knowledge bases) in future phases without fundamental redesign. 

4.7 The solution must ensure that all external data used for generation (layouts, brand rules, content) is retrieved from governed, centrally managed sources rather than user-side configuration. 

4.8 The solution consists of an API or MCP server for integration with other applications that want to trigger the creation of a presentation. The end-result of this flow should be comparable to the output of the user-driven flow.

5. User Interaction Model 

Primary stakeholders: End users, UX teams, Digital workplace teams

5.1 The solution must provide a chat-based interface embedded directly inside Microsoft PowerPoint for initiating and controlling AI-driven presentation edits. 

5.2 The system shall display the agent’s messages, reasoning summaries (where relevant), and status updates in an easy-to-understand format. 

5.3 The system shall present a clear plan for multi-step operations and provide Approve/Reject controls before making large changes. 

5.4 The solution must indicate the current phase of an operation (analyzing, planning, executing, verifying) so users understand progress. 

5.5 The solution must support session reset so users can start a fresh conversation and scope without reloading the add-in. 

5.6 The vendor must ensure accessibility best practices in the UI (e.g., keyboard navigation, screen reader compatibility) consistent with enterprise standards. 

5.7 The solution must provide clear and actionable error messages when operations fail or connections are lost, including instructions for recovery. 

5.8 The system shall support localization and translation of the UI text and messages where applicable. 

6. Administration and Platform Governance 

Primary stakeholders: IT administrators, platform owners, AI governance teams

6.1 The system shall allow administrators to control access to AI features (e.g., by user group, role, or license type). 

6.2 The solution must provide configuration surfaces for controlling which templates, brands, and layouts the agent may use. 

6.3 The system shall allow administrators to configure LLM model parameters and endpoints (e.g., model version, deployment name, temperature) within supported ranges. 

6.4 The solution must provide administrative visibility into AI usage metrics (e.g., number of runs, tokens consumed, tool calls). 

7. Security, Privacy, and Compliance 

Primary stakeholders: Security teams, risk management, compliance officers 

7.1 The solution must operate within a (multi-tenant) architecture that logically isolates data, sessions, and configuration between tenants. 

7.2 The system shall use existing enterprise authentication and authorization mechanisms (e.g., single sign-on) for all agent and add-in endpoints. 

7.3 The solution must ensure that screenshots and other transient assets are stored in secure, access-controlled storage (e.g., Azure Blob) with short-lived URLs and expiration policies. 

7.4 The vendor must support data residency and regional deployment options consistent with enterprise requirements for LLM and storage resources. 

7.5 The system shall ensure that only authorized users in a tenant can invoke AI presentation generation features for that tenant’s content. 

7.6 The solution must prevent cross-tenant data leakage in all AI prompts, tool calls, model contexts, and logs. 

7.7 The system shall support optional disabling or redaction of full conversation logging while still capturing aggregate telemetry. 

7.8 The vendor must adhere to enterprise security review and penetration testing requirements, and provide evidence of remediation of identified issues. 

7.9 The solution must ensure that customer data, prompts, and generated outputs are not used to train or fine-tune shared or public AI models. 

7.10 The solution must support deployment using customer-controlled LLM environments (e.g., Azure OpenAI), ensuring data remains within enterprise boundaries. 

7.11 The system shall ensure that the agent only accesses and uses content that the requesting user is authorized to access, respecting enterprise identity and permissions. 

7.12 The solution must implement controls to filter or block prompts and outputs that violate enterprise policies (e.g., sensitive data, inappropriate content). 

7.13 The vendor must provide clear documentation of where data is processed and stored, and support regional deployment options aligned with enterprise requirements. 

7.14 The solution must implement safeguards against prompt injection and malicious input, ensuring that external or embedded content cannot override system instructions or access controls. 

8. Architecture and Deployment 

Primary stakeholders: Enterprise architects, cloud platform teams 

8.1 The solution must support integration with customer’s own instance of Azure OpenAI. 

9. Performance and Scalability 

Primary stakeholders: Platform engineering teams, IT operations

9.1 The solution must support concurrent agent sessions without degradation of performance beyond agreed thresholds. 

9.2 The vendor must demonstrate horizontal scalability of backend components (API, agent orchestration, telemetry) under realistic enterprise loads. 

9.3 The solution must maintain responsiveness of the PowerPoint client UI during agent execution, including non-blocking update strategies for the canvas. 

9.4 The vendor must provide guidance and configuration options for handling large presentations (e.g., over 100 slides) without exceeding model context limits. 

9.5 The solution must support scaling of LLM calls for both the main agent and any sub-agents (e.g., layout selection) without mutual interference.