Digital transformation RFP: the AI UX assessment question set the existing 60-question playbook does not cover

The 60-question agentic AI RFP playbook is the publication’s reference for the procurement-team evaluation of an agentic AI vendor; it has accumulated 38 Microsoft Copilot grounding citations against agentmodeai content in the three-month window ending 25 May 2026, anchoring it as the buying-committee’s RFP starting point.

The 60 questions there cover governance (regulatory posture, audit rights, breach disclosure), technical depth (architecture, observability, scaling), procurement (pricing model, term, exit, renewal), and audit (SOC 2, ISO 27001, sector-specific certifications). The dimension the existing playbook treats only at the workflow-design level is the user-interaction surface. The Microsoft Copilot grounding query “digital transformation RFP AI UX assessment questions” (3 citations against agentmodeai content) surfaces the gap directly; the buying committee for a digital-transformation RFP that includes agentic AI is asking the user-interaction question more directly because the agent is the new UI primitive the customer’s users will interact with.

This piece extends the existing 60-question playbook with 15 UX assessment questions in five categories. The buying committee uses these alongside the existing playbook; the structural argument is the UX dimension becomes substantive when the agent replaces or augments a substantial portion of the existing user interface, and the year-one user-adoption rate depends materially on the design quality the vendor’s RFP response signals.

Category one, interaction modes (3 questions)

The interaction-mode questions resolve the buyer’s question of how the customer’s users will actually experience the agent in their daily workflow.

Question 1. What interaction modes does the agent support at GA: voice, text chat, embedded inline-suggestion (the Copilot-in-Word style), autonomous-action (the agent acting without user prompting against pre-approved policies), or others? What is the modality-switching support: can a user start in text, escalate to voice, and continue with the same context, or is each mode a separate session?

Question 2. What user-agency primitives does the agent expose: the user’s ability to pause an in-flight agent action, reverse a completed action (within a defined window), redirect a misinterpreted query, or veto an autonomous action before it executes? How are these primitives surfaced in the UI: visible controls, voice commands, keyboard shortcuts, or post-hoc audit-trail review only?

Question 3. What is the agent’s handoff model to a human, and what is the latency commitment? The options range from synchronous live handoff with full conversation context transfer (the highest-maturity tier; the human picks up the conversation where the agent left off with no replay required) to asynchronous escalation with ticket context (the conversation is summarised and queued for human review) to none-of-the-above (the agent terminates and the user starts over with a human). The latency commitment per handoff path is the customer-side question that the vendor’s RFP response should answer in measured seconds, not in qualitative language.

Category two, error-handling and recovery (4 questions)

The error-handling questions resolve the buyer’s question of what happens when the agent gets it wrong, and how the user experience recovers.

Question 4. How does the agent handle tool-use failures when a downstream tool the agent depends on is unavailable, timing out, or returning an error? What is the user-facing experience: does the agent surface the failure transparently, attempt a workaround silently, escalate to a human, or fail silently with the user not knowing the agent’s output is based on incomplete information?

Question 5. How does the agent handle ambiguous queries where the user’s input is interpretable in multiple ways? What is the disambiguation surface: does the agent ask the user to clarify, guess-and-confirm (proceed with the most likely interpretation but verify), or proceed silently with the most likely interpretation?

Question 6. How is the agent’s confidence in its output communicated to the user? The options are numerical confidence score (the high-maturity tier; the agent explicitly states “I’m 70% confident in this answer”), qualitative hedging language (the agent uses softening phrases like “I think” or “based on what I found”), visual cue (a UI element conveys uncertainty without explicit language), or none-of-the-above. What is the customer’s ability to configure the confidence-communication policy across user populations (showing numerical confidence to technical users, qualitative hedging to general users)?

Question 7. What is the audit-trail of agent decisions the user can review? The conversation history is the minimum tier (the user can see what the agent said and what the user said in response). The decision-explanation surface is the maturity tier above (the user can see why the agent chose a specific tool, why it returned a specific answer, what alternatives it considered). The customer’s evaluation should ask the vendor to demonstrate the decision-explanation surface on a real conversation rather than describe it.

Category three, accessibility (4 questions)

The accessibility questions resolve the buyer’s compliance and inclusion obligations.

Question 8. What is the agent’s WCAG 2.2 conformance level? The vendor should produce a Voluntary Product Accessibility Template (VPAT) showing the agent’s conformance to WCAG 2.2 Level AA at minimum; the buying committee’s accessibility office reviews the VPAT against the customer’s specific workflow.

Question 9. What is the agent’s Section 508 conformance for federal-civilian buyers, and what is the EN 301 549 conformance for EU buyers? Both standards extend WCAG 2.2 with sector-specific requirements; the VPAT should document conformance at the GA tier (not the developer-tier).

Question 10. What is the agent’s assistive-technology compatibility specifically with the customer’s standard tooling? The customer should specify the tooling (JAWS, NVDA, Voice Control, Switch Control, eye-tracking systems, alternative-input devices) and ask the vendor to demonstrate compatibility rather than assert it. The procurement-mature pattern is a live demonstration in the customer’s standard accessibility-testing environment.

Question 11. What is the agent’s keyboard navigation and screen-reader experience for the agent’s primary interaction surface? Specifically, the agent’s response can be navigated with keyboard alone (not just mouse), the agent’s UI elements have appropriate ARIA labels, the agent’s dynamic content updates announce to screen readers, and the agent’s modal dialogs and overlays are accessible.

Category four, continuity (2 questions)

The continuity questions resolve the multi-user and cross-device experience.

Question 12. How does the agent’s state and conversation-history transfer across devices and sessions? The user who started a conversation on the desktop and continues on mobile should be able to see the conversation history; the user who returns to the agent after a multi-day gap should be able to see the prior conversation. The cross-device experience commitment is what the customer should ask the vendor to demonstrate, not describe.

Question 13. What is the agent’s user-handoff model for sequential workflows where one user completes a step and the next user picks up the workflow with the agent’s context preserved? The procurement use case is the customer-service-agent-to-human-supervisor pattern, the engineer-to-engineer pattern in code review, the sales-rep-to-sales-engineer pattern in deal escalation. The agent’s handling of the handoff should preserve context without exposing private information across the handoff boundary.

Category five, design-system extensibility (2 questions)

The design-system questions resolve the customer’s UX team’s ability to integrate the agent into the customer’s existing brand and interaction patterns.

Question 14. What customisation depth does the vendor support against the customer’s design system? The options range from CSS theming only (the agent’s existing components are themed but not replaced) to component-level customisation (the customer can override specific components with custom React or Web Component implementations) to full component-replacement (the customer can build a completely custom interaction surface against the vendor’s API). The customisation depth determines whether the agent feels like a vendor product embedded in the customer’s environment or a native customer-built surface.

Question 15. What API surfaces does the vendor expose for the customer’s UX team to integrate the agent into existing UI patterns? The minimum is a public REST API for the agent’s invocation surface plus a webhook surface for the agent’s outbound events. The maturity tier above is a React or Web Component library that the customer can compose into existing UIs. The maturity tier above that is a design-system contribution where the vendor publishes Figma libraries, design tokens, and interaction patterns that the customer’s design team can extend directly.

How the buying committee uses the 15 questions

The 15 questions extend the existing 60 questions in the RFP playbook; the buying committee adds them as a separate section in the RFP document. The vendor’s response to each question is evaluated by a different reviewer than the technical-feature reviewer: the customer’s UX team, accessibility office, and design system stewards review the UX section.

The procurement-mature pattern is to ask the vendor to demonstrate (not describe) the answers to questions 4-7 (error-handling), 10 (accessibility), 12-13 (continuity), and 14-15 (extensibility). The demonstration produces a structurally different evaluation output than the description; vendors with strong narratives can describe well, but the demonstration surfaces the gaps between the narrative and the product.

The output of the 15-question exercise feeds into the MSA negotiation. The customer’s UX, accessibility, and design-system requirements become specific contractual obligations: the vendor commits to maintaining WCAG 2.2 Level AA conformance at GA, the vendor commits to the customer’s specified assistive-technology compatibility, the vendor commits to the design-system customisation depth the customer’s UX team requires. The AM-167 NHI procurement clause work describes the contract-instrument layer.

What this means for the Q3 2026 digital-transformation RFP procurement

The 2026 digital-transformation RFP that includes agentic AI procurement uses the existing 60 questions for the governance, technical, procurement, and audit dimensions, and adds the 15 UX questions for the interaction-design dimension. The 75-question total is materially larger than the typical SaaS-procurement question set; the larger size reflects the agentic AI’s role as a new UI primitive in the customer’s environment.

The buying committee that runs the full 75-question exercise produces a procurement that is defensible against the year-one user-adoption review (when the customer’s user population evaluates whether the agent is actually usable in their daily workflow) and the year-three accessibility-audit cycle (when the regulator or the internal compliance function asks for the WCAG-and-assistive-technology evidence). The procurement that runs only the original 60 produces a vendor selection that may pass the technical evaluation and fail the user-adoption review; the procurement that runs only the 15 UX questions produces a vendor selection that may pass the UX evaluation and fail the audit.

The sibling 60-question playbook is the primary RFP reference this piece extends. The AM-167 NHI procurement clause work covers the contract-instrument layer. The AM-181 strategic-narrative proof points cover the vendor-diligence work that runs in parallel with the RFP question set. Together the four describe the 2026 digital-transformation RFP discipline that the year-three audit and renewal will reward.