Enterprise HR had no shortage of Agentic AI ideas, but we needed a defensible way to decide what to build first and what to validate next. I led a prioritization study that aligned customer demand, trust constraints, and delivery realities into one decision model. The outcome changed the top 3 use cases we took forward into concept validation, and we backed the recommendation with execution proof across both employee and HR fulfiller workflows.

Role: Design Manager (study owner)
Type: Strategy, prioritization, and execution proof
Partners: UX Research (Collyn), Product and Design leadership (Sr Dir, VPs), Engineering leadership, Central Agentic AI team

1) What it looked like in product

Non-critical: resolve in the channel employees already use (Microsoft Teams)

For non-critical inquiries, the agent meets employees in Teams and guides them to resolution with policy grounding and clear choices.

What this flow proves

• Policy-grounded answer with a visible source

• Employee choice between waiting or escalating

• Closure loop once the path is confirmed

(Visuals: 3 screens)

• Policy answer + citation

• Wait vs escalate choice

• Close loop

Critical: assist the human HR agent in HR Agent Workspace

For critical cases, the AI does not replace the HR agent (human fulfiller). It reduces manual work by pre-routing, proposing a plan, and executing only after explicit human confirmation. Execution creates real workflow artifacts.

What this flow proves

• Background agents route and flag criticality before the human acts

• A step-by-step plan is proposed in the flow of work

• Execution creates tasks (and approvals) with an audit trail

(Visuals: 3 screens)

• Activity trace of agent routing and criticality

• Plan proposal + execute approval

• Tasks created with ownership and sequencing

Trust and control patterns

• Cited sources for answers and guidance

• Explicit human approval before execution

• Workflow-native outputs (tasks, approvals, activity log), not chat-only responses

2) Why this mattered

We were missing a shared decision model that could answer:

1. Which HRSD agentic use cases are worth investing in first

2. Why those are the right bets, with evidence

3. What would make a use case approval-ready early, not after months of work

This was less a research problem and more a product strategy decision problem that needed credible UX evidence and clear tradeoffs.

3) The decision problem

We had more plausible Agentic use cases than we could invest in at once, and each area had a different top priority. We also had central approval gates that could block work late if risk, data readiness, or feasibility were unclear.

This required a decision system that could survive leadership review, not a list of opinions.

4) What I owned

As the Design Manager, I owned the initiative end-to-end and partnered closely with UX Research.

• Planned the study structure, narrative, and checkpoints with leadership

• Co-shaped survey framing and interview approach with Research

• Participated in select interviews and synthesis to validate real workflows and constraints

• Consolidated signals into a prioritization model and sequencing recommendation

• Presented the readout to Product and Design leadership and aligned on next steps

• Ensured the output met expectations from Engineering leadership and the central Agentic AI team

5) Approach: two signals, one decision

Customer signal

A structured customer survey to capture relative demand across a set of Agentic AI use cases.

Operator and feasibility signal

Interviews and checkpoints with internal stakeholders and operators to pressure-test feasibility, governance risk, and what would be approval-ready.

Then we combined both streams into one prioritization view to avoid opinions versus opinions.

6) The prioritization model (how tradeoffs were made)

To prevent cool demos from winning, we evaluated each use case using consistent dimensions:

• Customer value and frequency (recognizable pain, repeated volume)

• Workflow leverage and scalability (reusable pattern vs one-off)

• Trust, risk, and compliance (what can go wrong, where human gates are required)

• Knowledge and data readiness (coverage, freshness, explainability)

• Feasibility and approval readiness (dependencies and gating constraints)

This framework turned debates into transparent tradeoffs and made sequencing defensible.

7) What changed (the outcome)

The prioritization output changed the top 3 use cases we advanced into concept validation. The final shortlist balanced customer demand, trust and governance requirements, knowledge readiness, and approval readiness.

Before (initial instinct):

• Performance Management Agent

• Succession Planning Agent

• Recruiting Sourcing Agent

After (evidence-backed prioritization):

• Agent Zero Auto-Resolution

• Offboarding Agent

• Onboarding Personalization Agent

We moved from strategic-sounding agents to trust-building fundamentals once customer demand, ROI clarity, and approval readiness signals made it clear that high-volume workflows would drive adoption first.