A Framework for Choosing Winning AI Use Cases [Agent Readiness Part 3]
Overview
This is Part 3 of a three-part “Agent Readiness” series on the AI Daily Brief, hosted by Nathaniel (from Superintelligent) and featuring Nufar Gaspar, an AI strategy consultant associated with Superintelligent. The episode focuses on use case readiness: how to identify, select, manage, and track AI agent use cases within an enterprise. The series culminates here with the most practical and actionable instalment, offering frameworks for deciding where to invest in agentic AI and how to measure the results.
Source video URL: (not provided)
Prerequisites
- Basic familiarity with what AI agents are and how they differ from traditional automation and simple chatbots
- Understanding of general enterprise software adoption and change management concepts
- Familiarity with ROI measurement in a business context
- Awareness of prior episodes in this series:
- Part 1: Cultural dimensions of agent readiness (the “change framework”)
- Part 2: Data and technical readiness (data issues as the primary blocker)
Main Points
1. The Four-Phase Use Case Process: Identify → Select → Manage → Track
- Nufar frames the entire episode around four sequential steps for managing agent use cases.
- The goal is not just to find one good use case, but to build and continuously improve a portfolio of use cases.
- This process applies whether a company has zero agents in production or several.
2. Sourcing Use Case Ideas: Bottom-Up Is Better
- The worst source of use case ideas is top-down executive mandate (“we need to build an agent”).
- The best sources are mid-level and frontline employees, who know what is feasible and must be bought in to execute.
- An ideation sprint is a recommended technique: educate staff on what agents can and cannot do, collect a large inventory of ideas across the company, then apply a coarse prioritization filter.
- Aggressively eliminate any idea that could be solved with simpler automation or a decision tree.
3. Characteristics of Good (and Bad) Agent Use Cases
Good use case signals:
- Involves complex, highly variable decision-making (e.g., resolving customer issues that differ each time)
- Humans are bottlenecks due to limited professional capacity (e.g., legal contract review)
- Requires 24/7 response capability (employee or customer support)
- Benefits from high personalization at scale (e.g., tailored outreach with the right offer at the right time)
- Involves work employees want to offload (repetitive/tedious tasks), which builds momentum for more sensitive use cases
- Is measurable — agents are goal-driven, so if you cannot define and measure success, agents are inappropriate
- Has some tolerance for errors
Bad use case signals:
- The process can be described as a fixed decision tree with limited branches — use simpler automation instead
- Zero tolerance for errors (e.g., payroll calculation must be 100% correct — use deterministic automation)
- The process exists only in one person’s head and is not documented — agents require well-defined, documented processes
4. The Most Universal Agent Use Cases
Nufar identifies a short list of use cases that appear in nearly every enterprise audit:
| Use Case | Description |
|---|---|
| FAQ / Policy Bots | Internal or external agents answering complex, nuanced questions from rich information sources |
| Company Knowledge Retrieval | Enabling employees to access fragmented internal data across systems |
| Operational Workflow Automation | Eliminating drudge work at the team level (e.g., automated status reporting) |
| Market Watchers | Monitoring competition, regulation, and industry news continuously |
Additional common use cases surfaced in audits include: customer support, software engineering, marketing content generation, sales, contract and regulatory review, data cleaning, and industry-specific growth opportunities.
5. Managing Use Cases Like an Investment Portfolio
- Balance between efficiency-focused use cases (doing more with fewer resources / cost reduction) and growth-focused use cases (top-line revenue impact).
- Companies are typically over-indexed on efficiency; the largest value often lies in growth use cases.
- Nufar introduces a matrix adapted from the 1970s BCG Growth-Share Matrix, plotting use cases on complexity vs. value:
HIGH VALUE
|
[Moonshots] | [Focus Area]
High complexity, | High-ish value,
high value | moderate complexity
(centralized | (primary focus)
AI team needed) |
__________________|__________________
|
[Avoid] | [Low Hangers]
High complexity, | Low complexity,
low value | lower value
| (self-serve via
| agent platforms)
LOW VALUE- Low hangers: Great starting points; over time should be self-served via agent-building platforms rather than requiring a central AI team.
- Focus area: Where most resources should concentrate — sufficient value with achievable complexity.
- Moonshots: High risk, high reward; often correspond to radical process shifts or major growth opportunities; require a centralized, professional AI team.
- Also balance vertical vs. horizontal agents and build vs. buy decisions.
- Companies with low readiness scores should start with low hangers and focus areas before attempting moonshots.
6. ROI Measurement Framework
Nufar provides a concrete formula for measuring agent ROI:
Agent ROI = Return − Investment
Return = Benefits (usage × impact)
− Cost of errors and incidents
− Uncertainty discount (be conservative)
Investment = Build/buy cost
+ Ongoing usage cost (tokens, models, tools)
+ Maintenance cost (owner time, user time)
+ Sufficient buffers (investment is routinely underestimated)Key principles:
- Measure continuously, not just at pilot — production behaviour with real users frequently differs from pilot results.
- Be conservative when making judgment calls on benefits; do not ignore the cost of agent errors (e.g., refunds triggered by mistakes).
- Account for cost variability: model costs are falling, but more sophisticated agents consume more tokens, often balancing out.
- 2026 is expected to be a “show me the money” year where ROI becomes a central organisational question.
7. Industry Context: ROI Timelines Are Accelerating
- A KPMG annual CEO study (surveying ~1,300 CEOs of $500M+ companies) showed a dramatic shift:
- 2024: ~65% expected 3–5 years to realise AI ROI; ~20% said 1–3 years.
- 2025: ~67–69% now expect ROI in 1–3 years; ~19% expect it within 6–12 months.
- Morgan Stanley has publicly stated their AI investment costs have been recovered — they are ROI positive.
- Traditional ROI frameworks are poorly suited to agentic AI; enterprise leaders are actively seeking new measurement heuristics.
8. Efficiency vs. Growth: Why Both Are Required
- Efficiency gains (e.g., 50% more marketing content) will become table stakes — competitors will achieve the same, so no differentiation results.
- A professional services firm was told by its largest client to deliver the same work for 50% of the price in 2026 — illustrating that efficiency is a survival floor, not a competitive advantage.
- Growth use cases — particularly long-tail opportunities (upsells, edge cases, previously uneconomical niches) — represent the largest hidden value that agents unlock.
- The correct posture is to do both, using the portfolio approach to make “do everything” manageable.
Key Concepts
- Agent Readiness (Use Case dimension): Whether a company has identified, selected, and is executing on a sufficient set of well-chosen agent opportunities.
- Ideation Sprint: A structured workshop where employees are educated about agents, then collectively generate a large inventory of use case ideas for subsequent prioritisation.
- Low Hangers: Low-complexity, lower-value agent use cases that are quick wins; best suited to self-serve agent platforms over time.
- Focus Area: The primary zone of investment — moderate-to-high value use cases with achievable complexity.
- Moonshots: High-complexity, high-value use cases with significant risk and reward; require a centralised AI team and substantial investment.
- Efficiency Use Cases: Agent applications aimed at reducing cost or resource usage (doing more with less).
- Growth Use Cases: Agent applications aimed at increasing revenue or top-line impact; often underexplored relative to their value.
- Agent ROI Formula: A structured calculation of return (benefits minus error costs and uncertainty) minus total investment (build, usage, maintenance), with conservative buffers.
- BCG Growth-Share Matrix (adapted): A classic 2×2 strategy tool repurposed here to plot agent use cases by complexity vs. value to guide portfolio balance.
- Long-Tail Use Cases: Previously uneconomical opportunities (e.g., micro-upsells) that become viable at scale through agents.
- Vertical vs. Horizontal Agents: Vertical agents serve a specific industry or function; horizontal agents serve broad, cross-functional needs.
- Build vs. Buy: The decision of whether to develop an agent internally or procure an existing solution.
Summary
Nufar Gaspar concludes the three-part Agent Readiness series by providing a practical, four-step framework — Identify, Select, Manage, Track — for systematically choosing and governing AI agent use cases. She argues that the best use cases emerge bottom-up from employees, not top-down mandates, and that good candidates involve complex variable decisions, human bottlenecks, 24/7 demands, or high personalisation needs, while cases requiring zero error tolerance or undocumented processes should be avoided or handled with simpler automation. She recommends managing the resulting use case inventory like an investment portfolio, balancing low-hanging wins against high-value focus areas and long-term moonshots, and ensuring an equilibrium between efficiency and growth use cases — the latter being where the largest and most frequently overlooked value resides. Finally, she stresses that rigorous, conservative, and continuous ROI measurement is non-negotiable: as agentic AI matures and ROI timelines accelerate industry-wide, organisations that have built credible measurement systems will be positioned to adapt and compound their advantage, while those relying on intuition and usage metrics alone will struggle to justify or improve their investments.