The Real Reason AI Fails at Scale: A Practical Framework for Achieving Agentic AI ROI

AI should be the unlock for scale, efficiency, and better decision-making. Yet inside most enterprises, AI doesn’t break through. Pilots stall. Value evaporates. Teams lose confidence.

In Episode 5 of Prompting Potential, Sarah Karthigan — VP of AI at Weatherford and former AI leader at ExxonMobil and Chevron — explains why AI fails inside large organizations and how leaders can finally convert ambition into measurable Agentic AI ROI.

Her message is blunt: AI doesn’t fail because the models are weak. It fails because enterprises deploy AI into environments that were never designed for intelligent systems.

AI Inherits What’s Broken

Many companies attempt to leapfrog into AI while their foundations are still fragmented. Data is inconsistent. Processes are disconnected. Core systems are legacy. AI ends up inheriting the same constraints humans struggle with.

As Sarah puts it, no company has “perfect data,” and waiting to fix everything first means you will never start. The answer is to tackle data and process issues in the context of real use cases, not in isolation.

Her approach: improve the foundation as you build — not before.

Why Most AI Programs Stall

Across decades in energy, Sarah has seen the same three failure patterns repeat:

1. Weak data foundations paired with unrealistic expectations

   Leaders assume AI can overcome structural gaps. It can’t. AI amplifies both strengths and weaknesses.

2. Building AI in silos without real user involvement

   The fastest way to kill adoption is to build a “perfect” solution without the operators who will use it.

   If end users aren’t involved from day one, it never becomes their solution — so it never scales.

3. Governance that slows the organization instead of enabling it

   Governance isn’t friction — it’s an accelerator when done right. Clear guardrails build trust, and trust accelerates deployment.

A Framework for Real AI Value: ROE, ROI, ROF

Sarah uses Gartner’s three-tranche value model to explain where AI creates measurable returns:

• Return on Employee (ROE) — productivity gains, everyday AI, time saved.

• Return on Investment (ROI) — operational gains, revenue improvements, reduced cycle time.

• Return on Future (ROF) — new business models and disruptive bets.

Most enterprises fail because they only invest in the first bucket — productivity — and never build the operational conditions or governance to reach the ROI and ROF levels.

The Path to Agentic AI ROI

The lesson is clear: AI succeeds when leaders redesign the environment around it.

That means:

• Fixing the parts of the foundation that matter for each use case.

• Involving operators early so they co-create solutions.

• Establishing governance that enables speed with oversight.

• Shifting from “AI as a tool” to Agentic AI — intelligent agents that orchestrate people, decisions, and workflows.

When those elements align, AI stops being an experiment and becomes a system that produces real return.

Agentic AI ROI isn’t a mystery. It’s the outcome of intentional design.

Watch the Full Episode

See the full conversation with Sarah Karthigan, VP of AI at Weatherford, in Episode 5 of Prompting Potential. In this discussion, she breaks down why AI succeeds — or stalls — inside complex enterprises, and what leaders must redesign to achieve meaningful Agentic AI ROI.