The 2021 collapse of ‘Zillow Offers’ serves as a stark reminder of what can go wrong when AI is deployed without the right safeguards. Built to automate home buying, Zillow’s algorithms frequently misjudged market conditions – causing the company to overpay for thousands of homes. With unsold inventory piling up, Zillow took nearly a billion dollars in losses, laid off 2,000 employees, and eventually shut down the program.

While this played out in the consumer space, similar risks are surfacing inside enterprise AI pilots across banking, pharma, and manufacturing where agentic models are being rushed into production without the governance. However, real-world adoption of Agentic AI presents a complex landscape. What looks promising in a lab can fall apart in production. The gap between PoC and scalable success remains wider than most leaders realize.

The AI adoption journey

According to Gartner’s 2024 survey, a striking 80% of Generative AI proof of concepts (PoCs) fail to make it to full-scale production. Most enterprises begin their AI journey by collaborating with existing technology vendors and engaging strategy consulting firms to craft a tailored AI roadmap. This typically leads to the identification of a few proof-of-concept (PoC) initiatives aimed at testing AI’s viability within specific business processes.

However, the initial excitement often gives way to sobering realities. One of the first hurdles is accuracy and efficacy. Off-the-shelf AI solutions, regardless of the underlying large language models (LLMs), frequently deliver only 75-80% accuracy when deployed. For AI to be truly useful in business-critical scenarios, enterprises must aim for 95% or higher accuracy. Achieving this requires addressing both data quality and algorithmic sophistication – a dual challenge that many organizations underestimate.

The data dilemma

A major barrier to effective AI implementation is the nature of enterprise data. Historically, data has been curated for human consumption – structured to support manual analysis and decision-making. AI, on the other hand, thrives on digitally accessible, high-quality data that can fuel autonomous decision-making. This mismatch creates what is known as the “Data for AI” problem. Enterprises must invest in digitalizing their processes, segmenting truly digital data, and building robust data pipelines. Ensuring data quality, traceability, and lineage is essential for developing trustworthy AI systems. Depending on an organization’s data maturity, this transformation can be both time-consuming and costly.Governance and responsibility

Another critical challenge is establishing a governance framework for autonomous AI agents. While “agentifying” business processes can empower human workers and enhance productivity, it must be done responsibly. Without proper oversight, AI decisions can lead to unintended consequences. This is where Responsible AI comes into play. Enterprises must implement guardrails that ensure AI decisions are observable, auditable, and aligned with ethical standards. Continuous monitoring by human decision-makers is essential to maintain control and accountability.

A real-world success story: AskChemille

A standout example of Agentic AI in action is AskChemille, developed by a leading chemical manufacturer to transform how prospects engage with complex polymer solutions. In just six months and with less than half of a full-time equivalent’s cost, the company deployed an AI-powered search-and-answer experience designed to achieve two primary goals: reduce a traditionally 12-month sales cycle that relied heavily on experienced chemistry SMEs, and promote a self-service portal for prospects seeking tailored, expert-level guidance.

Powered by a Small Language Model (SLM) trained to match the expertise of a PhD chemist, AskChemille can accurately answer technical questions and recommend customized solutions, delivering a 75% improvement in sales cycle time and achieving a full ROI within the first quarter of deployment. This showcases how Agentic AI can leapfrog traditional knowledge transfer processes and deliver immediate, high-impact business value by autonomously adapting to user needs and driving outcomes.

The ART of AI success

To navigate these challenges, enterprises should adopt the ART framework focusing on Accuracy, Responsible, and Trustworthiness for AI. These three pillars form the foundation of successful AI implementation. Ignoring any of them introduces significant risk and undermines the potential benefits of AI. To accelerate AI adoption and maximize its impact, organizations should consider the following strategic actions:

• Educate Leadership on AI’s True Potential
• Move beyond the hype around LLMs and chatbots.
• Emphasize AI’s role in scaling human capabilities and enhancing decision-making.
• Integrate machines as trusted collaborators in driving informed decisions
• Invest in Data Maturity
• Digitalize core processes and identify gaps in data readiness.
• Build infrastructure that supports high-quality, AI-ready data.
• Define AI Vision Through Business Outcomes
• Align AI initiatives with key growth, profitability, and operational metrics.
• Ensure AI efforts are outcome-driven, not technology-driven.
• Focus on interoperability and orchestration of agents.
• Articulate ART Goals
• Customize agentic experiences based on business needs.
• Define trust goals, talent requirements, and observability standards.
• Involve business leaders in AI governance through councils and cross-functional teams.

3 Steps to Productionizing AI

To move from PoC to production, enterprises should build on the ART framework and follow three essential steps:

Step 1: Pilot with purpose: Anchor your use case to a clearly defined business outcome.

Step 2: Pick the right tech stack: Choose a platform that supports agent interoperability and future scalability.

Step 3: Co-create for value: Partner with a service provider who can share ownership of value realization. (See sample process below.)

Agentic AI holds immense promises for enterprises, but its adoption requires a thoughtful, strategic approach. By focusing on accuracy, responsibility, and trustworthiness, organizations can unlock AI’s full potential while mitigating risks. The journey may be complex, but with the right vision and investments, enterprises can build intelligent systems that truly empower their workforce and drive sustainable growth.

The author is Uday Hegde, Co-founder & CEO, USEReady

Disclaimer: The views expressed are solely of the author and ETCIO does not necessarily subscribe to it. ETCIO shall not be responsible for any damage caused to any person/organization directly or indirectly.

Join the community of 2M+ industry professionals.

Subscribe to Newsletter to get latest insights & analysis in your inbox.

All about ETCIO industry right on your smartphone!