[ With hundreds of billions invested globally on AI pilots, the failure rate for scaling custom enterprise AI tools remains remarkably high. This Pilot-to-Production Chasm of struggling to translate initial excitement into scalable, transformative business value has created what is now known as the "GenAI Divide".

To explore the issues and potential solutions to crossing this AI technology chasm, we are happy to bring you this detailed and insightful guest article from a leading AI pioneer working on applications for the financial services industry and Institute member, Nathan Stevenson, CEO of ForwardLane – an AI-powered, purpose-built analytics and insight automation platform for financial professionals in asset management, wealth management, and insurance enterprises.

His analysis reviews: The Anatomy of Failure: Core reasons Why Pilots Stall; The Agentic Bridge: Architectural Solutions; Strategic Imperatives for Crossing the Chasm; and concludes that the future of business intelligence will be defined by the emergence of the Agentic Economy - a new economic paradigm and fundamental architectural shift toward Agentic AI where AI agents become active, autonomous participants that inherently address the core technological and organizational barriers that cause traditional pilots to stall.]

 

Bridging the Pilot-to-Production Chasm in Enterprise AI: The Agentic Imperative Imperative 

by Nathan Stevenson, CEO ForwardLane

The rapid evolution of Artificial Intelligence (AI), particularly Generative AI (GenAI), has spurred unprecedented enthusiasm and investment across enterprise sectors. Yet, despite billions of dollars pouring into pilots - an estimated $30–40 billion globally - most organizations find themselves stranded on the wrong side of the Pilot-to-Production Chasm, struggling to translate initial excitement into scalable, transformative business value. This struggle has created what is now known as the "GenAI Divide".

The harsh reality is that the failure rate for scaling custom enterprise AI tools remains remarkably high, hovering around 95%. While generic LLM chatbots like ChatGPT and Microsoft Copilot boast high adoption rates (around 83% pilot-to-implementation rate), they often fall short in complex, mission-critical workflows because they lack the persistent memory and customization required for enterprise-grade tasks. For large enterprises (firms exceeding $100 million in annual revenue), the time lag from pilot to full implementation is often nine months or longer, starkly contrasting with the average 90-day timelines reported by top-performing mid-market companies.

This article argues that the solution to bridging this chasm is not merely incremental improvement but a fundamental architectural shift toward Agentic AI. Agents are autonomous, goal-driven systems that inherently address the core technological and organizational barriers that cause traditional pilots to stall.

The Anatomy of Failure: Why Pilots Stall

The consensus among industry leaders and researchers is that the core reason pilots fail to scale is not primarily infrastructure, talent, or even regulation, but a fundamental learning gap.

1. The Learning Gap: Static Tools in a Dynamic World Most early GenAI systems are inherently static; they are designed without the ability to retain feedback, adapt to changing context, or iteratively improve over time. This limitation proves fatal for mission-critical enterprise work, particularly in regulated industries like financial services, which rely on persistent knowledge and contextual awareness.

Users accustomed to the flexibility of consumer LLMs quickly become frustrated with internal enterprise systems described as brittle or overengineered. As one corporate lawyer noted, a system that "doesn't retain knowledge of client preferences or learn from previous edits... repeats the same mistakes and requires extensive context input for each session". This inability to learn means that for complex, multi-week projects, human judgment still dominates AI preference by a 9-to-1 margin.

2. Poor Workflow Integration and Customization Successful transformation requires AI to be embedded into existing workflows, not merely "bolted on" as a separate tool. Many vendor-pitched solutions fail at scale because they are misaligned with actual operational procedures and lack deep integration with core internal systems such as Customer Relationship Management (CRM) or Loan Origination Systems (LOS). Enterprises demand systems that minimize disruption to their current technology stack.

3. The Governance and Risk Hurdle The transition from a proof-of-concept (PoC) environment to a fully deployed production system requires rigorous governance, transparency, and accountability—requirements that are non-negotiable in regulated sectors. The inherent autonomy of next-generation AI agents introduces greater governance challenges related to accountability and liability. Gartner forecasts that over 40% of Agentic AI projects will be discontinued by the end of 2027 due to rising expenses, vague business benefits, or insufficient risk management. Addressing this requires moving beyond reactive compliance to establishing robust Human-in-the-Loop (HITL) oversight and clear guardrails from the outset.

The Agentic Bridge: Architectural Solutions

The emerging class of Agentic AI systems provides the architectural solution to the learning gap that defines the GenAI Divide. Agentic AI represents a "paradigm shift" that goes beyond traditional automation, enabling systems that autonomously plan, reason, and execute tasks toward defined objectives.

Jensen Huang, CEO of NVIDIA, suggested that Agentic AI is just starting to revolutionize the way companies work internally. This transformation is driven by core capabilities that directly counter the weaknesses of static GenAI tools:

• Autonomy and Proactivity: Unlike reactive, prompt-based AI, agentic systems are proactive, capable of taking initiative and executing actions without constant human intervention. They interpret objectives, understand context, and orchestrate complex, multi-step tasks autonomously.
• Persistent Memory and Learning: Agentic AI is designed with persistent memory and iterative learning capabilities, allowing it to adapt to user feedback, outcomes, and dynamic contexts. This continuous evolution is essential for high-stakes enterprise tasks.
• End-to-End Workflow Orchestration: Agentic AI’s true potential lies in coordinating complex, end-to-end workflows across multiple systems and tools. In banking, this multi-agent network could perform tasks like continuous KYC maintenance, where one agent pulls data, another scores risk, and a third files regulatory updates—all without human handoffs.

Leaders must recognize that agents exist on a spectrum: from Taskers (simple automation) and Collaborators (working interactively with humans, maintaining long-term memory) to highly sophisticated Orchestrators (coordinating multiple agents and tools at scale). The orchestration layer, which utilizes standardized frameworks for secure and explainable agent interactions, such as the Model Context Protocol (MCP) and Agent-to-Agent (A2A) standards, is critical for building a scalable Agentic Web.

Strategic Imperatives for Crossing the Chasm

Successfully crossing the chasm requires executives to adopt a new operating model focused on integration, accountability, and adaptive technology.

1. Embrace a “Buy-and-Embed” Strategy Internal AI development efforts fail twice as often as projects executed through strategic external partnerships (33% vs. 66% success rate in one sample). Successful buyers recognize this pattern and choose to partner with vendors who offer deep customization and fast time-to-value. They treat AI vendors less like software sellers and more like business service providers, holding them accountable to operational outcomes, not just model benchmarks. For example, firms like ForwardLane emphasize an integration-first architecture using protocols like MCP, Postman, Zapier, and Merge.dev to quickly onboard new data and extend platform functionality.

2. Prioritize High-Impact, Low-Risk Use Cases to Build Trust To gain early momentum and executive buy-in, enterprises should start by focusing on use cases that enhance advisor productivity (often referred to as co-pilots or internal assistants) before deploying AI directly to clients.

• Augmenting Front-Office Professionals: Financial firms like Morgan Stanley, Merrill (BofA), and UBS have deployed GenAI tools that automate administrative tasks such as meeting transcription, CRM note-taking, and drafting follow-up emails, saving advisors up to two hours per week in administrative time. Merrill’s pilots alone showed a 30% reduction in call-prep time.
• Automating Back-Office Functions: Agentic AI offers dramatic and sustainable returns in back-office automation, often yielding better ROI than front-office marketing efforts. Examples include automating loan pre-qualification, verifying identity/income, and orchestrating credit memo generation, moving from a manual, multi-step process to an orchestrated agent network.
• Deep Investment Research: Startups like Rogo and Hebbia leverage AI agents to speed up investment research, drafting reports, and performing due diligence, saving private equity firms 20 to 30 hours per deal.

3. Modernize the Technology Foundation for Agentic Readiness Agentic AI requires robust and scalable infrastructure, often leveraging cloud-based solutions. Organizations must:

• Ensure Data Quality and Accessibility: A high-quality, accessible, and unbiased data foundation is essential. Agents rely heavily on access to unstructured data (documents, voice recordings, transcripts) for contextual reasoning, requiring organizations to invest in ingestion pipelines.
• Embrace Interoperability Standards: Agent-to-agent communication must move beyond custom integrations. Protocols like MCP (Model Context Protocol) enable agents to dynamically access and inject relevant data from external functionalities into their reasoning processes via standardized communication layers. ForwardLane’s Hypebase utilizes its graph-based Knowledge Store as an auditable repository where every insight is traceable to source data, addressing critical trust and compliance concerns.
• Shift Engineering Paradigm: Software development needs to evolve to manage the full lifecycle of AI agents, focusing on how they are tested, monitored, and safely deployed as they learn and adapt over time.

4. Build Compliance and Trust into the Core For regulated industries, governance must be proactive, not an afterthought.

• Human-in-the-Loop (HITL) Oversight: Agentic AI should complement human expertise, not replace it entirely, especially for high-risk or sensitive decisions. Banks must maintain humans at key decision points—referred to as "Human-on-the-Loop" for monitoring and supervision—to ensure accountability and risk reduction.
• Explainability and Auditability: Transparency is critical to customer trust. Agentic systems must provide detailed audit trails and provenance logs that link every output or decision back to its source data.
• Mitigate Bias and Ensure Fairness: Agents rely on historical data, which may contain biases leading to discriminatory decisions in areas like lending. Organizations must implement bias-detection strategies and ethical AI practices, ensuring systems align with principles of fairness and inclusivity.

Conclusion: The Race to the Agentic Web

The Pilot-to-Production Chasm is a wake-up call, indicating that traditional enterprise AI strategies are insufficient for the current wave of autonomous intelligence. The future of business intelligence will be defined by the emergence of the Agentic Economy, a new economic paradigm where AI agents become active, autonomous participants in complex financial ecosystems.

The transition is moving from simple GenAI tools to deeply embedded, learning-capable, custom systems. This shift decentralizes action, moving from human prompts to autonomous protocol-driven coordination enabled by frameworks like A2A and MCP. Organizations that successfully cross the GenAI Divide are those that empower frontline managers, partner strategically, and choose adaptive, memory-rich agentic solutions.

The window for crossing the GenAI Divide and establishing a dominant position in the Agentic Web is rapidly narrowing. Leaders must act decisively now—rethinking architecture, codifying knowledge, and investing in governance—or risk falling permanently behind competitors who will soon leverage agents that operate 24/7, converting knowledge into coordinated action across the entire enterprise value chain. The future of competition will be determined not by who experiments the most, but by who successfully deploys and scales intelligent autonomy.

 

The Institute for Innovation Development is an educational and business development catalyst for growth-oriented financial advisors and financial services firms determined to lead their businesses in an operating environment of accelerating business and cultural change. We operate as a business innovation platform and educational resource with FinTech and Financial Services firm members to openly share their unique perspectives and activities. This interview is for informational purposes. The goal is to build awareness and stimulate open thought leadership discussions on new or evolving industry approaches and thinking to facilitate next-generation growth, differentiation, and unique client/community engagement strategies. The Institute was launched with the support and foresight of our founding sponsors — Ultimus Fund Solutions, FLX Networks, TIFIN, Advisorpedia, Pershing, Fidelity, Voya Financial, and Charter Financial Publishing (publisher of Financial Advisor and Private Wealth magazines).

Institute for Innovation Development www.innovationdevelopment.org @innovationIID   IID©2025