Organizations are struggling to convert enterprise AI adoption into measurable P&L impact. The issue often arises more from the AI architecture and operating model than the AI model itself. Durable ROI comes when enterprises shift from isolated AI projects to vertical AI engines built on solid data and business systems.

For technology leaders, the mandate is clear: design AI as an enterprise capability, not a collection of pilots. This means connecting AI across relevant business disciplines so each use case becomes faster to scale, easier to govern, and more likely to create business value.

Key takeaways:

• Horizontal AI is critical for engagement and early exploration, but vertical deployments often extract more business value.
• For AI initiatives, business should drive the “What?” and technology should focus on the “How?”
• Orchestration is important for AI organizations as well as agents.

Horizontal vs vertical AI: Turning adoption into business ROI

Essentially, horizontal AI is the application of general-purpose systems to broad classes of problems, while vertical AI uses purpose-built models trained on domain-specific data.

Horizontal AI is an important catalyst for enterprise experimentation. It gives employees immediate access to powerful capabilities and encourages grassroots innovation. However, leaders should not conflate broad usage with business value.

When every function independently builds its own AI workflow, the organization quickly accumulates waste in the form of duplicate efforts, inconsistent business rules, fragmented vendor engagements, and competing compute consumption.

The surface-level activity looks impressive, but the economics remain weak because each team is paying to rediscover similar patterns in isolation.

The next phase of AI maturity requires a deliberate shift from isolated deployments to vertical AI engines embedded in core business workflows. These engines should be built on governed data products, reusable domain logic, and standardized controls.

Furthermore, common business concepts such as customer, asset, and invoice should be modeled and accessible by AI agents. This is where AI ROI begins to compound: every new use case becomes faster to deliver, safer to scale, and more consistent in its outputs because it draws from the same trusted foundation.

For example, consider a Financial Planning and Analysis (FP&A) scenario. In a horizontal model, financial analysts may combine spreadsheets, forecast assumptions, and market commentary in a generic AI tool to predict corporate growth or risk exposure.

Other teams may produce separate views using sales, workforce, or operational inputs. The isolated insights are helpful for local use cases, but the enterprise still lacks an integrated, repeatable methodology.

In a vertical model, teams build on shared enterprise data. The AI connects and analyzes all the disparate inputs as a cohesive whole. It applies approved assumptions, highlights material gaps, explains key drivers, ties findings to source records, and recommends tradeoffs within the FP&A workflow.

Each meaningful output is integrated back into the data foundation for re-use. There is no single vertical architecture that works for all organizations. The challenge lies in developing a robust enterprise solution, synthesizing the right AI functions with critical processes to extract business value.

From bypass to breakthrough

The promise of low-code/no-code AI has encouraged many business teams to believe they can bypass IT entirely. That belief is understandable, but it is rarely sustainable.

A business user can configure a prompt, evaluate a workflow, and demonstrate a compelling prototype within days. The difficulty begins when that prototype must scale across multiple systems.

This is often where AI pilots hit the execution wall. The solution is to redefine the relationship of the business with IT as platform partners.

Business teams should own use-case ideation, business requirements, and outcome validation. IT should provide modular, reusable components that span multiple functional orgs such as secure APIs, approved data environments, cost governance, and integration patterns.

For example, a procurement team may want an AI agent to summarize supplier risk and recommend contract actions. If the team works independently, the agent may depend on exported spreadsheets, partial contract extracts, and manually gathered vendor information.

The prototype may be useful, but it will be difficult to trust, audit, or scale. With IT as a platform partner, the agent can securely draw from approved supplier and contract sources, apply consistent business rules, and route recommendations efficiently into the procurement process.

The business still owns the decision logic and user experience, while IT ensures the foundation is reliable, secure, and reusable. This creates faster innovation without sacrificing enterprise control.

Orchestrating AI for enterprise ROI

The next frontier of AI business value is orchestration. Enterprises already have the tools, models, and enthusiasm to begin. What currently separates leaders from laggards is the ability to orchestrate data, platforms, governance, workflows, and financial accountability into a robust, repeatable system for value creation.

Technology executives, CAIOs, and CDOs should expand their agendas beyond experimentation to scaling enterprise capabilities. This means:

• Identifying where AI functions repeat across domains
• Turning those patterns into reusable AI products
• Establishing collaborative platforms that allow business teams to innovate without wasting resources

For example, consider the development of a unified demand forecasting and customer engagement engine for a major retailer. The GM defines the business requirements, such as a 15% reduction in inventory costs paired with a 5% increase in personalized marketing conversions.

In support, the CDO integrates point-of-sale data, supply chain logistics, and customer loyalty histories into a single data product. The CTO then creates a small network of AI agents to predict localized inventory shortages and automatically launch targeted promotional campaigns to reduce overstock.

By standardizing this workflow, the trio enables local sales and marketing teams to rapidly and repeatedly deploy similar campaigns.

Moving forward

Over time, AI tools will evolve to solve more of these problems automatically. AI has already moved beyond rows and columns into the richer language of business: conversations, images, and activity signals.

Like an executive gaining broader context, AI technology becomes more useful when it can see the whole business landscape.

That said, enterprises cannot wait for the technology curve to solve the operating challenge. AI will become more powerful, but it will not soon remove the need for disciplined enterprise design and governance.

Today’s call to action is to build the operating model before the portfolio fragments. Treat every major AI initiative as both a business outcome and a reusable enterprise asset.

This is how organizations move from promising demos to durable, compounding ROI.

*The views expressed in this article are our own and do not necessarily reflect those of the organizations we represent.

About the Authors:

Jane Chen is Vice President of Data and Analytics at American Tower, where she leads the modernization of the company’s data and intelligence platform, building AI-ready data foundation and enabling reusable data products that accelerate business value. With more than 20 years of experience across technology, healthcare, retail, consumer products, and telecommunications, Jane has led large-scale data, analytics, and AI transformation initiatives that drive growth, operational excellence, and innovation. She partners closely with executive leadership to modernize technology foundations, build trust in data and AI, and foster a data-driven culture while developing high-performing teams.

Alan Lee is a C-suite executive and board-level advisor focused on leadership, strategy, innovation, and enterprise value creation in the AI era. He chairs the Board of Trustees at the NSF Institute for Pure and Applied Mathematics and previously served as CTO of Analog Devices and founder and head of AMD Research.