Enterprise AI’s “last mile” problem: Why ROI arrives through governed AI-driven applications

Enterprise companies are not challenged by a lack of AI ambition. Most organizations have been investing in platforms, people and projects to keep the Board and executive leadership feeling like they are keeping pace in the most disruptive technology revolution in history.

The truth of the matter is most of these companies lack the tools to turn experimentation into repeatable business outcomes at scale. BARC research shows that as more projects are deployed, companies are relying less on IT as a partner and moving toward outside resources to find a more successful path forward. 22% of survey respondents are somewhat or very dissatisfied with their experience working with IT to deliver AI projects [1]. Companies that focus on delivering AI-driven applications can sidestep these issues and leverage applications to extend and augment governance requirements for users while scaling faster than those that don’t.

This scale and repeatability challenge represents the last mile of enterprise AI for companies that don’t and or can’t adopt AI-application development. It’s where ROI is either realized or quietly disappears. A consistent pattern across all industries shows teams are producing AI projects faster than they can be governed or operationalized. Fear of missing out (FOMO) is outpacing control and exposing companies to significant risk. In short, the faster AI gets, the more the enterprise needs integrated guardrails that can keep up.

AI-driven applications are the scaling vehicle

AI-driven applications can be widely defined. Agentic workflows, custom applications tailored for subject matter experts and applications that leverage statistical analysis and data science algorithms. These applications are built to be:

• Used by business users, not just technical teams
• Operational, not experimental
• Continuously improved, not periodically rebuilt
• Governed by design, not governed after the fact

Fast, governed, application development matters because scaling AI is less like deploying one system and more like managing a diverse portfolio of AI and data assets. As that portfolio evolves, risk becomes less about a single bad model and more about systemic issues: inconsistent controls, unclear accountability, weak monitoring, and uncontrolled proliferation.

When the AI portfolio is poorly managed or governed, enterprise leaders immediately experience lower ROI, or in some cases, no ROI. If your company is struggling to scale AI and projects are languishing in proof of concept (POC) and prototype phase, there isn’t an easy path to ROI.

BARC research highlights how wide the scaling gap remains. Research findings indicate that agentic AI is already moving beyond an experimentation phase, with 32% of organizations running AI agents in production and 26% piloting. Additionally, in a soon-to-be-released study by Domino Data Lab and supported by BARC, respondents were clearly moving faster than their governance strategy with 32% of practitioners building, deploying and monitoring AI Agents with partially formalized governance strategies.

And while all of this is happening, the data shows only about 21% have operationalized the foundations needed to scale safely. Those foundations include leadership alignment, security and compliance, data access policies, legal alignment, governance frameworks, standards and ethics, and enterprise technology architecture. That mismatch, rapid adoption and slower control is where last-mile ROI breaks down.

Governance should enable innovation, not slow it down

Building AI-powered applications within a governance framework is the answer for companies needing to scale. Governance shouldn’t slow your progress. It should enable it, allowing more applications to reach the people who need them and creating repeatable, governed workflows. When done correctly, risk is reduced, output is increased and controls become automated, minimizing rework and maximizing AI value. Governance is the foundation to AI innovation.

BARC research on data products offers a useful adjacent signal. Organizations that establish governed data products company-wide are far more likely to operationalize AI at scale. The report showed that 85% of respondents with company-wide data products report three or more AI projects in production in comparison to just 25% of companies without governed data products. [2] Data products aren’t the same as AI-driven applications, but the underlying principle is similar. Packaging capabilities into reusable, governed units correlate with production scale and ROI expectations.

Why AI-driven applications can reduce risk while scaling impact

There is a common fear that scaling AI inherently increases risk. In practice, the opposite can be true if scaling happens through governed applications rather than unmanaged proliferation.

When AI reaches the business through well-designed applications, risk can fall because:

• The user experience can constrain unsafe behaviors and encourage correct workflows
• Data access can be controlled centrally instead of through ad hoc retrieval
• Monitoring and auditability can be standardized
• Validation can be industrialized rather than bespoke
• Compliance documentation can be generated as part of delivery, not as an afterthought

This becomes increasingly important as the audience of AI expands beyond technical teams. If business adoption grows but controls do not, risk becomes the reason scaling stops.

A simple framework for the last mile

Here’s what you need for driving ROI and scale for your enterprise AI strategy.

Seamless AI application development

• Fast and iterative development tools for applications and agents
• Standardized access to data, models and algorithms
• Reusable processes for deployment and monitoring

Integrated governance

• Integrated policy guidelines and enforcement
• Automated governance alerting and monitoring

Scaling Strategy

• Get away from experiments and move to application and product-oriented development
• Embrace governance as a value-add and road to ROI

This is what it takes to reach the last mile, where enterprise AI stops being a collection of pilots and becomes a scalable engine for productivity and measurable ROI.

[1] BARC Research, Lessons from the leading edge, December 2025 Shawn Rogers and Merv Adrian n=421
[2] BARC Topical Survey: “Data Products and Data Contracts in 2026: The Foundation for AI Success” (Carsten Bange, Florian Bigelmaier, March 2026) n=308