Managing AI Project Risk: An Executive Summary for Leaders of Middle Market Firms

Based on research by Albert Segars and Arvind Malhotra, UNC Kenan-Flagler Business School¹

Artificial intelligence technologies form the leading edge of business innovation, as AI investments represent extraordinary opportunity as well as existential risk, particularly for leaders of middle market firms. Operating with tight margins and subject to fierce competition, these organizations cannot afford to invest capital in technologies and systems without a real ROI. They also cannot afford to fall behind advancing industry benchmarks and standards, which are increasingly determined by AI.

A 2025 report² from the Massachusetts Institute of Technology finds that 95% of AI projects fail, indicating the hazards involved in these undertakings. Yet Segars and Malhotra’s research shows that companies that take a structured, well-reasoned approach to evaluating AI project-level risks and opportunities achieve much higher success rates. Drawing from surveys and interviews with more than 400 project managers leading AI initiatives, this research provides a practical framework for assessing and managing AI implementation risk. Middle market business leaders can use this framework to mitigate an AI project’s inherent perils and maximize its benefits.

An Integrated – Not Fragmented – Approach to AI

Firms should think of AI as a capability that managers integrate across systems and processes already in use, rather than a single technology that is added to existing structures. Because AI works with us and not for us, its adoption requires a shift in thinking as well as distinct approaches and assumptions from other technologies. AI integration also compels business leaders to give greater care and attention to human-technology interfaces, especially in the initial stages of adoption.

To understand AI technology integration, it is first helpful to categorize AI applications into four classes and overlay these groupings with existing information technology systems.

AI Application	What It Does	Examples
Natural Language Processing	Enables computers to understand and generate human language.	Virtual assistants, sentiment analysis, research agents
Machine Learning	Identifies patterns in data to support prediction and improve decision-making.	Fraud detection, health monitoring, digital operations, autonomous vehicles
Generative AI	Creates content based on patterns learned from large-scale datasets.	Code production, knowledge mapping, chatbots, report generation
Computer Vision	Allows digital technologies to interpret and analyze visual data.	Object detection, classification, tracking, virtual reality, digit identification, emotional recognition

Information technology systems include: Software development, infrastructure, cybersecurity, cloud, automation, data processing and analytics, digital transformation, enterprise resource planning, and legacy system replacement or upgrade.

Middle market leaders must assess the most suitable integration points in their firm’s existing technology stack to make sound decisions about potential AI projects. To find the most promising starting point for AI integration, experts recommend that managers look for internal systems that involve sorting and analyzing copious amounts of data. These often include HR enterprise systems, such as payroll, benefits and retirement accounts, which AI technologies can help streamline and demystify for employers and employees alike. Other primary integration points involve cybersecurity, including fraud detection, as AI technologies have proven successes in this domain.

AI Projects Require a Robust Risk Benefit Analysis

To assess whether an AI project is worth pursuing requires a strategic review of the project’s potential benefits and associated risks. AI technology adoption can create value for firms by providing four main benefits — speed, accuracy, revelation and adaptability — enabling organizations to operate more efficiently, make better-informed decisions, generate new insights and effectively respond to changing conditions. Realizing these benefits requires managers to candidly evaluate project-specific risks. Segars and Malhotra identify three key risk factors that determine AI project success: problem definition, technical experience and project scope.

1. Speed: AI integration and automation can generate faster production processes, analysis and decision cycles.
2. Accuracy: AI technologies can improve data quality, resulting in better code, improved insights and enhanced decision-making.
3. Revelation: AI tools can organize and analyze large quantities of data, revealing signals, meaning and implications of the information on hand.
4. Adaptability: AI can build operational capabilities allowing firms to customize and self-correct.

The Three Key Risk Factors of AI Project Integration:

1) Problem definition: How well is the problem or opportunity understood?

• Business leaders must clearly define the problem they are trying to solve and the potential benefits of solving it before investing in an AI solution. In the problem-defining process, successful organizations employ information discipline, using “fast and frugal” decision-making that focuses on essential inputs and avoids analysis paralysis.

• Grouping the four AI benefits (speed, accuracy, revelation, adaptability) with the five W’s of problem analysis (what, who, where, when, why) is a proven, effective process for information gathering. Middle market firms should leverage the productivity of small, focused teams to improve information discipline and achieve clear problem definition.

2) Technical experience: Does the implementing organization have the skills and knowledge resources to address the problem and fulfill the opportunity?

• Firms often overestimate their readiness for AI technology implementation, as leaders assume that their IT staff also has AI expertise. This is often not the case — AI is a rapidly evolving field where specialized expertise remains in short supply.

• Organizations that successfully integrate AI technologies tend to build a portfolio of AI technical capabilities. Organizations develop this portfolio using three approaches: upskilling employees in data, analytics and IT; partnering with vendors, consultants and peer organizations; and leveraging generative AI to design, test and refine other AI applications.

• Firms should employ a combination of these knowledge portfolio approaches customized for each project. Specialized integrations may require deep internal expertise while commodity-like projects can be outsourced. For middle market firms, this means matching capabilities and expertise source with project type.

3) Scope: What is the project magnitude in terms of the human resources, process change and capital involved in its execution?

• The most mistake-prone factor of AI project risk assessment, a project’s scope is commonly underestimated. It is helpful to divide scope along two dimensions:
  • Reach: How many groups, departments or organizations does the project span? Higher reach means greater coordination, more authority boundaries to navigate and higher risk.
  • Range: How much will the project change existing methods, processes and ways of working? Higher range means more behavioral change required and a paradox: Realizing AI’s full benefits often requires significant process redesign.

• Together, reach and range determine project scope and risk. The greater the scope, the greater the risk and the greater the potential benefits. The best approach balances risks and rewards while impartially assessing scope, planning for the requisite behavioral change, and investing in the governance and stress-testing needed to manage the project.

Figure 1: Project Risk Matrix

The risk matrix in Figure 1 classifies AI project types according to risk level. This matrix gives business leaders a framework for evaluating AI initiatives according to each project’s relative riskiness, helping managers compare projects, prioritize investments and identify the necessary steps to reduce vulnerabilities. It is important for business leaders to note that while lower-risk projects are often preferred, especially for firms taking on their first AI initiatives, higher-risk projects usually offer greater and more transformative benefits. Managers therefore should not always avoid these high-risk, high-reward initiatives. No matter the risk level, it is crucially important for business leaders to approach AI projects with a measured understanding of the risks involved and a plan for mitigating these risks throughout implementation.

Successful Approaches to Each Project Type

Building an AI Foundation Through Low-Risk Project Implementation

Successful organizations tend to use lower-risk projects to build their capabilities before taking on higher-risk, higher-reward projects. Firms amass wins by taking on lower-risk (Types 1 and 2) projects, using these initiatives to build technical, cognitive and behavioral capacity. Leveraging these new capabilities, implementing companies then take on more transformational projects. Segars and Malhotra’s research finds that the firms achieving the best results are those that build momentum through a focused portfolio of achievable projects. This approach is particularly salient for middle market firms, whose leaders have little margin for error.

Seven Rules of Thumb for AI Project Portfolio Management

Segars and Malhotra’s research distills seven practical rules that middle market leaders can apply to their AI investment decisions:

1. Beware of overinnovation: Too many high-risk projects (Types 7 and 8) can drain resources and stall progress.
2. Beware of underinnovation: Too many low-risk projects (Types 1 and 2) generate activity without breakthrough impact.
3. Practice antifragility management: Managers can move high-risk projects toward less risky territory through deliberate intervention. These projects do not have to be abandoned.
4. Guard against overconfidence: External shocks (e.g., tariffs, regulation, technology shifts, competitive moves) can turn easy projects into hard ones overnight, highlighting the need for continuous monitoring.
5. Harness existing knowledge: When the organization has technical capability but unclear problem definition (Types 5 and 6), focus on translating know-how into application.
6. Find a helping hand: When the problem is clear but capability is lacking (Types 3 and 4), seek external expertise.
7. Bring order to chaos: A project portfolio clustered in any single quadrant of the risk matrix may indicate a lack of strategic balance. Aim for a mix of initiatives that matches your organization’s risk-bearing capacity.

Implications for Middle Market Business Leaders

AI initiatives fail when organizations act hastily and without sound opportunity and risk appraisal informing their strategy. Successful AI implementation results from a disciplined approach to risk assessment and mitigation.

Segars and Malhotra’s research recommends that managers approach AI initiatives with a holistic perspective and look to integrate AI technologies into existing systems (e.g., analytics, enterprise resource planning, legacy platforms, cybersecurity), areas with well-understood problems and extant technical capabilities. Initial successes build the foundation for more ambitious initiatives. Technical experience, grown through upskilling, external partnerships, and the use of AI tools, is a more durable source of competitive advantage than any specific technology, meaning managers should develop their team and partnerships before scaling projects.

Firms should continuously monitor AI project operations and adjust their processes as needed. Managers must regularly review their AI portfolio according to this research framework’s risk matrix and intervene when necessary. This strategy requires honest assessment of organizational capabilities, focused problem definition, realistic scoping and the patience to build proficiencies through iterative processes.

Middle market firms can leverage their small and nimble teams to make quick decisions with direct executive engagement, giving these organizations a structural advantage in AI adoption. It is unlikely that any single AI project will prove transformational, yet AI can deliver extraordinary value through a well-managed portfolio of integration initiatives.

---

Written by:

Arvind Malhotra, H. Allen Andrew Distinguished Professor of Strategy and Entrepreneurship, UNC Kenan-Flagler Business School

Albert H. Segars, PNC Distinguished Professor of Strategy and Entrepreneurship, UNC Kenan-Flagler Business School

Carlin Rosengarten, Technical Business Writer, UNC Kenan-Flagler Business School

---