Manage Your AI Investments Like a Portfolio

Business leaders now face intense pressure to transform their organizations with AI, even though the technology, public attitudes, and the competitive landscape are all still in flux. The result is often too many pilots with too little coordinated oversight. Without a way to systematically decide where to start, how fast to move, and when to stop, AI efforts quickly become a drain on attention and resources rather than a source of advantage. A familiar pattern recurs across many companies: isolated, piecemeal deployments, limited buy-in by senior executives, and weak linkage to strategic goals.

In 2018, one of us (Tom Davenport) argued that companies pursuing AI should create portfolios of projects based on business needs rather than rolling the dice on moonshots. That article identified the need for systematic portfolio development but stopped short of detailing how to build and manage such portfolios. This article completes that picture.

Organizations need to follow a disciplined step-by-step portfolio approach that treats AI innovation as a structured pipeline of projects that is managed by applying coherent, repeatable principles. This approach enables leaders to allocate scarce resources strategically, secure and maintain executive sponsorship across multiple initiatives, and sequence the right projects at the right time rather than chasing disconnected proofs of concept.

Advertisement

The approach described here builds on proven portfolio management approaches, including those that members of the author team have implemented across large organizations such as Northrop Grumman, PepsiCo, and units within the U.S. Army. While AI-specific deployments are still emerging, the adaptation presented here incorporates insights from both past and present implementations.

Why Take a Step-by-Step Approach to AI Portfolio Management?

Processes for managing R&D and new product development have long employed tightly defined go/no-go gates—known as “stage gates”—to increase the likelihood of worthy innovations reaching the market. Combining these tools for managing project progression with a portfolio management approach to AI innovation addresses core failure modes for AI implementations. This approach ensures that every candidate project is judged not only on its standalone merits but also in relation to competing opportunities, enterprise priorities, and cross-initiative dependencies. At the same time, because projects progress through defined gates that assess their feasibility and strategic fit, organizations are protected from the distractions caused by a flood of strategically disconnected pilots.

The portfolio approach:

• Elevates AI from a series of departmental experiments to a board-level strategic imperative.

• Allows executives to view all current and planned AI initiatives on a single dashboard, including their interdependencies, resource requirements, and strategic alignments.

• Formalizes a strategic approach to time-horizons across projects, enabling leaders to select a mix of projects across:

  • near-term implementations that build confidence and capabilities

  • medium-term initiatives that require deeper integration but deliver more substantial transformation

  • longer-term projects that offer transformational potential

• Surfaces interdependencies across projects, allowing organizations to sequence initiatives so that earlier projects build foundational capabilities required by later, more sophisticated implementations.

Portfolio management has long enabled organizations to manage innovation in a holistic manner—capturing both top-down strategic imperatives and bottom-up opportunities, while unifying efforts horizontally across departments and developing strategies that build capabilities over time. The AI-specific portfolio approach adapts time-tested portfolio principles to AI’s unique characteristics.

How the Portfolio Runs

The step-by-step portfolio management approach views the connected portfolio through a dual lens: first, as an advancement pipeline with clear gates through which projects must pass; and second, as a whole-portfolio dashboard that shows balances across risk/return, time horizon, capability areas, and mission alignment. This dual perspective enables both rigorous project-level discipline and strategic portfolio-level optimization.

Three interconnected mechanisms make this system work in practice:

Buy/Sell/Hold scoring ranks backlog items against objective criteria to determine relative priority. These rankings transform subjective debates into structured conversations about trade-offs.

• Strategic alignment measures how well an initiative serves core business objectives.

• Feasibility assesses technical capability and organizational readiness.

• Risk-reward profiles weigh potential upside against implementation challenges.

• Resource requirements assess needs across financial, human, and technical dimensions.

Stage gates apply progression tests at each portfolio transition.

• Is required data available, of sufficient quality, and properly governed?

• Are necessary skills in place or clearly sourced?

• Does the business process the AI system supports need to be redesigned?

• Have ethical guidelines and security controls been established?

• Does the business case remain valid?

An electric utility company in the Northeast U.S. established a disciplined progression approach for its AI initiatives that relied on go/no-go reviews. The company was relatively early in its AI journey and realized that it needed a means of evaluating possible projects from ideation to production deployment. The utility’s AI Center of Excellence (AI CoE) gathers ideas for use cases from different teams. The first stage is evaluation by AI CoE members of whether the proposed idea meets a business need. If it passes this review, a proof of concept is developed. The project then advances to an evaluation involving senior management, the regulatory team, the legal team, and AI CoE members. If approved, the project is then sent to the CEO and Chief Information Officer to evaluate potential return on the investment and to take the use case into production. The company feels that disciplined progression through defined gates is particularly important when regulatory issues come into play. A lighter version of the process will be appropriate for industries with lower regulatory burdens.

Regular reviews rebalance the entire portfolio by adding promising initiatives, scaling successful pilots, redirecting struggling efforts, retiring misaligned projects, and recalibrating objective scoring criteria. These reviews assess both individual project health and overall portfolio coverage:

• Are we maintaining appropriate balance across time horizons?

• Do we have sufficient investment in foundational capabilities?

• Do early-warning triggers, such as schedule slippage, cost overruns, or performance shortfalls suggest that intervention is needed?

Portfolio Management in Practice: Lloyds Banking Group

Lloyds Banking Group’s “GenAI Control Tower” demonstrates how AI portfolio management can work in practice. This cross-functional forum prioritizes use cases across the organization, allocates resources, and ensures alignment with strategic priorities. The Lloyds model explicitly balances long-term transformation with short-term value delivery, building capability incrementally over time, while recognizing that rapid changes in available technologies can warrant abandoning ongoing projects and switching rapidly to new use cases. Each AI initiative must pass through rigorous reviews – including risk assessment, legal review, and controls for ethics, bias, and security – before advancing to production. A centralized playbook for AI development alongside clear decision rights on whether to build or buy solutions ensures that Lloyds’ portfolio of AI initiatives advances in a structured yet adaptive manner, exemplifying how large enterprises can manage diverse AI portfolios at scale.

The Four Portfolio Stages

We recommend moving ideas through four distinct innovation portfolio stages, each with its own mindset, assessment artifacts, and go/no-go stage gate. These stages align with the OPEN framework (Outline, Partner, Experiment, Navigate) for AI innovation outlined in a previous HBR article, but the broad approach is framework-agnostic and can be adapted to other methodologies.

Stage 1: The Opportunity Portfolio (Outline). This serves as a centralized location for the intake and triage of initial ideas. Activities include problem framing, initial value and risk hypotheses, and dependency mapping. The output is a scored, ranked backlog of outline ideas.

Stage Gate: Test of strategic fit and technical feasibility.

Stage 2: Design & Partnership Portfolio (Partner). Outlined projects now enter a more detailed shaping phase, with innovation teams developing extensive business cases that articulate expected benefits, required investments, and success metrics at a much greater level of detail. They map dependencies and develop capability plans that identify what technical skills, data assets, and infrastructure will be required. The teams then establish preliminary governance models that define decision rights, risk oversight, and compliance requirements. Partnership considerations are critical at this stage. Even organizations with strong internal capabilities typically need external partnerships—with technology vendors, academic researchers, ethics advisors, or industry peers—to realize their AI ambitions. It is also essential to ask what relationships will emerge between humans and AI systems as the project develops, as this will impact staffing levels, work patterns, reporting structures, and individual roles.

Stage Gate: Confirmation that required data is available and properly governed, necessary skills are in place or clearly sourced, ethical and security controls are defined, and the business case remains compelling.

Stage 3: Experimental/Prototyping Portfolio (Experiment). The key to successful AI experimentation is structuring experiments as learning journeys rather than validation exercises. Each experiment should test not just technical feasibility but also enterprise viability and human desirability. Ask: Can the AI actually perform the intended function? Can we integrate it with existing systems and processes at reasonable cost? Will users adopt it and find value in it? Rapid, bounded trials enable this multidimensional learning. Teams might develop paper models to test conceptual approaches, build minimum viable products to validate core functionality, or deploy limited pilots to gather real-world performance data. Multiple parallel experiments can explore different technical approaches or implementation strategies.

Stage Gate: Rigorous system testing and red-team scrutiny (deliberate attempts to break or misuse the system).

Stage 4: Scale & Operate Portfolio (Navigate). Production deployment marks a fundamental shift in focus and usually entails a considerable increase in cost and development time. To enter production, an AI system typically requires scaling for multiple users, upskilling employees, redesigning processes to take advantage of new capabilities, and integration with the existing technical environment. This stage emphasizes service management, continuous monitoring, and knowledge capture. Teams establish guardrails that prevent system misuse, implement cost-tracking to understand total cost of ownership, and measure actual mission impact rather than assuming benefits.

Regular reviews of the whole portfolio are essential for maintaining strategic alignment as both technological capabilities and organizational needs evolve. The portfolio must remain balanced across time horizons and risk/return levels, with steady advancement of high-potential projects through the pipeline and the addition of fresh ideas to the Opportunity Portfolio. This creates the continuous innovation flow needed to ensure that organizations remain at the leading edge of a rapidly evolving field.

Project advancement is determined by resource availability and objective scoring against fixed criteria. When capacity opens—for example, when an experimental project is advanced to production or a new team is stood up—leaders select the next item from the highest-scoring backlog entries that have passed the relevant gate. The scoring criteria ensure that portfolio balance is maintained across business units, time horizons, and risk profiles while the resource-constrained flow of projects ensures that initiatives can be executed effectively.

. . .

A disciplined, step-by-step portfolio approach to AI innovation represents a fundamental shift in how companies must think about technological innovation. By treating AI initiatives as an interconnected investment portfolio rather than as isolated projects, organizations create the visibility, discipline, and strategic coherence necessary to deliver sustained transformation. This approach bridges the gap between technical possibility and business reality: it elevates AI decisions to the C-suite level where they belong, ensures continuous executive sponsorship through structured stages, and provides the transparency needed for leaders to make informed trade-offs between competing priorities. Most importantly, it creates a sustainable pipeline of innovation projects that balances transformative moonshots with practical quick wins, ensuring that organizations can build capability while also delivering value.

Copyright 2026 Harvard Business School Publishing Corporation. Distributed by The New York Times Syndicate.

Explore AAPL Membership benefits.