Most Chief Data Officers (CDOs) do not struggle with AI because they lack ideas. In most organizations, the challenge is the opposite. There are more ideas than capacity, and the real difficulty lies in turning those ideas into outcomes that scale.

Somewhere between experimentation and deployment, AI portfolios become crowded with proofs of concept, priorities lose definition, and measuring impact becomes difficult. While teams stay busy, the business struggles to see the value.

This pattern is widely documented. A 2025 MIT study found that 95% of generative AI pilots fail to produce measurable ROI, largely due to leadership and problem‑framing failures rather than technical limitations. Similarly, RAND research shows that over 80% of AI projects fail because organizations attempt to apply AI to poorly understood or broken processes.

For CDOs, this is the moment where leadership becomes decisive. The question is no longer whether something can be built. The question is whether it should be built, and what must change in the underlying process for the value to materialize.

While the first part of this two-part series explored why many organizations struggle to realize returns from AI investments, this article focuses on the practical side of portfolio design. It explains how data and AI leaders can build an AI portfolio that consistently delivers measurable business value, with process evaluation and transformation at the center.

Start with measurement, not use cases

It is tempting to begin with a long list of use cases. That approach almost always leads to a fragmented portfolio and unclear return on investment. The more disciplined path, and the one that separates high‑performing data organizations, is to define measurement before defining the portfolio.

Every initiative must be evaluated across two dimensions:

1. Technical and operational performance: This includes the cost, data quality, latency, model performance, reliability, and integration complexity. These are essential for delivery, but they do not tell the business whether the initiative mattered.
2. Business impact: What business process will change, what economic value will be created, and how success will be measured in the first three, six, and twelve months.

This discipline mirrors what leading enterprises have learned. At a recent CDO conference, transformation leaders emphasized that technical metrics like “models run in 2 hours instead of 24” are irrelevant unless they translate into faster decisions, lower cost per transaction, or revenue impact.

When expectations are defined upfront, the business case becomes easier to articulate, and the measurement becomes more credible. But measurement alone is not enough.

AI does not create value in isolation. It creates value when a business process changes. This is why process evaluation and transformation must be treated as a core enabler, not an afterthought.

Process evaluation and transformation as the missing link

Many organizations skip the step of evaluating the process that AI is meant to improve. This is one of the primary reasons AI initiatives stall. If the process is broken, fragmented, or inconsistently owned, the AI will not scale, no matter how strong the model is.

This is not theoretical. A Fortune 500 financial services company spent $2.3 million on a customer‑service chatbot pilot that failed completely because the underlying workflow was inconsistent and undocumented.

For CDOs, this means every AI initiative must include a review of the process it touches. The review must determine whether the process owner has the authority to change it, how the “future state” process is defined, and what upstream and downstream dependencies need to be modified.

It must also identify what work will disappear, shift, or be redesigned and the impact on the skills needed to support the future process.

The most successful CDO organizations treat process transformation as a first‑class component of the AI portfolio. It is not a business‑side activity, but a prerequisite for value.

This aligns with findings from Forbes Technology Council: organizations that attempt to eliminate technology debt before addressing process and culture debt often amplify complexity rather than reduce it.

Where use cases should come from and how to evaluate them

Once measurement and process evaluation are in place, the next question is how use cases enter the portfolio.

In product‑driven organizations, the product owner is often the natural entry point because they own the outcome and often the profit and loss. In other organizations, decision rights may sit with a business unit leader, an operations head, or a process owner.

The CDO’s role is to ensure that decision rights are clear, business ownership is explicit, technology partners are involved early, and data and architecture constraints are understood. The CDO must also ensure that process transformation requirements are defined at the start.

This is also where the build‑versus‑buy decision becomes central.

Historically, building models by hand was the default. Today, the ecosystem has matured significantly. Buying tools to automate data engineering and model building is often faster, more scalable, and more cost‑effective for both generative and predictive AI.

The decision to purchase allows for more of the project’s resources to be spent on the process engineering and change management.

Building still matters for highly specialized needs, but the decision must be balanced across cost, speed, scalability, vendor maturity, integration complexity, and the degree to which the business is willing to adapt its processes to the tool.

These conversations involve tradeoffs and discomfort. That discomfort is a sign that the organization is taking the decision seriously.

Cutting through the noise and identifying which POCs matter

Many CDOs inherit portfolios with dozens or even hundreds of proofs of concept. It looks like momentum, but often it is noise. The real work is determining what to scale and what to stop.

The first filter is economic value. The second is the probability of realizing that value. This is where process evaluation becomes decisive.

The CDO must determine whether the process is ready for change, whether the business owner is committed, whether upstream and downstream processes are aligned, whether governance is manageable, and whether deployment is feasible within the existing architecture.

Real‑world examples show the importance of this discipline. JPMorgan’s AI suite reached 230,000 employees and is projected to deliver $2 billion in value, and it is delivered as a governed program rather than a collection of disconnected pilots.

Three small initiatives with clear paths to deployment often outperform one ambitious project that the business is not ready to adopt. A mature CDO organization also has the discipline to stop work.

Ending low‑value initiatives is a sign of strength, not failure.

At the same time, a healthy portfolio includes exploration. Some ideas will not pan out, while others will unlock entirely new capabilities. The goal is to achieve a balance between near‑term value and long‑term innovation.

What a well‑balanced AI portfolio looks like

Over time, a mature AI portfolio becomes a map of the enterprise. Imagine a matrix with business functions on one axis and core processes on the other. Across that matrix, patterns begin to emerge. You can see where AI is deployed, where it is emerging, and where gaps remain.

This eventually becomes a heat map of enterprise readiness and opportunity. Some areas move faster than others, but the trajectory is clear.

This matters because focusing on only one part of the business limits impact. True value comes from embedding AI across functions and processes; not everywhere at once, but with a clear roadmap.

This approach mirrors the AI transformation frameworks which emphasize readiness assessment, use‑case prioritization, and phased scaling as the backbone of sustainable AI value creation.

Early wins create momentum. Over time, the portfolio becomes a strategic asset that supports long‑term competitiveness. When managed this way, reporting becomes easier.

You are not explaining isolated projects, but telling a coherent story about how AI is transforming the business.

The bottom line for CDOs

A high‑performing AI portfolio is not defined by the number of use cases it contains. It is defined by clarity of purpose, discipline of execution, strength of business alignment, commitment to process transformation, and measurable impact. Everything else is noise.

This is the moment for CDOs to lead with precision rather than volume. The organizations that win will be the ones that treat AI not as a collection of experiments but as a strategic capability tied directly to business outcomes and grounded in the processes that create value.

Note: This article is part two of a two-part series exploring how to build a transformative AI portfolio. Part 1 was about “What an AI Portfolio Really Is and Why Only 5% of Companies See Measurable Impact.”