Upgrading the Mid-Year Review for Your AI Portfolio

JUN 15, 2026  ▪  KATHLEEN G. CO  ▪  PMI PHILIPPINES DIGITAL STRATEGY LEAD  ▪  SINGAPORE 

Every mid-year, project and portfolio management professionals go through a familiar ritual. We pull out our dashboards, review delivery milestones, check budget burn rates, and ask whether our strategic initiatives are still on track. For most portfolios, it works. But for AI portfolios, something critical is missing from that review. And most organizations won't notice until it's too late.

The gap sits with the leaders responsible for governing AI transformation. As we hit the midpoint of yet another year where AI has moved faster than anyone anticipated, that gap needs honest attention.

TRANSFORMATION IS OUTPACING LEADERSHIP

AI has moved from experiment to enterprise infrastructure faster than most of us expected. In financial services, healthcare, and shared services — sectors many of us work in or support here in the region — AI isn't a pilot anymore. It's embedded in credit decisions, customer service workflows, fraud detection, and resource forecasting. The pace of that growth has quietly outrun the governance structures organizations put in place to manage it.

I recently completed research at Singapore Management University (SMU) in collaboration with Boston Consulting Group (BCG) pointing to a persistent problem: leaders accountable for these systems are frequently unprepared to govern them. AI governance demands a different kind of leadership capability than what most executives were developed for, and most leadership programs haven't caught up yet.

"A 2026 Grant Thornton survey of banking leaders puts a number on it: only 18% say they're fully confident they could pass an independent review of their AI controls within 90 days, and 50% say governance and compliance barriers are actively limiting their AI performance. That's a leadership readiness problem, and it won't close by throwing more budget or better tools at it.

WHAT YOUR MID-YEAR REVIEW IS PROBABLY MISSING

Traditional portfolio reviews are built around execution metrics: schedule adherence, scope control, benefit realization, risk status. These matter, but for AI portfolios they're not enough. Here are three failure modes that won't show up on your standard mid-year dashboard. Chances are, at least one is already present. What makes them particularly hard to catch at mid-year is that most are seeded much earlier at project initiation, during model evaluation, or at the point of first deployment, and they often only become visible once the system is embedded in operations.

DELEGATED ACCOUNTABILITY - FAILURE MODE ONE

A senior leader approves an AI initiative, assigns a team, and gets periodic status updates. On paper, everything's green. In practice, that leader can't independently evaluate whether the AI system is producing reliable outputs, whether human oversight is genuinely working, or whether the ethical guardrails set at the start are still holding. Accountability has been passed down so many times that the person at the top is signing off on outcomes they can't actually assess. The authority to approve an AI system and the substantive understanding of what you've approved are two different things. Most leaders conflate them. Think about what that looks like in a real review. The PMO reports adoption is up, the system is live, stakeholders are satisfied. Yet nobody in the room can answer: what happens when this system gets it wrong, who's responsible, and how would we even know?

OPTIMISM BIAS IN AI ROI REPORTING - FAILURE MODE TWO

Mid-year reviews love good numbers: users onboarded, processes automated, hours saved. These are real. They often mask a harder question, though: is the organization building genuine AI capability or accumulating AI dependency? The two look identical on a dashboard, and the difference only becomes visible when something goes wrong.

In 2026, Amazon discovered this the hard way. The company built an internal leaderboard to track employee AI adoption. Workers began gaming it, manufacturing token counts rather than doing genuine AIassisted work. The metric became the target, and the target stopped measuring what mattered. Meta hit the same wall shortly after, when an employee AI usage leaderboard went viral and was shut down within days. When two of the world's most AI-sophisticated organizations can't reliably measure AI initiatives, the challenge for the rest of us deserves serious attention.

GOVERNANCE DRIFT - FAILURE MODE THREE

AI systems aren't static. They get retrained, updated, expanded. The governance boundaries agreed at project approval are rarely revisited as the system evolves. By mid-year, an AI system that started in a low-risk corner of operations may be making decisions that warrant significantly higher oversight. Nobody flagged it because the review process was never designed to ask that question.

Researchers call this pattern performative compliance: governance structures that exist on paper but quietly stop functioning in practice, while every status report still reads green. This is compounded by model drift: the gradual degradation of AI performance as real-world conditions diverge from the data the system was originally trained on. This can occur invisibly while governance reviews focus only on process compliance.

The legal world is catching up to what happens when leadership stops reviewing what AI produces. In 2024, Air Canada was held liable in a Canadian tribunal for a refund its customer service chatbot invented. The airline's position that the chatbot was a separate entity responsible for its own statements was rejected outright. Leadership owned the output, whether they reviewed it or not.

THE LEADERSHIP CAPABILITY GAP

So what does good AI governance leadership actually look like? The same BCG-SMU research identified five capabilities that separate leaders who govern AI well from those who govern it in name only. These aren't technical skills; they're governance capabilities that have to be deliberately developed.

At scale, the risk isn't that oversight structures don't exist. It's that they quietly normalize into procedural exercises while everything on paper still looks fine. Structured AI project management methodology addresses the delivery layer; these capabilities address the governance layer above it, and both are necessary.

THE EVOLVING REGULATORY LANDSCAPE

AI deployment has outpaced regulatory development almost everywhere. Across industries and sectors, organizations are already operating AI systems in production while many of the governance frameworks intended to oversee them are still being finalized.

Regulators are responding, but at different speeds and from different starting points. Internationally, ISO/IEC 42001 and the OECD AI Principles have established a clear baseline: human accountability for AI outcomes is non-negotiable, regardless of jurisdiction. The EU's AI Act goes further, imposing direct obligations on organizations deploying high-risk AI systems. In May 2026, its implementation was postponed to give organizations, regulators, and the supporting infrastructure more time to prepare.

In Singapore, the Monetary Authority of Singapore has moved more decisively. Its 2025 AI risk management guidelines place explicit accountability on designated senior leaders, emphasizing personal responsibility rather than relying solely on institutional or committee oversight. Closer to home, the Bangko Sentral ng Pilipinas has signaled that AI governance requirements for Philippine financial institutions are forthcoming, with particular attention to algorithmic bias, model accuracy, and ethical use.

The challenge is that AI adoption is already well underway. BSP 2024 data indicates that >44% of supervised institutions are already using AI in production. While regulations continue to evolve, the direction is unmistakable: governing AI is becoming a leadership responsibility, not merely a compliance exercise.

For project and portfolio leaders operating in complex environments, this means making decisions today that will need to withstand tomorrow's regulatory scrutiny. That gap between what AI systems are already doing and what governance frameworks formally require is precisely where many of us are already operating.

THREE KEY AI PORTFOLIO QUESTIONS TO INCLUDE IN YOUR NEXT REVIEW

You don't need to overhaul your governance framework to start effective AI reviews. Add these three questions to your mid-year portfolio conversation and see what surfaces

ONE - THE EXPLANATION TEST - - - - - - - - - - - - - - - - - - - - - - -

For each significant AI initiative, can the accountable leader explain, in their own words and without prompting from a technical team member, how the system works, where the human checkpoints are, and what would trigger an escalation?

TWO - THE BOUNDARY AUDIT - - - - - - - - - - - - - - - - - - - - - -

Have the governance boundaries for each AI system been reviewed since original approval?

THREE - THE HORIZON CHECK - - - - - - - - - - - - - - - - - - - - - -

When did your leadership team last engage directly with AI governance developments, through emerging frameworks, industry incidents, or regulatory updates, rather than waiting to be briefed? These questions won't fix everything, but they'll tell you quickly whether something needs attention. Put simply, if your answer needs a briefing note, accountability has already drifted further than you think.

RAISING THE STANDARD

Our profession has always known that delivering a project isn't the same as delivering value. We've spent decades building the discipline to close that gap through governance, stakeholder management, and benefits realization frameworks.

AI is asking us to extend that discipline into new territory. It's no longer just about being on time and on budget. It's about whether the leaders responsible for AI portfolios can genuinely account for what those portfolios produce to their boards, their regulators, and the people their systems affect.

As you head into your mid-year reviews, consider whether leadership readiness belongs on your agenda alongside delivery metrics. It's a core governance obligation, not an optional add-on. In AI transformation, what you don't govern will eventually govern you.

Disclaimer: The research referenced in this article was conducted as part of an SMU-BCG AI leadership readiness study focused on Singapore's financial sector. The author writes in her personal capacity.

ABOUT THE AUTHOR

KATHLEEN G. CO is a Singapore-based digital and AI transformation leader with 25+ years of experience advising leaders across Fortune 500 and regulated industries. She recently founded Citrus Tech and completed AI leadership readiness research in collaboration with Boston Consulting Group at Singapore Management University. If this article resonates with challenges you're navigating, Kathleen would welcome a conversation. Connect with her directly on LinkedIn www.linkedin.com/in/kathleengiselle.