Overview

Bashir Farahani oversees AI investment at a regional bank with about $8 billion in assets. Three years ago, his team had fifteen AI projects running simultaneously - fraud detection, credit scoring, a chatbot, document processing, branch optimisation, and nine others. Every project had a sponsor, a team, and a budget line. None had a strategic relationship with the others. "We were treating innovation like a wish list," he told me. "We funded everything that seemed promising and hoped some of it would be transformative." At the end of the year, three projects were in production, four had been quietly abandoned, and the other eight were in various states of limbo. Total investment: $14 million. Measurable business impact: modest improvements to fraud detection. Everything else was still a pilot.

Innovation portfolio management is the discipline of treating your AI initiatives the way a fund manager treats investments - as a portfolio that needs deliberate balance, not a collection of independent bets. That shift in framing changes how you approve projects, how you resource them, and how you decide when to stop.

Why Portfolios, Not Projects

When AI initiatives are managed as individual projects, each one is evaluated on its own merits. The strongest advocates win the most resources. Projects are rarely killed because killing them feels like failure. Over time, the portfolio drifts toward the pet interests of the most persuasive people - not toward the organisation's strategic priorities.

Portfolio thinking reframes the question. Instead of "is this project worth doing?" the question becomes "does this project make our portfolio better?" A project that would be approved on its own merits might be rejected if the portfolio already has too much exposure in that category. A project that seems modest in isolation might be essential because it fills a strategic gap.

Think of it like a financial portfolio. A rational investor does not put 80% of their money in a single asset class, even a promising one. They balance across asset classes to manage risk and capture opportunities at different time horizons. An AI portfolio should work the same way.

Classifying AI Initiatives: The Three Horizons

The most widely useful classification framework divides AI initiatives into three horizons based on time to value and risk level.

Horizon 1: Incremental improvements to existing operations. These are AI applications to processes you already run - making them faster, more accurate, or cheaper. Fraud detection rule enhancement, invoice processing automation, predictive maintenance for known equipment failures. Time to value: 6-12 months. Risk: low. Expected return: modest but reliable.

Horizon 2: New AI-enabled capabilities. These create new products, services, or operational models that did not exist before. A personalised credit product priced dynamically using real-time data. A customer-facing AI advisor that goes beyond FAQ answering. Time to value: 12-24 months. Risk: medium. Expected return: significant if successful, but with meaningful execution risk.

Horizon 3: Transformational bets. These are moonshots - AI initiatives that, if successful, could fundamentally change the organisation's competitive position or business model. They may fail. They require protected resources and different success metrics than operational projects. Time to value: 3-5 years or more. Risk: high. Expected return: very large if successful, zero if not.

A healthy portfolio has initiatives in all three horizons. Bashir's mistake was having fifteen Horizon 2 projects and no clear Horizon 1 quick wins to build organisational confidence, and no Horizon 3 moonshots to create genuine differentiation. He had chosen the middle-risk, middle-return tier for everything - and got middle results.

Setting Portfolio Balance

The right portfolio balance depends on your organisation's risk appetite, competitive position, and cash flow situation. There is no universal answer, but some principles apply broadly.

Conservative organisations - regulated industries, incumbent market leaders, organisations with tight cost structures - typically run portfolios weighted toward Horizon 1, with selective Horizon 2 investments and minimal Horizon 3. A reasonable starting point: 70% Horizon 1, 25% Horizon 2, 5% Horizon 3 by budget.

Challenger organisations - those fighting for market share against established players, or those facing genuine existential threats from digital competitors - need to accept more risk. A more aggressive allocation might be 40% Horizon 1, 40% Horizon 2, 20% Horizon 3.

Bashir recalibrated after his three difficult years. He consolidated his fifteen Horizon 2 projects to four, added two deliberate Horizon 1 quick wins that could show ROI within six months, and ring-fenced a small budget for one genuine Horizon 3 experiment - AI-driven real-time lending decisions. "I had to stop saying yes to everything interesting," he said. "That felt like leadership. It was actually just noise."

Resourcing the Portfolio

One of the most common portfolio errors is distributing resources evenly across initiatives. If you have twelve projects and a fixed budget, giving each 1/12 of the budget means you have twelve underfunded projects. None can move fast enough to succeed. All are vulnerable to the first budget pressure.

Portfolio resourcing should follow conviction. The projects you believe in most should have enough resource to execute at pace. Projects you are less certain about should be appropriately small - big enough to generate real learning, small enough that failure is affordable.

Horizon 1 projects should be staffed for speed: clear scope, dedicated team, short sprint cycles. The goal is to generate value quickly and build organisational momentum.

Horizon 3 projects should be resourced differently: small dedicated team, ring-fenced budget, and protected from the quarterly pressure to show ROI. Applying standard business case metrics to a Horizon 3 experiment will kill it. The right metric for a Horizon 3 project is learning, not return.

Managing the Pipeline: Intake, Review, and Kill Decisions

A portfolio is not a static list. It should have a managed pipeline - a process for how new ideas enter, how they are evaluated, and how decisions are made to advance, hold, or kill them.

Intake: new AI proposals should enter a standardised evaluation process that asks: what horizon is this? What is the strategic fit? What are the key risks and success criteria? This does not need to be a lengthy process - a two-page proposal template and a 30-minute review meeting is sufficient for most initiatives.

Quarterly review: portfolio composition should be reviewed at least every quarter. Projects should be assessed against their milestones. Projects that are significantly behind plan, or whose strategic rationale has weakened, should be considered for termination or restructuring - not just extended.

Kill decisions: the hardest portfolio discipline is killing projects that are not working. The psychological cost of admitting failure - to sponsors, to teams, to leaders - creates enormous pressure to continue investing in struggling projects. The portfolio framing helps: a kill decision is not an admission of failure, it is a reallocation of resources from a project that is not meeting expectations to one that might. Bashir created a quarterly "portfolio health meeting" at which every active project was reviewed against a simple traffic-light status. Red projects were discussed openly, with a clear question: what would need to change for this to become green, and is that change realistic?

Metrics for Innovation Portfolios

Innovation portfolios should not be measured by operational metrics alone. A portfolio that produces three Horizon 1 successes every year looks great on a short-term ROI dashboard but may be hollowing out the organisation's long-term competitive position by neglecting transformational bets.

Measure the portfolio across three dimensions: *health* (is the balance across horizons appropriate?), *velocity* (how quickly are projects progressing from idea to production?), and *value* (what is the measurable impact of projects that have reached production?).

>
"A portfolio with no failures is a portfolio that took no real risks. That is not an innovation portfolio. That is a list of safe bets." - Bashir Farahani, three years into building his bank's AI portfolio

Key Takeaways

• Manage AI initiatives as a portfolio, not a collection of independent projects. Portfolio thinking asks whether each new initiative makes the overall portfolio better - not just whether it is worthwhile in isolation.
  - The three-horizon framework provides a practical classification tool. Incremental improvements (Horizon 1), new capabilities (Horizon 2), and transformational bets (Horizon 3) require different resourcing, timelines, and success metrics.
  - Distributing resources evenly across all projects funds none of them adequately. Portfolio resourcing should follow conviction - give high-priority projects enough resource to move at pace, and keep exploratory projects appropriately small.
  - A managed pipeline prevents portfolio drift. Standardised intake, quarterly reviews, and explicit kill criteria maintain the discipline that informal review cannot sustain.
  - Kill decisions are portfolio management, not failure. Projects that are underperforming against clear criteria should be terminated and resources reallocated. Frame this as portfolio stewardship, not individual project failure.
  - Measure health, velocity, and value - not just ROI. A portfolio with no failed projects has taken no meaningful risks. Some failure rate is evidence that the portfolio is genuinely innovative.