AI Strategy & Business Case

Managing AI in the Project Portfolio

Marin Niehues9 March 202612 min read

Two colleagues planning with sticky notes on a whiteboard

You manage AI in the project portfolio by treating every application as an investment with a named owner in the business and a KPI, and by re-weighting the portfolio every quarter. Most companies have yet to take this step: they use AI in many areas but achieve a measurable impact on results in only a few applications. Typically, initiatives start in individual units, are funded from separate budgets and end months later as pilots that nobody operates. This article shows how executives and divisional heads place AI in the project portfolio, by which rules they allocate funds and how they measure success.

What is AI portfolio management?

AI portfolio management is the joint steering of all AI initiatives in a company as a single set of investments, on which the executive team decides at fixed intervals using uniform criteria. It answers three questions: which initiatives are running, what do they cost in operation, and what impact do they show in the business metrics?

A classic IT portfolio holds initiatives with a defined scope and an end marked by an acceptance date. AI applications follow a different course. Their benefit appears only in live operation, their quality fluctuates with data and usage, and after launch they generate running costs that appear in no project plan. Treating AI applications like an implementation project means budgeting the effort up to the pilot and losing control afterwards.

Dispersion adds to the problem. A survey by the German Federal Statistical Office shows that in 2025, 26 percent of companies with 10 or more employees already used AI, rising to 57 percent among companies with 250 or more employees. AI ideas arise in many units at once. Each unit selects tools according to its own needs, and the executive team learns of many initiatives only when a license comes up for renewal. A portfolio creates a shared view of the status of initiatives and their impact. Its guiding question is which applications reach operation and remain effective there. How the portfolio fits into AI management as a whole is set out in the article AI management: what executive leadership must steer.

Four portfolio classes with their own rules

A single yardstick for all AI initiatives leads to poor decisions, because an experiment and a core process must be judged by different criteria. A division into four classes gives each group its own expectations of benefit and risk.

  • Foundations: platform and data access that several applications use. The measure is reuse, meaning the number of applications running on the same foundation.
  • Efficiency: applications that make an existing process faster or cheaper. The measure is the benefit per transaction.
  • Growth: applications that generate revenue or retain customers. The measure is the contribution to a business metric.
  • Exploration: limited trials with a fixed budget and an expiry date. The measure is what the organization has learned and whether an initiative for one of the other classes emerges.

Each class receives a share of the total budget, set by the executive team once a year. This prevents the loudest requests from dominating the portfolio, and it shows whether the organization is investing in efficiency or in growth.

These ranges are starting values for the first budget year. From the second year, the executive team shifts the shares according to the measured results of each class.

How much budget should go into exploration?

Reserve 10 to 15 percent of the AI budget for exploration, divided into fixed amounts per trial with an expiry date of no more than three months. The range follows the 70-20-10 rule of thumb common in innovation management, under which most funds go to the core business and a small, protected share goes to new ventures.

Two criteria determine the choice within the range. If the company has no AI application in operation yet, the lower limit applies, because foundations and first efficiency applications take priority. If three or more applications already run reliably, the upper limit is appropriate, because the organization can then move insights from trials into operation quickly.

The rule for ending a trial is decisive. An exploration initiative ends on its expiry date with one of two decisions: it moves with a business case into the efficiency or growth class and passes through the gates there, or it is closed with a short record of what was learned. An extension without a change of class is ruled out.

Gates instead of open-ended funding

Staged release of funds is the most direct remedy for growing graveyards of pilots, because it ties funding to proven results. Every application passes through gates at which the executive team, or a steering group it appoints, decides. Funds flow up to the next gate. An initiative that misses a gate is adjusted, stopped or moved to another class. All three outcomes count as the result of good management.

The four gates follow the path of an application from its first description into operation. At the first gate, the question is whether problem and benefit are described robustly. At the second, a prototype demonstrates the benefit in the real process with real data. At the third, it is decided whether the business unit adopts the application and a team takes over operation. At the fourth, the executive team checks after an agreed period whether the application reaches its KPI. What a robust business case for these decisions looks like is covered in AI business case: calculating benefits and costs reliably.

KPIs that make a portfolio manageable

A portfolio needs a few KPIs, collected in the same way for every application. A set on three levels has proved its worth.

At application level, usage rate and output quality count: how many of the intended users work with the application regularly, and how often must a person correct the result? At business level, the metric the business unit already owns counts, such as lead time or closing rate. At portfolio level, the share of initiatives that reach operation counts, together with the ratio of running costs to realized benefit.

The benefit must become visible in the business unit's own metric. A separate AI statistic convinces nobody in the budget discussion, whereas an improved lead time in the business unit does, because the unit endorses the result itself.

Cadence and decision rights

A portfolio lives on rhythm. A quarterly review with the executive team and divisional heads has proved its worth, in which all applications are discussed on the same one-page format. Between reviews, a portfolio office maintains the data and prepares the gates with short decision papers.

Decision rights must be clarified in advance. Who releases funds? Who may stop an application? Who decides in conflicts over scarce specialists? Without these rules, every decision slips to the next meeting, and the portfolio grows without anyone being accountable for it. A proven allocation assigns the release of funds to the executive team, the acceptance of the benefit to the business unit, and operation to the team that will later run the application.

Which tools are suitable for AI portfolio management?

Up to around 20 initiatives, a well-maintained spreadsheet or the existing ticketing system is sufficient; beyond that, extending the existing project portfolio management software with an AI label pays off. A company generally needs no dedicated AI portfolio software, because AI initiatives compete with all other investments for the same funds and should appear in the same view.

The fields a tool holds matter more than the tool itself. The following checklist shows what every entry in the portfolio must contain as a minimum.

Niehues Consulting as a partner for the AI portfolio

Niehues Consulting combines the tasks an AI portfolio requires in a single advisory firm: AI management with business analysis, business case and prioritization, the development of prototypes with transfer into operation, and delivery management with organizational design. Marin Niehues comes from the delivery perspective and, in projects ranging from VC-funded scale-ups to DAX 40 groups in 20+ countries, has set up organizations so that initiatives arrive in operation.

The basis is the Niehues Delivery Model with its phases understand, build and introduce. Three of its test questions matter most for the portfolio. Which business outcome is to change measurably? This question decides at gate 1 whether an initiative enters the portfolio at all. Has the business case been confirmed with real values? It decides on further release at gate 2. Who is accountable for operation from day one? It prevents a successful prototype from being left without ownership at gate 3.

The management model is defined together with the executive team and tested on running initiatives before the portfolio office takes it over. The entry point is a free 30-minute initial consultation, in which you clarify the state of your portfolio and the first steps.

Frequently asked questions

How many AI initiatives should a portfolio contain at once?

A portfolio should contain only as many initiatives beyond gate 2 as operating teams and business units can take over, as a rule no more than one initiative in introduction per operating team. If more are running, pilots queue up in front of the third gate. The number of exploration trials, by contrast, is limited by the class budget.

Who should be accountable for the AI portfolio?

Overall accountability rests with the executive team or a member it appoints. Operationally, a portfolio office maintains the data and prepares decisions. The business unit that uses an application is accountable for its benefit.

Does AI belong in the existing project portfolio or in a separate one?

AI belongs in the existing project portfolio, with its own label and the four classes. AI applications then compete with other investments for the same funds, and the executive team sees dependencies on systems and data. A separate portfolio hides these dependencies.

How often should the portfolio be reviewed?

The portfolio as a whole is reviewed every quarter. The steering group decides individual gates in between as soon as an initiative meets the criteria, so that good initiatives do not wait for the next date. The executive team resets the budget shares of the classes once a year.

How do you prioritize AI projects?

Prioritize AI projects by value, feasibility and risk, assessed with the same criteria for all initiatives in a class. Value is measured by the business KPI, feasibility by data and available specialists, risk by legal and operational exposure. A detailed assessment scheme is offered in prioritizing AI use cases.

What is a stage-gate process for AI projects?

A stage-gate process releases funds for an AI project only in stages, each up to the next decision on continuation, adjustment or termination. For AI, the process needs a fourth gate after launch, because the benefit only shows in operation. Predefined test questions apply at every gate.

Why do AI projects get stuck in the pilot phase?

AI projects get stuck in the pilot phase when nobody has taken over operation before launch and no gate decides on continuation. Often there is also no operating budget and no KPI against which the business unit can measure the benefit. Causes and remedies are described in why AI pilots fail to reach production.

Let’s talk about your initiative.

In 30 minutes we clarify where your organization stands and which step will pay off most now.

Request a consultation

30 minutes · free · no obligation