The AI ROI problem is really a project portfolio problem

AI investment is entering its less glamorous phase.

The question is changing from “What can this model do?” to “Which of our AI initiatives deserves more time, money, and attention — and which one should we stop?”

That shift is visible in two recent Gartner announcements. A September survey found that only 22% of organisations had successfully scaled AI across multiple business units or adopted an AI-first approach, even as 85% of functional leaders planned to increase spending in 2026. Gartner also reported that high performers treated AI as a portfolio of value, tracking returns and reallocating or discontinuing underperforming initiatives. Read the Gartner survey.

On September 24, Gartner made the finance implication explicit: CFOs need realistic time-to-value expectations, a balance between quick productivity gains and longer-term outcomes, and more deliberate portfolio management. Read the finance update.

The headlines sound like a finance problem. In practice, they expose a project problem.

AI return cannot be managed after the work has started if the initiative has no named owner, no baseline, no delivery shape, no evidence plan, and no agreed point at which somebody decides whether to scale, change direction, or stop. The model may be impressive. The investment is still undefined.

The AI portfolio is already here

Most organisations did not announce an AI portfolio. They accumulated one.

One team added a drafting assistant. Another started an automation pilot. A product group began an internal evaluation. Finance subscribed to a tool. Engineering changed its workflow. Operations created a small experiment that now depends on a customer-facing process.

Each decision may have been sensible in isolation. Together, these initiatives consume delivery capacity, create dependencies, introduce operating costs, and compete for the same specialists. That is a portfolio whether or not it has been given a portfolio name.

A portfolio is not just a list of vendors or model names. It is a set of investments that can be compared because each one has a shared minimum record:

The record does not need to predict the future perfectly. It needs to make the current bet legible enough for a team to learn from it.

That is why AI ROI belongs in the project operating model. The return is shaped by the work around the model: requirements, integrations, process changes, testing, training, adoption, maintenance, and the people who must decide what to do with the output.

Orbyna's Project Management System is built around that connected record. Its live capabilities cover portfolios, roadmaps, goals, backlogs, timelines, dependencies, quality, time, reporting, and project knowledge. Those capabilities do not make an AI promise on their own. They give an organisation a place to define and manage the work that an AI investment creates.

A budget line is not a business case

“We spent 500 dollars on an AI tool” is a procurement fact. It is not an ROI calculation.

The real investment may include the time spent selecting a use case, preparing data, changing a workflow, integrating a system, reviewing outputs, training a team, handling exceptions, and maintaining the process after the pilot. Sometimes the licence is the smallest line in the business case.

The same is true of the return. A faster draft is not automatically a business result. The useful question is what changed because the draft arrived faster:

• Did a team complete more accepted work with the same capacity?
• Did cycle time fall without quality deteriorating?
• Did the organisation avoid an external cost or reduce rework?
• Did the initiative unlock a capability that another project can use?
• Did a customer, employee, or decision-maker receive a better outcome?

These are project questions because they depend on how the work was scoped and delivered. If the baseline was never recorded, the team will end up arguing about impressions. If the effort is not attached to the initiative, the organisation will report a tool subscription while hiding the labour required to make it useful.

Orbyna's issue estimates, time tracking, budgets, workload views, and project reports help keep delivery effort visible beside the intended outcome. The system will not replace a finance model or measure a vendor invoice that is outside it. It can make the operational denominator harder to lose.

Measure three clocks

AI initiatives often have one date: the date someone expects value. That is too thin for a useful portfolio review.

Track at least three clocks:

1. The delivery clock. How long does it take to scope, build, integrate, test, and release the change?
2. The value clock. How long until the intended business outcome can be observed with reasonable confidence?
3. The learning clock. How soon will the team have enough evidence to decide whether the original hypothesis is still worth pursuing?

The clocks do not move together. A small workflow change may ship in two weeks but need two months of usage data. A foundational data project may take longer to deliver while making several future initiatives possible. A pilot may fail quickly and still be a good investment if it prevents a much larger commitment to the wrong problem.

This is where project structure matters. A backlog can show the work required before a pilot is meaningful. A timeline can show dependencies and critical path. A sprint can contain a bounded learning goal. A report can compare committed work with completed work and scope changes. A retrospective can capture what the team learned before the next investment decision.

The point is not to turn every experiment into a bureaucracy. It is to stop using one calendar date as a substitute for a decision model.

Choose bets by evidence, not excitement

The most visible AI demo is rarely the same thing as the most valuable initiative. A portfolio review should make room for different kinds of bets without pretending they have the same time horizon.

One useful classification is:

• Near-term improvement: a bounded change with a measurable productivity, cost, or service target.
• Strategic capability: work that may take longer but changes what the organisation can offer or decide.
• Foundation: data, integration, knowledge, process, or security work that enables several future initiatives.
• Learning bet: a deliberately limited experiment designed to resolve an important uncertainty.

The category changes how the initiative should be reviewed. A near-term improvement may need a short value window. A foundation project may be judged by the number and quality of downstream opportunities it unlocks. A learning bet should have an explicit evidence threshold and a small enough scope that stopping is affordable.

Orbyna's portfolio rollups, project goals, milestone dates, dependencies, and reports give these bets a common operating surface. Leaders can ask not only which project is late, but which project is consuming capacity, which dependency is holding up several initiatives, and which outcome has not yet produced credible evidence.

That is a more useful conversation than ranking projects by how often they appear in an executive presentation.

Give every initiative a decision path

An initiative that can only move from “pilot” to “scale” has already made stopping feel like failure. That creates a predictable bias: teams keep reporting activity because the operating model has no respectable state for learning or pausing.

Define the decision path before the first build begins. It could be:

Hypothesis → Scoping → Pilot → Measuring → Scale / Change / Pause

The names can vary. The important part is that each transition has a reason and an accountable decision-maker.

For example, moving from Pilot to Measuring might require a working workflow, a named user group, a baseline, and a test plan. Moving from Measuring to Scale might require evidence of the target outcome, acceptable quality, a support owner, and an estimate of the ongoing cost. Moving to Pause should preserve the learning and explain what would need to change before the idea is reconsidered.

Orbyna supports custom workflows, allowed transitions, conditions, required fields, roles, permissions, and activity history. That lets the decision path live with the project rather than in a quarterly slide deck. Automation can also create follow-up work, notify an owner, or move an item when a defined condition is met, while the history keeps the change attributable.

The goal is not to automate investment decisions. It is to make the decision points visible enough that humans can make them deliberately.

Keep effort attached to the outcome

AI projects often become invisible when their work is spread across several teams. A data engineer prepares an integration. A product manager changes the process. A subject-matter expert reviews outputs. A project lead tracks adoption. Finance sees the licence. No single view shows the whole investment.

That is how an initiative can appear cheap while consuming a large amount of organisational capacity.

Keep the surrounding work attached to the same project record. Link the requirements, implementation issues, dependencies, tests, decisions, and follow-up actions. Assign the work to the people doing it. Record estimates and actual effort. Keep the current scope separate from the history of changes.

This does not mean every contributor needs to log every minute forever. It means the organisation should be able to answer a bounded question such as: “What did this AI pilot require from the team, what changed during delivery, and what evidence supports the next investment?”

In Orbyna, boards and backlogs show movement, timelines show coordination, issue history shows change, time tracking shows effort, and reports show delivery patterns. The value is not another dashboard. It is the ability to move from a portfolio question to the underlying work without starting a reconstruction exercise.

A practical project card for an AI bet

Before approving another AI initiative, create a project record that can survive beyond the excitement of the launch announcement.

• State the business outcome in observable terms, not only the model or feature being introduced
• Record the baseline and the measure that will show whether the outcome changed
• Name one accountable owner and the people who can approve scope or investment changes
• Define the first bounded slice of work, including what is explicitly out of scope
• List dependencies, required integrations, affected workflows, and the evidence needed for acceptance
• Set a realistic value window and an earlier learning checkpoint
• Define the conditions for scaling, changing direction, or pausing
• Track delivery effort and ongoing operating cost beside the expected return

That card is intentionally ordinary. It can describe an AI-assisted process, a new integration, a data foundation, or a conventional software improvement. The point is to make the investment comparable to other work without pretending every outcome can be reduced to one number on day one.

The portfolio becomes the strategy

The current news is not simply that organisations are buying more AI. It is that the cost of weak selection is becoming visible at the same time as the cost of experimentation is falling.

When initiatives are cheap to start, the organisation needs a stronger way to decide what deserves continuation. That does not require certainty before the first experiment. It requires a record that can distinguish a promising result from a busy project, a useful failure from an abandoned one, and a long-term foundation from a short-term productivity win.

AI ROI is therefore not mainly a dashboard problem. It is a project design problem.

Give every initiative an outcome, an owner, a delivery path, a baseline, a learning checkpoint, and a decision that can be revisited. Then connect those initiatives to the wider portfolio so capacity, dependencies, effort, quality, and evidence are visible together.

That is the operating discipline Orbyna's Project Management System is designed to support: one continuous record for direction, work, ownership, time, quality, evidence, and outcomes. Explore Orbyna or talk to us about a project portfolio.