AI Initiative Triage: A CEO's Decision Guide

Your AI budget probably grew this year. So did the number of AI projects that went nowhere.

According to BCG’s 2026 AI Radar, corporations expect to double their AI spending from 0.8% to 1.7% of revenues. But S&P Global found that 42% of companies abandoned most of their AI initiatives in 2025, up from just 17% the year before. More money in, more projects killed. The problem isn’t AI itself. It’s how companies pick and manage AI projects.

Three out of four CEOs now say they are their organization’s main AI decision-maker, according to BCG. If that includes you, the AI portfolio is your portfolio. And like any portfolio, it needs regular triage.

Why AI Spending Doubles While Results Don’t

The short version: companies are treating AI like a shopping spree instead of an investment strategy.

A typical mid-size company in 2026 has a handful of AI experiments running at once. A chatbot here, a content generation tool there, an analytics pilot in finance, maybe a document-processing agent somewhere else. Each one got approved on its own, usually because a department head made a convincing pitch or a vendor’s free trial quietly turned into a paid subscription.

No single project is the problem. The problem is that nobody manages the portfolio. Every initiative competes for data, attention, and budget, and nobody has a centralized view of what’s actually producing results.

RAND Corporation research backs this up: more than 80% of AI projects fail, roughly twice the failure rate of non-AI technology projects. The causes aren’t technical. They’re organizational: unclear objectives, poor data foundations, no defined success criteria, and nobody with the authority to kill projects that aren’t working.

The Three Categories Every AI Project Falls Into

When you look at your AI initiatives as a portfolio, every project falls into one of three buckets. Your job is to sort them correctly and act on the sorting.

Fund: Projects with a clear path to measurable value. They have a defined business problem, a baseline you can measure against, an accountable owner, and data that’s clean enough to work with. These get budget and attention.

Scale: Projects that proved their value in a pilot but need organizational investment to grow. The technology works, but now you need process changes, training, and data infrastructure to make it part of daily operations. The investment here is different: not money for tools, but time for adoption.

Kill: Projects with no realistic path to business value. Maybe the experiment didn’t pan out. Maybe they’re solving a problem nobody actually prioritized. Maybe the timeline keeps stretching without delivering results. These need to stop before they consume more resources and before they erode your team’s confidence in AI altogether.

Most CEOs are comfortable with “fund” decisions. “Scale” and “kill” are harder. Scaling requires patience and organizational change, and that doesn’t come naturally when you want fast results. Killing requires admitting a project didn’t work, which can feel like admitting a bad decision. But you need to do both.

How Do You Know Which AI Projects to Kill?

Look for these five signals. Any two of them together should trigger a real conversation about whether the project deserves another quarter.

No baseline was ever established. If you can’t answer “what was the metric before AI, and what is it now?” then you can’t prove value. Projects without baselines are unfalsifiable. They can always claim they’re “helping” without evidence.
No process owner. An AI tool without someone responsible for its outcomes is an orphan. Somebody in IT set it up, but nobody in the business owns the result. These projects drift until someone notices the subscription renewal.
The data foundation is missing. According to a 2026 survey by Writer, 73% of organizations report data quality as their biggest AI implementation challenge. If the data feeding your AI project is incomplete, inconsistent, or siloed, the AI will produce unreliable outputs. Better algorithms won’t fix bad inputs.
The ROI timeline keeps extending. First it was “we’ll see results in Q2.” Then Q3. Then “we need another six months.” When a project keeps pushing its value delivery date forward, the initial hypothesis was probably wrong, and the team hasn’t adjusted.
It’s solving a problem nobody prioritized. Some AI projects exist because someone saw a demo and got excited. The technology is interesting, but the business problem it addresses wasn’t on anyone’s priority list before the vendor showed up. If nobody was losing sleep over this problem before AI entered the picture, question whether you should be spending money on it.

What Separates AI Projects That Scale

Roughly 67% of organizations remain stuck between pilot and scaling phases, unable to turn promising results into repeatable business processes. Getting past that gap takes more than technology spending. It takes organizational investment.

The AI projects that actually move from pilot to production tend to share a few things:

They solve a decision, not a task. A task-focused AI project automates something a person used to do: data entry, report generation, classification. A decision-focused AI project changes how people make choices. The first saves time. The second changes outcomes. Decision-focused projects scale better because better decisions lead to better results, which generate better data, which improve the AI over time.

They have executive sponsorship that lasts past launch. Plenty of AI pilots get C-suite attention during the approval phase. Few keep that attention during the messy adoption phase, when users need training, processes need redesign, and early results look underwhelming. The CEO’s role here isn’t to manage the project. It’s to keep removing obstacles and keep reinforcing why the change matters.

They connect to data people trust. In our experience with mid-size businesses, the single biggest predictor of AI project success isn’t how sophisticated the model is. It’s whether the team already trusts the underlying data. If your finance team doesn’t trust the numbers in your ERP, they won’t trust an AI that reads those same numbers.

They replace a workflow, not just a tool. When AI scaling works, the old way of doing things stops. If people still maintain the spreadsheet “just in case” alongside the AI tool, adoption hasn’t happened. Real scaling means the AI is the process, not a layer on top of it.

The CEO’s AI Portfolio Review

You already review financial performance quarterly. Your AI portfolio deserves the same treatment. Here’s a straightforward framework for a quarterly AI review.

For each active AI initiative, ask:

What specific business metric is this project improving?
What was the baseline, and what’s the current performance?
How many people actually use this daily (not “have access to it,” but use it)?
What’s the total cost this quarter, including people’s time?
What needs to happen to scale this to the next level?

Then sort into actions:

Double down on projects with demonstrated impact and a clear scaling path
Set a deadline for projects showing promise but still in pilot. Give them 90 days to hit a specific threshold
Kill projects that can’t answer the first two questions. If you don’t know the metric or the baseline after a quarter, you won’t know after two

Keep the portfolio small. Mid-size companies that succeed with AI rarely run more than two or three initiatives at once. Spreading attention across six or eight projects means none of them get the organizational support they need.

When AI Becomes a Competitive Advantage

Most companies still use AI to cut costs: fewer hours on data entry, faster report generation, less manual processing. That’s a reasonable starting point. But BCG’s research suggests a bigger strategic shift: “Leaders who use AI only to cut costs will eventually be outlearned by leaders who use AI to increase enterprise intelligence.”

Cost-cutting AI produces a one-time savings. Intelligence-building AI produces an advantage that grows. When your team makes better decisions because AI surfaces patterns they couldn’t see in spreadsheets, that edge compounds over time. Your pricing gets sharper. Your forecasts get more accurate. Your resource allocation gets tighter.

The CEO’s role in this shift is to move the AI conversation from “how do we save money?” to “how do we make better decisions?” That doesn’t mean abandoning ROI measurement. It means expanding what you measure. Track decision speed. Track forecast accuracy. Track whether your team asks different questions now than they did before AI.

The companies that turn AI into a lasting competitive advantage aren’t the ones that spend the most. They’re the ones that triage honestly, scale deliberately, and measure what actually matters.

Frequently Asked Questions

How many AI initiatives should a mid-size company run at once?

Two to three active initiatives is the practical ceiling for most mid-size businesses. Each AI project needs executive attention, data infrastructure, and change management support. Spread those resources across too many projects and you get shallow adoption everywhere, deep adoption nowhere.

What percentage of AI projects fail?

RAND Corporation research found that more than 80% of AI projects fail, roughly twice the failure rate of non-AI technology projects. The main causes aren’t technical. They’re organizational: unclear objectives, missing data foundations, no defined success metrics, and nobody with the authority to stop underperforming projects.

How long should an AI pilot run before you expect results?

Ninety days is a reasonable window for a well-scoped AI pilot to demonstrate measurable impact. If a pilot can’t show improvement against a baseline metric within one quarter, it’s worth questioning whether the problem, the data, or the approach needs to change before investing further.

What’s the biggest barrier to AI adoption in mid-size companies?

Data quality is the most commonly cited barrier, with 73% of organizations reporting it as their top AI implementation challenge. But the real issue runs deeper: having clean data, defined processes, trained users, and a clear owner for each AI initiative. The technology itself is rarely what holds things up.

Should CEOs personally oversee AI projects?

CEOs should own the AI portfolio, not individual projects. That means setting strategic direction, reviewing performance quarterly, making fund-scale-kill decisions, and making sure the organization backs what you’ve decided to pursue. Day-to-day project management belongs with the team, but the CEO’s sustained attention signals that AI adoption is a real priority, not a passing interest.

How Pluto Fits Your AI Portfolio

One reason AI initiatives stall is that they demand too much infrastructure before they deliver any value. Months of integration work, custom data pipelines, and training cycles eat budget before anyone sees a result.

Pluto works differently. It connects to your existing ERP and lets you ask business questions in plain language. No separate data warehouse. No custom dashboards. No six-month implementation. Your team can start getting answers from your business data in days, not quarters.

That changes the math on triage. Instead of betting on a large, risky AI initiative with uncertain payoff, you start with something your team can use right away: asking questions about margins, receivables, shipment status, or resource utilization in the same way they’d ask a colleague. The value shows up in the first week, not the first year.

If you’re evaluating where your AI budget should go next, we’re happy to walk you through it.

The hardest AI decision a CEO faces isn’t which project to fund. It’s which one to stop. Get that right, and the ones you keep will finally have the resources they need to deliver.