In this article
Innovation Readiness Assessment
A staff survey that shows, department by department, where work stalls.
Assess Your OrganizationKey takeaways
Most AI projects fail for reasons in the organization, not in the model.
AI makes a working system faster and a broken one more confusing.
Name the problem without the word AI before you choose a tool.
Ask the people who do the work first. They decide whether a tool is used.
Judge an AI project by what changed in the work at 30 and 90 days.
Why AI Projects Fail: What the Research Shows
Definition
AI project failure
An AI project fails when it is abandoned, never reaches daily use, or reaches daily use without changing a result the organization can measure.
Most organizations now use AI in some form, and most boards ask the same question: what is our AI plan? Far fewer can show what AI has changed. This article explains why AI projects fail, where the causes sit, and what to fix before you add another tool.
93%
of Canadian business leaders surveyed said their organizations use AI in some form, up from 61% a year earlier
2%
said their organizations are seeing a return on their generative AI investments
Source: KPMG in Canada, survey of 753 business leaders, November 2025
Other studies point the same way. RAND reported in 2024 that, by some estimates, more than 80 percent of AI projects fail, twice the rate of information technology projects that do not involve AI. MIT’s Project NANDA, in preliminary findings published in July 2025, reported that 95 percent of the organizations it studied were getting zero return from generative AI, despite 30 to 40 billion dollars of enterprise investment. Each study defines failure differently, so the figures are not directly comparable.
What the studies share matters more than the numbers: the causes they name sit mostly in the organization, not in the technology. Of the 65 experienced data scientists and engineers RAND interviewed, 84 percent named failures driven by business leadership as the primary reason AI projects fail. The MIT authors wrote that the gap “does not seem to be driven by model quality or regulation, but seems to be determined by approach,” and that most enterprise tools failed on brittle workflows and a poor fit with day-to-day operations.
AI Does Not Fix a Broken Process
At Unbounded Thinking we see the same pattern in the organizations we work with. AI makes a working system faster and a broken one more confusing. A team that already decides well gets more done. A team with no owner for its ideas gets more ideas, more drafts and the same weak decisions.
Key idea
AI does not fix a broken operating model. Without a working system underneath, you are not transforming it. You are digitizing confusion.
When the work is fragmented, ownership is unclear and nobody can see what is working, an AI tool speeds those problems up. Reports are drafted faster and read no more often. Requests are sorted faster into a queue that nobody owns. From the outside it looks like progress, while the same questions remain: what is working, where value is created and what should change.
Signs that a process needs fixing before it gets an AI tool:
No owner. Nobody is accountable for the result the tool is meant to improve.
No common way. Each person does the work differently, so there is no single process for a tool to support.
No baseline. Nobody can say how long the work takes today or what it costs.
Scattered data. The information the tool needs lives in personal spreadsheets and inboxes.
Stalled decisions. Work waits on one person’s approval, and a faster draft will wait just as long.
The Five Causes Behind Most AI Project Failures
RAND’s interviews found five leading root causes of failure. In plain terms:
The wrong problem. Leaders misunderstand or miscommunicate what the AI should solve, so a model is built for the wrong measure or does not fit the way the work is done.
Not enough data. The organization lacks the data needed to train an effective model.
Technology first. The project chases the latest technology instead of a real problem for the people who will use it.
Missing infrastructure. There is no reliable way to manage the data and put a finished model into daily use.
A problem too hard for AI. The task is beyond what current AI can do.
Only two of the five are technical. The wrong problem, technology first and a problem too hard for AI are decisions people make before any work begins. That matches what we see: organizations rarely run short of ideas or tools. They lack the system underneath them, and they decide poorly before the work begins.
Boston Consulting Group made a similar point in its 2022 rules for AI in business: about 10 percent of the effort goes into the model, 20 percent into data and technology, and 70 percent into new business processes and the way business functions operate. A plan that spends most of its attention on the tool has those proportions reversed.
In practice. Before approving an AI project, ask for the problem in one sentence that does not use the word AI. If no one can write it, the project is starting from the tool.
Before AI, Start With Your People
The people who do the work know where the time goes, which steps nobody needs and where a tool would help or get in the way. They are also the people who decide, every day, whether a new tool is used or quietly worked around. Asking them first is the cheapest research an AI project can do.
Instead of
Choose a tool, then look for a use
Ask whether teams are using AI tools
Pilot with the most enthusiastic team
Count licences and logins
Try this
Start from a problem people name in their own work
Ask whether the organization is ready
Pilot with a team that will live with the result
Count what changed in the work
The people who do the work know where a tool would help.
AI can draft, sort and suggest. People notice the problem, judge what matters, decide and adopt the change. The difference between an idea and an innovation is adoption, and that holds for an AI tool as much as for any other change. In our training we call the distance between a tool that works and a tool people use the Innovation Gap.
Adoption follows the same rules as any change at work: it goes better when it happens with people rather than to them. Our article on how to get employee buy-in for change sets out the steps.
What to Fix Before You Add Another AI Tool
The steps below come before the choice of tool. Most take days rather than months, and each one lowers the risk of the project that follows.
1
Name the problem in plain words
Write one sentence: who has the problem, what it costs today in hours, errors or delays, and how you will know it is solved. Leave the word AI out of it.
2
Fix the process before you speed it up
Map the work as it is done today, remove the steps nobody needs and agree who owns the result. Then decide where a tool would help.
3
Ask the people who will use it
Bring three to five people who do the work into the choice. Ask where their time goes, what they would hand to a tool and what they would never trust a tool to do.
4
Measure before you change anything
Record how long the work takes and what it costs now. Without a baseline, nobody can show later what the tool changed. Count time before money: hours saved show first.
5
Test small, with the team that will keep it
Try the tool on one real task for a few weeks, with one owner, in the team that will use it every day. A test that works only for enthusiasts says little about everyone else.
6
Check use and results at 30 and 90 days
Agree the dates in advance. Look at whether people use the tool when no one is checking and what changed in the work. Then scale it, change it or stop it.
Step 5 is where many AI pilots stall. Our article on why pilot projects do not scale explains why the readiness of the team that receives a pilot matters more than the pilot itself.
How to Tell Whether Your Organization Is Ready for AI
Readiness for AI is mostly readiness for change. Before an AI project starts, check that each of these is true.
Before an AI project
The problem is written in one sentence without the word AI.
The process the tool will support has an owner and works without it.
People who do the work helped choose the tool.
There is a baseline for the time and cost of the work today.
The first test is small enough to finish in weeks.
Dates are set to check use and results at 30 and 90 days.
A missing item is cheapest to find before the purchase. To see where your organization stands before you add another tool, take the free Innovation Readiness Snapshot.
In the Age of AI, People Still Do the Work That Matters
The organizations that get the most from AI are rarely the ones with the most tools. They are the ones that already hear from their people, decide well and keep what works. AI makes that system faster. Without it, AI makes the confusion faster.
That is the case for human-centered innovation: start with the people and the work, then add technology where it helps. Our page on Human-Centered Innovation Management explains how we build that system with organizations.
Innovation Readiness Assessment
Measure Innovation Readiness in Every Department
Staff across every department assess process, people, projects and performance. Leadership receives a department-by-department view of where work stalls, with a prioritized 30- and 90-day action plan.
Frequently Asked Questions
Why do AI projects fail?
Most AI projects fail for reasons in the organization rather than the technology: the wrong problem, a tool chosen before the problem, a broken process, no owner and too little say for the people who will use it. Missing data and infrastructure are the main technical causes.
What is the AI project failure rate?
Estimates vary with how failure is defined. RAND reported in 2024 that, by some estimates, more than 80 percent of AI projects fail. MIT’s Project NANDA found 95 percent of the organizations it studied getting zero return from generative AI, and KPMG found 2 percent of Canadian organizations seeing a return on it.
Why do AI pilots fail to scale?
A pilot usually runs with an enthusiastic team, extra attention and a simplified process. When it moves to the teams who must use it every day, their workflow, ownership and readiness decide whether it survives. MIT’s 2025 findings point to brittle workflows and a poor fit with day-to-day operations.
What is the human side of AI adoption?
It is everything that decides whether people use an AI tool once it works: whether they helped choose it, whether it fits how they work, what they lose when it arrives and whether anyone checks that it is used and useful.
How do you prepare employees for AI?
Ask them first. Involve the people who do the work in naming the problem and choosing the tool, fix the process before adding the tool, train people on their own tasks, and check use and results at 30 and 90 days.
How do you know if your organization is ready for AI?
An organization is ready for an AI project when the problem can be stated without the word AI, the process has an owner and works without the tool, the people affected have been involved, there is a baseline and the first test is small.
Limited-time Q4 offer
Innovation Process Certification at 50% Off
Your team can build this on a real problem in our two-day Innovation Process Certification, for up to 10 people. Alberta organizations that book by 30 November pay half our medium business price.
About the author
Shannon Phillips
Co-Founder and Innovation Strategist
Designs structured innovation systems for enterprise organizations and helps leadership teams set priorities that reflect how the organization operates.
Related Articles
Adoption and Change
Why Pilot Projects Do Not Scale: The Readiness of the Team That Receives Them
Adoption and Change
How to Get Employee Buy-In for Change: Do It With People, Not to Them
Knowledge Hub