On this page
Most AI projects fail before the build even starts. The decision to skip a proper scoping phase is where the money disappears, not in the model, not in the infrastructure, and rarely in the code itself. The most common failure driver is misaligned purpose: leaders and technical teams do not agree on the problem the project is meant to solve, so success is never clearly defined and accountability never lands. That is a scoping problem. It is diagnosable, and it is fixable, but only before you commission a build.
How Bad Is the Problem, Really?
More than 80% of AI projects fail to deliver intended business value, roughly twice the failure rate of comparable IT projects without AI, according to RAND Corporation's 2024 study. For generative AI specifically, 95% of organizations see no measurable P&L return from their pilots, with only about 5% capturing value at scale (MIT Project NANDA, 2025). Those numbers have not budged despite better tools and growing in-house expertise. 84% of AI project failures are primarily attributable to leadership decisions rather than technical limitations, according to RAND Corporation. Picking a model is a technical decision. Defining what problem you are solving, confirming the data exists to solve it, and agreeing on what "done" looks like, those are leadership decisions. They happen in the scoping phase, or they do not happen at all.
What a Scoping Phase Actually Does
A discovery phase is not a planning exercise. It is a diagnostic. The goal is to answer four specific questions before any budget is committed to building:
1. What is the actual problem? Not the symptom, not the feature request, the business problem. A team that says "we need an AI chatbot" has named a solution. The scoping phase works backwards to find out what problem they are actually trying to solve and whether a chatbot is the right answer.
2. Does the data exist to solve it? Organizations consistently underestimate the quality, access, and governance work that AI requires, and discover the gap only after committing resources. Scoping surfaces that gap in days, not months. If your data is missing, incomplete, or siloed, that is a finding, and it changes the project plan. See Clean Data = Smarter AI for a deeper look at what data readiness actually requires.
3. Is there an existing system to connect to? This is where a lot of proposals go sideways. One medical system came to us after an overseas firm quoted six weeks to connect to a data source. The other firm assumed an API existed. It did not. We looked at what was actually there and built a screen-scraping solution in two weeks. The six-week quote was not dishonest, it was built on an assumption nobody checked. Scoping checks those assumptions.
4. What does success look like in numbers? Organizations that report significant financial returns from AI are twice as likely to have redesigned workflows before selecting modeling techniques, according to McKinsey (2025). That redesign starts with agreeing on a metric. Without one, you cannot tell whether the project worked.
What Skipping It Actually Costs
The instinct to skip scoping is understandable. Budget is approved, the team is ready, and any delay feels like friction. At the start of many projects there is immense time pressure, the budget is finally approved, the business department has been waiting for months, and there is a strong temptation to shorten the analysis phase. The cost of that shortcut shows up later. Many budget overruns are not created during execution. They are built into the project foundation from the start. As software engineering researchers Barry Boehm and Victor Basili demonstrated, errors detected in later project phases are exponentially more expensive to fix than those caught during requirements analysis, in many cases, ten to a hundred times more costly. For AI projects, that multiplier is steeper because model behavior is harder to reverse than traditional code. A chatbot trained on the wrong data, or an automation built around a misunderstood workflow, does not just fail to work, it produces confident wrong outputs. Fixing it means going back to the data, back to the architecture decisions, and sometimes back to the problem definition itself. Pilot purgatory is when AI projects get stuck between technical validation and production deployment. MIT research shows 88–95% of AI pilots never reach production, projects demonstrate initial promise in controlled environments but fail to scale due to organizational readiness gaps, weak business alignment, or technical debt. Most of those gaps are diagnosable before the pilot starts.
What a Good Scoping Phase Produces
A scoping engagement should end with documents you can actually use, not a slide deck of observations. Here is what that looks like in practice:
- A written problem statement that the technical team and the business owner have both signed off on.
- A data audit confirming what data exists, where it lives, what format it is in, and what cleaning or governance work is needed before a model can use it.
- An integration map showing every system the AI needs to touch, how it will connect, and where API coverage ends and other approaches begin.
- A definition of success with specific metrics — not 'improve efficiency' but '30% reduction in manual review time on X workflow by Q3.'
- A realistic timeline and cost range that reflects what was actually found, not what the proposal assumed.
That last item matters more than it sounds. The build vs. buy vs. partner decision looks completely different once you know what your data situation actually is, which systems are involved, and how tightly the workflow needs to be controlled. You cannot make that call from a blank intake form.
How Long Does Scoping Take?
For a focused AI initiative at a startup or SMB, a proper scoping engagement typically takes one to three weeks. The output is a technical brief, a data readiness assessment, and an architecture recommendation with a costed build plan. That timeline can feel like a delay when you are eager to start. Weigh it against the alternative. S&P Global's 2025 survey found that 42% of companies abandoned most of their AI initiatives, up from 17% in 2024. Those abandoned projects all had launch dates. They did not have scoping phases. For longer-term engagements, a partner who knows your systems and your data history scopes faster each time. A technology relationship we have maintained since 2018 is a good example of this, by year two, the discovery work for new initiatives shrank significantly because the foundational questions had already been answered. New features, new workflows, new AI layers: the scoping is faster because the context already exists. If you are thinking about incorporating AI into an existing application, the diagnostic work is the same regardless of project size. The question is not whether to do it. The question is how much it costs to skip it.
Not sure where to start? A scoping conversation costs nothing. Talk to us