Someone in your organization is about to propose an AI initiative. Maybe it is already on the roadmap. Maybe a vendor just sent a deck. Maybe your CEO read something on a flight and forwarded it to the whole leadership team with three question marks.
Whatever the trigger, the question is the same: is this worth doing?
Most organizations answer that question badly. They bring in the vendor, sit through the demo, watch the AI do something impressive in a controlled environment, and then make a decision based on how the demo felt rather than what the numbers say. Six months later they are wondering why the project stalled, why adoption is low, and why the business case never materialized.
There are three questions that, if asked before any AI investment decision, separate the projects worth pursuing from the ones that will waste your time and budget. They are not complicated. They are just rarely asked.
Question One: Does It Solve a Problem That Already Costs You Money?
The best AI investments replace existing costs. The worst ones create new capabilities that nobody asked for.
Before any AI initiative gets approved, someone needs to answer this: what does this problem cost us today, in dollars, per year? Not "it wastes time" or "it creates inefficiencies." A number. If you cannot put a number on the cost of the problem, you cannot evaluate whether the AI solution is worth its price.
A manufacturing company spending two million dollars a year on quality inspection has a clear target. An AI system that reduces defects by 40 percent pays for itself within months. The math is simple because the baseline is known.
The companies that get this wrong start from the AI capability and work backwards to the problem. "We have this AI tool, what can we use it for?" That question almost never leads to a good investment. The right direction is always: what is breaking and how much is it costing us?
When a vendor pitches you, your first question should not be "what does it do?" It should be "what problem does it replace, and what does that problem currently cost us?" If they cannot answer that, the conversation should end there.
Question Two: Can You Measure Success in 90 Days?
AI projects that require 18 months to show results almost never show results at all. The technology moves too fast. The team loses focus. The business priorities shift. By the time the original deadline arrives, the project is either quietly cancelled or redefined to make the results look better than they are.
The most successful AI deployments share a common trait: measurable impact within one quarter. Not full deployment. Not perfect performance. Clear evidence that the approach works.
A logistics company testing route optimization should see fuel savings in 90 days. A customer service team deploying AI should see resolution time improvements in 90 days. A fraud detection system should catch something real in 90 days. If the 90-day check-in produces nothing but explanations for why results take longer, that is not a timing problem. It is a signal that the project does not work.
If your vendor or internal team cannot commit to a 90-day milestone with a specific, measurable outcome attached to it, they are not confident in their solution. You should not be either.
Question Three: Who Owns This After Launch?
AI systems are not software installations. You do not deploy them and move on. Models drift as the data they were trained on becomes less representative of current reality. Edge cases accumulate. Performance degrades in ways that are invisible until they are embarrassing.
The question "who owns this after launch?" is not about maintenance contracts. It is about accountability. There should be a person, with a name, with budget authority, who is responsible for the ongoing performance of the system six months after go-live. Not a team. Not a department. A person.
If that person does not exist at the time of approval, the investment will fail regardless of how well the initial deployment goes. The vendor will be gone. The internal champion will have moved on to the next project. And the system will quietly degrade until someone notices.
Before you approve anything, identify the owner. Get their name in writing. Make sure they have budget and authority. That conversation will tell you more about whether this project will succeed than any vendor demo.
The Filter
Three questions. What does the problem cost today? What is the measurable 90-day milestone? Who is the named owner post-launch?
Projects that can answer all three clearly succeed far more often than those that cannot. Projects that cannot answer any of them should not receive funding regardless of how compelling the technology demonstration was.
The AI revolution is real and the opportunities are genuine. The difference between the organizations that capture that value and the ones that spend money on impressive demos is often this simple: they ask the right questions before they write the check.
Need a research brief on an AI opportunity your organization is evaluating? Get a full Arlo Report delivered to your inbox in under 10 minutes, starting at $15. Order at arlobottman.com/research