There's a conversation we have often with potential clients. They call us with a specific idea: "we want to apply AI to this process". We listen, ask questions, look at the data. And, in more cases than we'd like, we end up saying the same thing: this isn't the right fit.
It's not the right fit because AI won't solve the problem. Or because it'll solve it worse than what's already in place. Or because the cost of building and maintaining it will be higher than the benefit. We say it with technical and business arguments, not as a commercial stance.
This article is about that: the cases where AI shouldn't be deployed, why it's so hard to say it openly, and how to identify whether your case falls into any of these scenarios before investing time and budget in something that won't make a real difference.
Why it's so hard to say "no" to an AI project
We live in strange times. Pressure to "do something with AI" comes from everywhere: boards that read Forbes articles, competitors announcing initiatives on LinkedIn, agencies promising full transformations with three prompts. The question stops being "does this solve a problem?" and becomes "why aren't we doing it yet?".
In this context, a consultancy that says "it wasn't needed" sounds bad. It sounds like a lack of ambition, like not knowing. And yet, it's exactly the opposite: saying no requires understanding the problem better than whoever sells the default yes.
Our position is clear. AI is a very powerful tool for certain kinds of problems, and a terrible investment for others. Distinguishing between the two is the first serious step in a project; skipping it is like starting a renovation without checking the foundations.
The five cases where we don't recommend applying AI
Based on our experience with our own projects and reviews of others' approaches, these are the five scenarios where, time and again, we've seen that AI doesn't add value:
1. When the process already works well with deterministic rules. If you have a clear rules system covering 95% of cases that only fails in edge situations, adding AI almost always makes things worse. AI introduces unpredictability: it can make different decisions in similar contexts, requires constant monitoring and evaluation, and forces you to maintain a new kind of infrastructure. If a well-designed spreadsheet or a Python script with five conditions solves the problem, leave it that way.
2. When the volume doesn't justify the operational cost. Every call to an AI model has a cost: in time, in money, in latency. If the process runs fifty times a year, automating it with AI is like building a crane to move a box. The execution time saved won't compensate the time and money spent setting up and maintaining the system. There's a minimum frequency or complexity threshold below which it never pays off.
3. When you don't have your own data providing real context. One of AI's most attractive promises is that it understands your business. But it only understands it if we give it the right pieces. If your company doesn't have structured data, consolidated historical records, or processed documentation of internal knowledge, what you'll have is a system giving generic answers with your brand on top. That's not differentiation, it's wallpaper.
4. When the decision has significant regulatory or legal consequences. Deciding who gets a loan, who passes a compliance check, who ends up on a risk list. In regulated sectors like finance or auditing, there are decisions that require full explainability, traceability, and explicit acceptance of responsibility. Current AI models don't offer these guarantees natively. You can use AI as support (preparing data, detecting patterns, generating drafts), but the final decision must remain in human hands with explicit criteria. If the process you want to automate is the decision itself, not the support, AI isn't the answer.
5. When the real goal is "do something with AI" rather than solving a concrete problem. This is the most common case and the hardest to detect from inside the company. It appears when the project starts due to external pressure (board, press, competition) and not from a need identified by the operational teams. The result is usually a pilot that never leaves the testing phase, a cost no one wants to explain, and a team that loses confidence in technology. If you can't describe the problem in a single sentence before introducing the word "AI", you're not ready for an AI project yet.
How to know if your case fits any of these
When evaluating a project, we ask three preliminary questions. Apply them yourself before looking for budget:
Question 1: if AI ceased to exist tomorrow, would you still want to solve this problem? If the answer is no, the project isn't born from a real need. It's born from the trend. Wait.
Question 2: can you describe the expected outcome in a concrete metric? "Reduce invoice review time from 4 hours to 30 minutes" is a metric. "Modernise the department" isn't. Without a metric there's no project: there's wishful thinking.
Question 3: do you have someone internally who can validate whether the system works or not? An AI system in production needs follow-up, adjustments, and periodic technical decisions. If there's no one inside the company capable of judging whether the system does what it should, the project depends 100% on the vendor. That's risk, not autonomy.
Our position
At MindRise we've said no to several projects in the past year. Not because they weren't technically feasible to build, but because the client wasn't going to come out winning. This is, probably, the most profitable decision we've made: prioritising projects where AI makes a real difference, where our hours add measurable value, and where the client walks away convinced the investment was worth it.
The most valuable question you can ask yourself before starting an AI project isn't "how do I apply AI here?". It's "do I actually have a problem that AI solves better than any other available solution?". If the answer is yes, let's talk. If not, we're glad we discussed it anyway.
If you're unsure whether your case fits any of the scenarios in this article, get in touch. We'll give you an honest assessment in a first call, with no commitment. Sometimes the best consultancy is confirming that you don't need a consultancy.