Rushing into AI without defining what it should actually do is one of the most reliable ways to burn budget and end up with nothing useful to show for it. The problem isn’t AI itself – it’s that “AI” has become a label slapped onto everything from basic scripts to sophisticated language models, and that vagueness causes real problems. This isn’t an anti-AI argument. It’s a pro-clarity one.
Here’s what happens when businesses skip the definition step – and how to avoid it.
When “AI” means everything, it means nothing
The term “AI” currently covers an enormous range of technologies – rule-based automation, recommendation engines, machine learning models, large language models like GPT-5.5, and fully autonomous AI agents. These are fundamentally different things with different costs, capabilities, and failure modes. Yet they all get called “AI” in boardrooms and budget meetings.
Think of it like calling everything with an engine a “vehicle.” Technically correct, but not helpful when you’re trying to decide whether you need a forklift or a company car.
In practice, this vagueness creates three concrete problems for anyone trying to make decisions:
- For buyers/clients, it makes vendor evaluation nearly impossible – you can’t compare solutions if you don’t know what problem you’re solving
- For internal teams, it creates confusion about what’s expected and how success will be measured
- For compliance, it makes it nearly impossible to assess risk, accountability, or regulatory fit
In Switzerland specifically, where compliance expectations are high and SMEs operate under tight regulatory scrutiny, “we’re implementing AI” is not an adequate answer to auditors, data protection officers, or procurement reviewers. Definitional vagueness isn’t academic – it creates real legal and financial exposure.
When a new client says “we need AI,” our team at what. always asks the same set of clarifying questions before anything else:
- What specific task or decision are you trying to improve or automate?
- Where exactly does the current process break down or slow down?
- Does this require judgment and adaptability, or is it a predictable, rule-based task?
- How will you measure whether it’s working?
- Who owns the outcome if it goes wrong?
These aren’t gatekeeping questions. They’re the foundation of any project that actually ships.
Why vague AI goals sink projects before they start
Companies often adopt AI because it feels urgent – competitors are “doing AI,” leadership is asking about it, and nobody wants to be left behind. That’s understandable. But adopting AI because it’s fashionable, rather than because you’ve identified a specific problem it solves, is how budgets disappear with nothing to show for them.
Without a clear definition of what the AI is supposed to do, three things become genuinely impossible:
| What breaks | Why it matters |
|---|---|
| Data strategy | You can’t identify what data you need, whether you have enough, or how to evaluate model quality without a defined use case |
| Integration | Fitting AI into existing workflows becomes a moving target – pilots stall, handoffs break, nothing goes live |
| Governance | If you can’t define what the system is doing, you can’t assess whether it’s compliant, explainable, or fair |
This connects directly to something we’ve written about before: the “garbage in, garbage out” principle. As we explored in Before You Touch AI, Fix Your Workflows First, AI built on broken or undefined processes doesn’t fix those processes – it just makes the confusion run faster.
What a useful AI definition looks like
Instead of arguing about what AI “is” in the abstract, the more productive question is: what does it need to do for your business, and does it actually require AI to do it?
Before starting any AI project, work through this checklist:
- What specific task should this system perform? (One task, defined precisely)
- Does it need to learn, adapt, or handle ambiguous inputs – or would a simpler rule-based automation do the job just as well?
- How much human oversight is needed, and at what points?
- How will success be measured, and within what timeframe?
- Is this meaningfully different from standard automation?
That last question is the most important one. If the honest answer is “not really” – that’s not a failure. It’s a signal to use the simpler, cheaper, more reliable tool.
This matters especially in the Swiss context. Labour costs here are among the highest in the world. Every poorly-scoped AI project is expensive twice over: once in direct spend, and once in the staff hours lost to a tool that doesn’t work properly. Being precise upfront is not over-engineering – it’s basic financial discipline.
It’s also worth noting that many problems framed as AI problems are actually integration problems. When a client tells us their team is spending hours manually reconciling data between their CRM, ERP, and accounting system, they often assume they need AI to fix it. Usually, they don’t. They need their systems to talk to each other. Our tools integration work resolves the majority of these situations without any AI model involved – and does it faster and more reliably.
What our AI automation expert says
What our AI automation expert says
Most clients don’t have an AI problem – they have a clarity problem. Once we define the actual task, the right solution usually becomes clear, and it’s often simpler than they expected.
Define first, build second
The companies getting real results from AI aren’t necessarily the ones moving fastest. They’re the ones who were clearest about the problem they were solving before they committed to a solution.
A precise brief – one that defines the task, the success criteria, the data requirements, and the governance expectations – is worth more than any tool. AI automation becomes genuinely powerful once that foundation is in place. Not hype, not a demo that gets shelved after a month. Real leverage, applied to a well-understood problem.
If you’re unsure whether what you’re looking at is an AI problem, an automation problem, or an integration problem, that’s exactly the right question to explore before spending anything. Get in touch with our AI automation team – we’ll help you find the honest answer.
