When the decision is a one-off
AI learns from recurring patterns. A project that happens once, choosing a vendor or a merger decision, for example, gives it nothing to learn from. The investment in building an AI system doesn't pay for itself on something that won't happen again.
When there isn't enough quality data
An AI system is only as good as the data it sees. A business with partial, disorganized, or too-sparse data will get unreliable answers from AI, sometimes delivered with misleading confidence. Sometimes the right first step is organizing the data, not deploying AI.
When the cost of a mistake is too high
On projects where clients need accessibility compliance, site terms, or a privacy policy, the first stage is always human: a proper review against current Israeli law, tailored to the type of client and site, and compliant with Israeli standards and, where relevant, international ones too. AI can help with initial research or drafting a first pass, but the real focus is genuine legal coverage for the client, to prevent problems down the road. That's exactly an example of a decision where the cost of a mistake is too high to leave to an automated system alone.
When a simple solution already solves it
My standing rule, which over time became something of a motto, is Keep It Simple. Where there's no real need for another third-party system, added complexity, or a new point of failure, it's better to avoid it. Sometimes a small change to the workflow solves the problem with no technology at all, and sometimes a narrow, targeted technical fix is enough, without building a whole AI system around it that isn't actually needed.
Quick check: 4 questions to ask before you start
Before you start planning an AI solution, it's worth running through these four questions. If the answer is 'yes' to more than one, there's probably a simpler alternative worth checking first.
| Question | If the answer is yes |
|---|---|
| Does this decision happen only once? | There's no recurring pattern to learn from, AI won't learn anything from it |
| Is the data partial, disorganized, or too sparse? | You need to organize the data first, not deploy AI on top of it |
| Could a mistake here cause legal or financial harm to the client? | You need human review with accountability, not just automated output |
| Is there already a simple process or fixed rule that solves this? | Better to use it than add an unnecessary AI system |
How I actually decide
In practice, the decision starts with a short, precise scoping call: what's happening today, what really needs to change, and what's realistic to do and in what order. From there the answer becomes clear: sometimes the conclusion is that AI is needed, sometimes simple automation, and sometimes just fixing an existing process. The goal is always the same: a solution the team can actually run day to day, not technology for technology's sake.