It's tempting to frame the fraud-technology conversation as rules versus AI, with rules cast as the outdated approach on its way out. That framing is both wrong and unhelpful. Rules are fast, explainable, and excellent at catching known, well-defined patterns — which is exactly why every mature fraud stack still runs on them.
Where Rules Remain Valuable
A rule is the right tool when the pattern is known, stable, and easy to state: a transaction above a threshold from a new device, a payment to a sanctioned entity, a velocity breach against a hard limit. Rules are cheap to reason about and easy to audit — real advantages a black-box model doesn't automatically have.
Where Rules Become Insufficient
Rules struggle with anything that requires context: a transaction that's only suspicious because of who the beneficiary is connected to, a session that's only suspicious because of how it deviates from one customer's own baseline, a ring of accounts that's only visible when viewed as a network rather than one at a time. That is where behavioural analytics, machine learning, and network analytics each add a layer rules cannot express on their own.
Rules remain an important layer. They should no longer be the only layer.
The credible position isn't 'replace your rules' — it's 'stop expecting your rules to do a job they were never designed for.' Layer behavioural and network intelligence on top, and let rules keep doing what they're good at.