In a workshop a department head asked us, half joking: “And where am I in this diagram?” We pointed to a node in the middle of the process. He was visibly relieved — and that is exactly what this is about today.

The loop is not a stopgap

Human-in-the-loop means that an automated process seeks human judgement at defined points: for approvals above a threshold, for unusual cases, for anything that requires fine judgement. For a long time this was seen as a transitional solution — what you do until the AI is “good enough”. The practice of 2025 tells a different story: the most successful automations are built as hybrids, on purpose and for the long term. Not because the technology is too weak, but because responsibility, liability and customer trust need human decisions at the right places.

Where humans belong — and where they don't

The art lies in placement. Putting people where they merely nod things through creates alibi controls and frustration. Deploying people where their judgement makes the difference — edge cases, exceptions, delicate communication — creates quality and acceptance. A good test: if the reviewer simply clicks “OK” in nine out of ten cases, the threshold is set wrong. If they have to intervene in every second case, the agent is not mature yet. In between lies the sweet spot, and it shifts over time — which is why it should be reviewed regularly.

Acceptance is not a side issue

Perhaps the greatest benefit is cultural: teams who know that the final decision in critical moments rests with them approach automation more openly and with more curiosity. The loop is therefore also a promise: the AI works for you, not around you.

Where in your processes the human is indispensable — and where they only click out of habit today: we are happy to map that with you.