The Ladder, Trust Before Automation

I've come to think of the relationship between people and automation as something closer to a courtship than a rollout, and I don't say that flippantly. You don't ask someone to trust you completely on the first date, and you shouldn't ask an operations team to hand over a decision to a system before you've earned the right to make that suggestion.

The way I sequence this for systems that advise people or take over meaningful operational decisions follows a pattern simple enough to describe over a pint. I do not mean that every automatic file transfer needs a five-stage courtship. A low-risk, well-tested mechanical step may simply deserve automating. This ladder is useful when I am asking people to trust a system with judgement they previously exercised themselves.

It starts with visibility, which is really just showing people something true about their own operation that they didn't have easy access to before. Nothing clever happens yet. You're just holding up a mirror. If people start checking that mirror on their own, without being told to, you've earned the right to move to the next stage.

That next stage is insight, where you start pointing out patterns inside the visibility that aren't obvious from the raw numbers alone. This is where people start referencing what the system shows them in ordinary conversation, the way you'd mention something a trusted colleague had noticed, rather than something a screen had generated.

From there comes assistance, where the system starts suggesting what somebody might want to do about what it's shown them. This is a meaningfully bigger step than it sounds, because you're no longer just informing, you're advising, and advice can be ignored in a way that information rarely is. If people start actually acting on what's suggested, that's the signal that trust has deepened.

Only then does prompting arrive, where the nudge shows up at the exact moment a decision needs making, rather than waiting for someone to go and look for it.

And only after prompts have proven themselves, repeatedly, in a way that people have come to rely on without resentment, do we even begin to talk seriously about automation, where the system finally acts on someone's behalf because the trust has been earned rather than assumed.

I've watched organisations jump straight to the last stage because the technology made it possible. I understand the temptation, but when people have not had time to understand or challenge what is happening, suspicion can undo months of otherwise good work.

People don't distrust automation because they're resistant to change in some general sense. They distrust automation that arrives before it's earned the right to be trusted, and that's an entirely reasonable thing for anyone to feel.

So the ladder is a useful pattern rather than a law every system must obey. The pace depends on the decision, its consequences and what people already understand. User confidence is only one part of that. A system can be reassuring and wrong, so tested reliability, appropriate controls and somebody clearly accountable still have to sit alongside the trust. Familiarity is not a substitute for checking that the thing works.

I have found that people are remarkably tolerant of imperfect technology when they understand what it is doing and retain a sense of control. The opposite is also true. A technically brilliant automation that makes a surprising decision without explanation can lose trust in seconds. Once that happens, every future suggestion gets judged through the memory of the one time the machine did something daft.

That is why explainability, in the ordinary human sense rather than the academic one, matters so much. A manager does not need a lecture on model weights. They need enough context to say, 'I can see why it thinks that.' Trust grows from those small moments of recognition. Automation is the final rung because by then the system has spent time being right in public.