Ask about the shape of the problem when you cannot hand over the material

The material is off the table before the question is

A great deal of the work worth thinking about arrives attached to material that cannot leave the building: a client contract, a colleague’s medical situation, an unannounced product, a dispute with named people in it, a dataset covered by an agreement somebody signed years ago. The decision not to paste it is usually the right one and it is not the end of the matter, because the question underneath is often entirely ordinary.

What people do instead tends to fall into two unhelpful shapes. They paste it anyway, having quietly decided the risk is theoretical, or they abandon the question and proceed on instinct, which was not obviously the safer option either.

There is a third move, and it is a skill rather than a policy. You can usually separate the structure of a problem from the instance that carries it, and the structure is the part you actually wanted help with.

Describe the structure, withhold the instance

Start by writing down what makes the situation difficult without naming anything. A dispute over whether a supplier missed a deadline is, structurally, a disagreement about what a document promised, complicated by a course of dealing that contradicts it. That sentence contains no identifying material and it is enough to have a genuinely useful conversation about how such disagreements usually resolve.

The discipline is to keep the properties that drive the answer and drop the ones that merely identify. Dates matter when the sequence matters, and become a liability the moment they are precise enough to locate the event. Amounts often matter as ratios rather than as figures, and a ratio identifies nobody.

Do this in writing rather than in your head, because the process of stripping the case down is itself clarifying. Half the time you discover the difficulty was somewhere other than where you assumed, and you have written the brief you should have written anyway.

A substitute has to behave the same way to be worth using

The obvious next step is to replace the real particulars with invented ones, and it works when the substitution preserves the mechanics. Changing names, places and industries is usually safe. Changing the numbers is safe only when nothing in the reasoning depends on their size, which is less often than people assume.

A substitute that behaves differently produces an answer that does not transfer, and the failure is silent, because the response will be perfectly reasonable about the fictional case you described. If the real contract has an unusual clause and your sanitised version does not, everything you get back is about a contract you do not have.

So sanitise deliberately and then read your version as though it arrived from a stranger. Ask whether anything you changed could flip the answer. That single check catches most of the substitutions that were about to waste your afternoon.

Ask for the method rather than the conclusion

When the material stays out, the request that pays best is procedural. Ask what questions a person handling this kind of situation would want answered first, what the usual failure modes are, what distinguishes the straightforward version from the difficult one, and what evidence normally settles it.

A method transfers across instances in a way that a conclusion does not, which makes it exactly the right output when the instance is missing. You then apply it yourself, with the confidential material in front of you, which is where it was always going to have to happen.

This also produces something you can keep. A checklist of the questions worth asking about a category of problem is reusable, and it is safe to store and share in a way that a discussion of a specific case never is.

Some questions do not survive the abstraction

Be honest about where this stops. Anything that turns on the exact wording of a document cannot be abstracted, because the wording is the case. Anything requiring the pattern across a real dataset cannot be answered from a description of the dataset. Judgements about particular people are not generalisable and should not be attempted from a sketch.

For that class of question the answer is that the tool is not available to you for this task, which is a legitimate outcome rather than a failure of technique. The remaining options are the ones that existed before: a colleague under the same obligations, a professional you can properly instruct, or your own reading of the material.

The point of the abstraction habit is not to get around a restriction. It is to notice that a fair number of questions people abandon for confidentiality reasons were never really about the confidential part at all.