Decision archaeology is the practice of reconstructing why a business decision was made: who decided, when, and what they were weighing at the time, from the trail already left in email threads, document version histories, and meeting notes. Most organisations retain the decision itself. Almost none retain the reasoning.

Why does the reasoning disappear?

Because nothing in a company's systems is designed to hold it. A policy document records the rule. A contract records the term. A ticket records the change. None of them record the trade-off that produced it. The result is a slow, invisible tax: teams re-litigate decisions that were settled years ago; new joiners inherit rules they can't evaluate; rules outlive the conditions that justified them; auditors ask why this is your standard and the honest answer is that nobody knows.

What decision archaeology actually looks like

An agent built for this reads across mail threads, document revision histories, shared drive comment trails, and meeting transcripts — assembling a decision record for a given rule or artefact. Not a summary of a document. A reconstruction of an event. A useful output captures what was decided, when and what triggered it, who participated, what options were considered, the stated rationale, whether a later decision superseded it, and how much of the record is directly evidenced versus inferred.

Where the value shows up

Onboarding. A new department head can read the reasoning behind twenty standing rules in an afternoon rather than absorbing them over eighteen months.

Removing dead weight. Once you can see the original justification, you can test whether it still holds.

Client and regulator questions. Why is this your standard becomes a retrievable answer with a date and an author attached.

Succession. When a long-serving person leaves, the decisions they made stay legible to whoever inherits them.

What you need for this to work

Three conditions matter: retention (if mail is purged after ninety days, there is nothing to read), concentration (it works best where discussion happened in a small number of systems), and a bounded starting question.

What this is not

It is not search. It is not summarisation. And it is not surveillance. The scope is decisions and their rationale, not individual behaviour.