Which decisions belong in a document and which belong in chat

A decision that will still matter in a month belongs in a document. A decision that only coordinates today belongs in chat. Most teams apply this backwards.

The refund rule that lived in a thread

Elena’s team sold annual plans with a thirty-day refund window. In October, support began granting refunds at fifty days. It applied to anyone who asked twice.

The exception was agreed in a thread of forty-one messages. It spread across two days and involved three people and one manager. No page was ever written.

In January a new agent refused a refund at day forty. She quoted the policy page. The customer said a colleague had promised it, and he was right.

What chat is good at

Chat is fast, low-commitment, and visible to whoever is awake. Those three properties make it the right place for coordination. None of them help a reader next year.

Who is picking up the call, whether the build is green, whether anyone has the login. All of it belongs there. All of it expires within days.

None of those answers needs to survive the week. Chat is honest about that by scrolling away. The archive is a courtesy, not a record.

What chat destroys over time

A thread has no title, no owner, and no version. The same question gets re-decided every few months by different people. Each round produces a slightly different answer.

Search works on words rather than meaning. A decision phrased as let us just do it is effectively unfindable. Nobody types those words looking for a rule.

The thread felt like a record because it was complete. Completeness is not legibility. The complete version was forty-one messages long, and nobody read past the first five.

The test of a decision that outlives the week

Elena’s test was one question. Will someone need to know this after everyone in the thread has forgotten it. If yes, it goes to a page.

The question is deliberately about the future reader. It is not about how important the topic felt. Importance is judged badly in the moment.

A pricing rule passes easily. A lunch plan does not. The test sorts them without anyone having to argue about status or seniority.

Where the boundary sits in practice

Coordination, scheduling, quick confirmations, and anything expiring within days stay in chat. That covers most messages. It should cover more than it does in most teams.

Policies, prices, access rules, and anything a customer can quote go to a document. The last clause is the useful one. A customer with a screenshot is the real test.

The boundary is not about length. A one-line pricing rule is a document. A thousand-word debate about lunch is not a document at all.

The thread that should have been a page

The refund exception had three properties that made it a document. It changed a rule, it applied to customers, and it had no end date. Any one would have justified writing it down.

Having all three made the omission a category error rather than a slip. The team made it repeatedly that autumn. Two other exceptions went the same way.

None of them felt like a decision at the time. Each felt like a favour to one customer. That is how most policy changes begin.

The page that should have been a thread

The opposite failure is a document for a decision lasting a week. It then sits in the index looking permanent. Readers treat it as current for months.

Elena found four pages describing a temporary shipping arrangement. It had ended in the spring. Two of the pages contradicted each other on the cutoff date.

Stale pages are worse than missing ones. A reader trusts a page and acts on it. A missing page at least prompts a question.

Writing the page after the fact

The refund page was written in November, six weeks late. The manager who made the call wrote it himself. It took eleven minutes.

The reasoning already existed in the thread. It only needed a shape and a date. Most missing pages are missing for this reason alone.

The expensive part was never the writing. It was six weeks in which two agents applied different rules. One of them lost a customer.

Who owns the page

Every page needs one name at the top, and that name must be a person. A team name means nobody updates it. Responsibility spread across five people is felt by none.

Elena’s rule was that the owner is whoever would be asked the question if the page did not exist. That names a person immediately. It also names the person who cares.

Ownership also sets a review habit. The owner checks the page when the rule changes. Without a name, the page is checked by nobody.

Linking rather than copying

The thread still happens. What changed is that its conclusion ends with a link to the page. Whoever closed the thread posts it.

Copying the text into chat would create a second version within a week. The two would drift apart. Nobody would know which one a customer had read.

One link, one source. The thread becomes a pointer rather than a competing record. That single habit removed most of the contradictions.

The rule the team adopted

Their rule fit in one sentence. If a customer could quote it, it goes on a page before the day ends. It is crude and it works.

It caught the refund case, the pricing exception, and the two access rules that followed. It missed nothing that mattered in six months. It also created no paperwork.

Crude rules get used. Nuanced ones get discussed and then ignored in favour of whatever is fastest. The team had tried nuanced before.

What still leaks into chat

Small exceptions still get agreed in threads, and some of them should. The team accepts a slow leak rather than a strict process. The leak has a known cost.

That cost is roughly one contradiction per quarter. They decided it was cheaper than reviewing every thread. A review process would have cost several hours a week.

The trade is a judgement rather than a principle. It would change if the product were regulated. It would also change if the team doubled in size.

Where is the line in your team between a decision that needs a page and one that only needs a reply, and who drew it?