Designing a research workflow that survives a deadline

Ask a firm to describe its research process and you will usually get a description of the good version: define the question, search primary sources, check the position is current, record what was found, form a view. Watch the same firm on a week when three things land at once, and none of it happens. Somebody searches, finds something plausible, and gets on with the drafting.

This is not a discipline problem. It is a design problem. The described process has costs that are tolerable on a quiet Tuesday and intolerable on a bad Thursday, so it is abandoned exactly when the risk is highest. A workflow worth having is one whose steps are cheap enough to survive the bad Thursday.

Scope the question before searching it

The most common failure is not searching badly. It is searching a question nobody has stated precisely, which means the searcher does not know when to stop.

The fix costs about ninety seconds: before opening anything, write down the question in one sentence, the facts it depends on, the jurisdiction, and the date the position needs to be assessed as at. That last element is the one most often skipped and most often fatal — research conducted against the current text when the facts arose four years ago is not partially useful, it is wrong.

Write it at the top of the note you are going to produce anyway. It is not overhead; it is the first paragraph.

Record the search, not just the finding

Most research notes record conclusions. Very few record how the conclusion was reached. This matters for two reasons, and only one of them is defensive.

The defensive reason is obvious: if the position is challenged eighteen months later, the note needs to show what was searched, in what sources, as at what date, and what was reviewed and set aside. A conclusion without that trail is an assertion.

The practical reason is better. A recorded search is resumable. When the matter comes back — and it does, usually with one fact changed — the person picking it up can see the ground already covered instead of starting again. This is where most of the compounding value in legal research actually lives, and almost no firm captures it.

A useful trail is short: sources consulted, the queries that produced results worth reading, the date, and the version of any provision relied on. Four lines, not four pages.

Capture the reasoning at the point of reading

The gap between reading a passage and writing about it is where most of the value evaporates. You read something, you understand immediately why it matters for this matter, and three hours later you can reconstruct only that it mattered.

Annotate as you read, in the file rather than in a separate document, and write the sentence that connects the source to the matter — not what the source says, which you can always re-read, but why it changes what you were going to do. "Our clause 14.2 gives a window rather than a fixed date" is worth more than any amount of quoted text.

Make the check for currency a step, not an assumption

Every workflow needs an explicit point at which someone confirms the sources relied on are still good. Not because people forget it, but because when it is implicit it is the first thing to disappear under pressure.

Make it a named step with a recorded outcome. And where the tooling allows, make it ongoing: if the platform can watch what a matter relies on and tell you when something changes, the check stops depending on anyone remembering to repeat it. A matter that runs for nine months will outlive at least one amendment to something it depends on.

Let the team see the work

In most firms, research is invisible to everyone except its author. The same question gets researched three times a year by three people who never learn of one another. Nobody is at fault; the information was never surfaced.

The fix is filing research against matters in a way colleagues can find, with permissions that respect confidentiality and any ethical walls in place. The mere fact that a question has been looked at before — even if the earlier note is not directly usable — is worth knowing, because it points at a colleague who has already thought about it.

What good looks like

A workflow that survives the bad week has four properties: the question is written down before the searching starts; the search leaves a trail as a by-product rather than as an extra task; reasoning is captured while it is fresh and attached to the file; and currency is checked as a step rather than assumed.

None of this requires more time than the informal version, once it is embedded. It requires the steps to be cheap. Every one of these habits fails the moment it costs more than about a minute, which is why the tooling question and the process question are the same question.