From a written brief to a real search

A brief is prose. A search is a query.

Every search begins as a sentence somebody wrote. “Senior backend engineer, fintech background, has run a small team, Bangalore or remote within India.” That is how briefsare written, and there is nothing wrong with it. It is simply not a search. A search is a set of constraints applied to a population, and the distance between the sentence and the constraints is where most sourcing quietly goes wrong.

Usually that distance is crossed out of sight. Something reads the brief, does whatever it does, and hands back a list. If the list is good you are pleased. If it is poor you have no way of knowing which part of your sentence caused it, so you rewrite the sentence and try again.

The brief becomes named axes

On EmployLabs the brief is turned into an explicit structured queryalong named axes — seniority, function, industry, location and the like. Each axis holds the value that was read out of your prose, written down rather than held in the model’s head.

So “senior backend engineer at a fintech” does not remain a phrase. It becomes a seniority band, a function, an industry and a place, each one a separate thing with a separate value. The interpretation is now an object on the screen instead of an assumption inside a process.

Every axis is shown to you, and every axis can be changed

The axes are not internal machinery we describe in a blog post. They are on the role, in front of you, before the search runs, and you can adjust any of them. The system has made a reading of your brief and it shows you the reading rather than only the consequence of it.

This is the difference between a tool you have to trust and a tool you can check. Trust is what you are left with when there is nothing to inspect, and it is a poor substitute for looking.

A black box cannot be corrected. It can only be asked again, differently, and hoped at.

When the results look wrong, you can see which axis is wrong

This is the part that changes a recruiter’s day. A disappointing result set is no longer one undifferentiated problem called “the AI got it wrong”. It is almost always one axis carrying a value you would not have chosen, and you can read the axes and find it.

• Everyone is too junior. Look at the seniority band. Your brief probably described the work rather than the level, and the level was inferred from the work.
• The companies are wrong. Look at industry. A brief that mentions a domain in passing can have that domain read as a hard constraint, or the reverse.
• Nobody would realistically move. Look at location. A place named once in a long paragraph carries a great deal of weight in a search, and it may be narrower or broader than you meant.

In each case the fix is a change to one value, not a rewrite of the brief. Rewording prose to move a search is guesswork, because you are editing the input to a translation you cannot see. Changing an axis is an edit to the search itself.

Why the prose still matters

None of this makes the brief unimportant. The axes are what a search can act on, and a brief carries a good deal that no axis holds — why the role exists, what the first ninety days look like, what would make somebody say yes. That context is used elsewhere in the pipeline and it is worth writing well.

What the axes do is stop the searchable part of the brief from staying implicit. Two things were being asked of one paragraph, and only one of them is a query.

The axes are also a shared language

There is a second, quieter benefit. A brief lives in one person’s head as much as on the page, and when a role changes hands — a colleague covering a week of leave, a hiring manager wanting to know why a particular profile keeps appearing — the prose does not travel well. Two people can read the same paragraph and picture different candidates, and neither of them is obviously wrong.

A set of named axes with values on them does travel. It is short enough to read in a minute and specific enough to argue with, which makes the disagreement a useful one: not whether the search feels right, but whether this seniority band and this industry are the ones the role needs. Most of the arguments worth having about a search are arguments about a single value, and prose keeps them from ever getting that far.

It is a small thing on one role and a large one on twenty. The searchable definition of a role stops being folklore and becomes something written down, in the same shape every time, which a person can check.

How to use it

1. Read the axes before you read the results. It takes a few seconds and it tells you what the search is about to do, while it is still cheap to change.
2. When a result set is wrong, change one axis and run it again. Changing three at once tells you the next set is different, not which change made it so.
3. If an axis keeps coming back with a value you did not intend, the brief is ambiguous at that point. Fixing the sentence there is worth more than fixing the axis every time.