In short
If the role has no established title and no keyword that reliably finds the right people, stop searching for a profile and start searching for a problem. Write down the outcomes, the specific evidence that would convince you, and two or three adjacent fields where people already solve a structurally similar problem — with an explicit test for what transfers. That document is what makes the search possible; a job advert is not.
Key facts
- The failure mode
- Keyword search returns the wrong people confidently
- Search by
- The problem and its constraints, not the job title
- Widens the pool most
- Naming acceptable adjacent backgrounds up front
- Decide before CVs arrive
- What evidence proves each must-have
Why keyword search fails on novel roles
Keyword search works when a role is common enough that the market has agreed a vocabulary for it. Early-stage technical roles frequently sit before that point: the title is invented, the stack is incidental, and the hard part is a combination nobody has packaged into a job description yet.
Search on those keywords anyway and two things happen. People who use different words for the same work are invisible, and people who use the right words for different work look like a match. The list gets longer and less relevant at the same time.
Search by problem instead
Five steps that take about an hour and change what the search returns.
Describe the hard part in plain English
Two or three sentences a non-specialist could follow. Name whether the difficulty is the domain, the scale, the constraints or the ambiguity — those lead to different candidates.
Write the outcomes, not the responsibilities
Three to five things that will be true in twelve months and are not true now. Rank them. Candidates can react to outcomes; they cannot react to a list of technologies.
Turn each must-have into evidence
For each requirement, write what would convince you: a system built at a comparable scale, a device that shipped, a migration led, a result they can defend in detail.
Name the adjacent fields
Where else do people solve a structurally similar problem? For each field, write what transfers and what does not, and the test you would apply.
Name places, not just skills
Two or three companies, research groups or open-source projects where this problem is familiar. This is what makes an outbound search possible at all.
Keyword thinking versus problem thinking
The same role, written two ways.
| Keyword-led | Problem-led |
|---|---|
| Senior engineer, 5+ years, named framework | Owns reliability of a system with hardware in the loop that changes weekly |
| Must have industry experience | Must have worked where a mistake is expensive and verification is part of the job |
| Strong communicator | Explains a trade-off to a non-specialist founder well enough for them to decide |
| Machine learning expert | Has taken a model from a notebook into something with evaluation and monitoring |
Illustrative wording only — write your own from your actual constraints.
Signs you are still keyword searching
- The requirements list years rather than evidence.
- You have rejected candidates for the stack when the hard part is not the stack.
- Nobody can say what would get someone rejected, so every screen is re-argued.
- The adjacencies were agreed in principle and rejected in practice.
- The advert would describe five different roles at five different companies.
Common questions
- Does this mean job titles do not matter?
- Titles still matter for the advert and for the candidate's own career, so pick a recognisable one. They just should not drive the search when the work is genuinely novel.
- How many adjacent backgrounds should I accept?
- Two or three, each with a written transfer test. More than that usually means the role has not been defined; fewer means the pool may be too small to hire from in your timeframe.
- What if the search is still slow?
- Treat a slow search as information. It usually points at the role definition, the range or the location constraint rather than at effort. Decide in advance the point at which you would change the plan.
Turn this into a brief
Answer six short sections and you have a technical hiring brief you can share internally or send straight to us.
Open the hiring brief builderAny percentages or figures shown in this article are illustrative examples used to explain the model. They are not quoted rates, market benchmarks or salary data.