"Hard to fill" is a label applied to five different problems, only one of which is a sourcing problem. Working the market harder on the wrong diagnosis is how a role stays open for seven months while everyone involved becomes convinced the talent doesn't exist.
This is a method rather than a tactic list: size the market, diagnose why it's hard, then apply the fix that matches — including the fix nobody wants, which is changing the role.
Step 1 — Size the market before you source
Before contacting anyone, estimate how many people plausibly qualify within your constraints (geography, work authorisation, seniority, comp band). Any search tool can produce this: run the honest version of your requirements and count.
| Qualified population | What that means | Approach |
|---|---|---|
| 5,000+ | Not a scarcity problem | Look at outreach, offer, and process — the market is there |
| 500–5,000 | Competitive but workable | Standard outbound with sharper targeting and faster process |
| 100–500 | Genuinely scarce | Named-account sourcing, artefact sourcing, relationship building over months |
| Under 100 | The requirements are the problem | Change the role: split it, relax a constraint, or build internally |
This exercise is worth an hour because it converts an argument into arithmetic. If 60 people in your metro area meet the brief, and 20 of them work at competitors who pay more, no tool and no amount of outreach discipline changes the outcome — and telling the hiring manager that early is the highest-value thing a recruiter does on a role like this.
Step 2 — Diagnose which kind of hard
| Cause | Symptom | Fix |
|---|---|---|
| Scarcity | Small qualified population; few results even on loose queries | Artefact sourcing, geography expansion, adjacent-skill hiring, build internally |
| Saturation | Population is large but nobody replies | Differentiate outreach; use warm channels; improve employer story |
| Compensation | Good replies, candidates drop at the comp conversation | Benchmark the band and take evidence to finance; nothing else works |
| Over-specified brief | Requirements list eight things; nobody has all eight | Separate must-haves from learnable; test against your successful hires |
| Unattractive offer | Interest until candidates learn about scope, manager, or process | Fix the role, the loop speed, or the story — not the sourcing |
The diagnostic questions are cheap. Where do candidates drop out? If nobody replies, it's saturation or targeting. If everyone replies and drops at comp, it's comp. If they drop after meeting the hiring manager, sourcing more people is an expensive way to repeat the experience.
Over-specification deserves particular scrutiny because it is the most common and the most fixable. Take the requirement list to the hiring manager with the market size attached: "there are 40 people with all six of these; there are 900 with five of the six — which one do we drop?" That conversation, backed by numbers, is far more productive than a debate about candidate quality.
Step 3 — Source from artefacts, not profiles
For genuinely scarce talent, the profile databases everyone licenses are the worst place to look, because every competitor is running the same query against the same index. The people you want are often visible instead through what they have produced:
| Artefact source | Works for | How to use it |
|---|---|---|
| GitHub, package registries | Engineers, infra, ML | Contributors to the specific libraries your stack depends on |
| Papers, arXiv, citations | Research, ML, science | Authors and their co-authors; recent-graduate pipelines via advisors |
| Patents | Hardware, chemistry, medtech | Inventor names and their current employers |
| Conference talks and meetups | Any specialism | Speaker lists from the last three years of relevant events |
| Professional registries and licences | Healthcare, trades, legal, finance | Public registers give verified credentials and often location |
| Community forums, Discords, Slacks | Niche tooling and specialisms | Participate first; recruit second. Extraction gets you banned |
| Industry awards and certifications | Specialist operations | Award shortlists are pre-vetted lists of practitioners |
Artefact sourcing needs Boolean and X-ray searching, not natural-language search over a vendor's index — the sources aren't in anyone's index, which is precisely the advantage. Among directory tools, AmazingHiring is the most explicitly built around technical artefact signal, and SeekOut has the deepest filter set for narrowing a scarce population. Our executive search and tech recruiting rankings cover the specialist end.
Step 4 — Engage on a longer horizon
With 100 qualified people in the world, a 20% reply rate is 20 conversations — and the same people will be your candidates for years. That changes the engagement model:
Personalise properly. With a list of 40, spend ten minutes per person. Reference their actual work: the talk, the library, the paper. Generic outreach to a scarce, saturated audience is worse than no outreach, because you have spent your one credible first contact.
Use warm channels first. Shared connections, alumni, community overlap, past applicants, and your engineers' networks. In scarce markets, referrals aren't a nice supplement — they're often the primary channel, and reply rates of 50–80% on warm intros dwarf anything cold.
Have the hiring manager reach out. For senior and scarce roles, a message from a practitioner peer materially outperforms a recruiter message. It is also the one tactic that is genuinely hard for competitors to copy at scale.
Accept "not now" as a result. Build a real pool: log the conversation, note the timing, and re-contact in two quarters. Half of scarce-market hiring is being the recruiter who was pleasant a year ago when the candidate's situation changed.
Automation helps less here than anywhere else in sourcing, and this is a case we'd make even though we rank agentic tools highly overall: agents earn their money on throughput over specifiable roles, and a 40-person market has no throughput to gain. Use tools to find the artefacts and to keep the follow-up disciplined; write the messages yourself. The boundary is discussed in AI sourcing agents explained.
Step 5 — When to stop sourcing and change the role
If the qualified population is under a hundred, or the comp band is below market, sourcing effort has a hard ceiling. The interventions that actually resolve these roles are structural:
Split the role. Two people with four of the six requirements each, instead of one unicorn with all six. Usually cheaper and faster than the unicorn search.
Hire adjacent and train. Identify the one or two genuinely hard-to-learn skills and treat the rest as trainable. This widens the population by an order of magnitude and is how most scarce-skill teams were actually built.
Expand geography. Remote or a new hub multiplies the market. See our remote hiring ranking for the tooling implications.
Fix comp with evidence. Take three anonymised examples of candidates who declined at the band, with market data, to whoever owns the budget. This is a recruiter's job on a hard role and it is more valuable than another 200 messages.
Use an agency for this one. A specialist retained search on a genuinely scarce role at 20% of salary can be rational — you're buying an existing relationship network, not effort. Compare against the arithmetic in our ROI model.
Build internally. Promote and backfill the easier role. Often the fastest path and almost always the cheapest, and it is the option that gets raised last.
The uncomfortable summary: on the hardest roles, the recruiter's most valuable output is often not candidates but evidence — market size, comp reality, and requirement tradeoffs presented clearly enough that the business changes the role. For tool selection on the roles that are workable, start with the 2026 ranking or the tool finder.