# The Talent Sourcing Playbook: From Role Brief to Ranked Shortlist

*You will run a full sourcing cycle for one role - brief, pool sizing, search, sequenced outreach, scored screen, calibrated shortlist - without burning a week on the wrong pool.*

- Canonical URL: https://www.refolk.ai/guides/talent-sourcing-playbook-2
- Pillar: Recruiting and sourcing
- Format: Playbook
- Published: 2026-09-16
- Last reviewed: 2026-09-16
- Reading time: 16 min
- Keywords: talent sourcing playbook, how to source candidates, sourcing process, build a candidate shortlist, candidate shortlist

## Key takeaways

- Size the pool before you write a single message: in Refolk's index there are 54,976 US Software Engineers with Python but only 614 with Rust, making the Rust req 89.5x rarer and a volume plan useless.
- Channel beats copy on first contact - the same candidates replied 16.6% of the time on LinkedIn versus 4.4% by email, so pick the channel before you polish the message.
- A shortlist is 4 to 8 people drawn from a longlist of 15 to 25, and you write three to five must-haves before you open a single CV or you will reverse-engineer people you already liked.
- Single messages are not campaigns: most replies land on follow-ups, and a 4-stage recruiter email sequence reaches a 32 to 35% reply rate against a 4.4% first-touch email rate.
- Track reply-to-screen, not reply rate - a 35% reply from the wrong pool that yields 5% screens is worse than a lower rate from the right one.
- Speed is a selection mechanism: top candidates leave the market in about 10 days and 57% abandon slow processes, so a 44-day average is structurally choosing from whoever is left.

This is the end-to-end method for sourcing one open role, from a signed brief to a ranked shortlist of four to eight people. It is for in-house recruiters, sourcers, talent leaders, and founders doing their own hiring. Run it in order and you will not waste a week on the wrong pool, send single messages that look like a bad market, or reverse-engineer a shortlist of people you already liked.

Talent sourcing is the proactive process of identifying and engaging potential candidates to generate a flow of applicants. That is different from recruiting, which focuses on filling the current opening. The distinction matters because sourcing done well front-loads the work: the pool you can reach and the criteria you agreed decide most of the outcome before the first message goes out.

## What a repeatable sourcing process actually looks like

A repeatable sourcing process is a fixed loop of six to seven stages, run the same way every time, so the outcome depends on the pool and the brief rather than on who happened to run it. The public consensus converges on the same shape, and step one is always a written brief, not a search string.

The loop: define the role and candidate profile, research and segment the talent pool, select channels and build search strings, personalize outreach in sequences, nurture and screen against a scorecard, then hand the longlist to the hiring manager and calibrate the shortlist. Measure and optimize sits on top, run quarterly rather than per-req.

There is one genuine order disagreement in the source material. Most guides put "define channels" before outreach, which this playbook follows. Some fold scorecard-building into step one rather than treating screening as its own step. The rule that resolves it: screening comes after outreach or application and before interviews. Screening removes the clearly unqualified; shortlisting picks the best of who is left. Keep them separate so you can see where candidates actually drop.

#### The sourcing loop for one role

1. **Brief** - Recruiter and hiring manager agree 3 to 5 written must-haves
2. **Segment** - Sourcer sizes the pool and lists target companies and titles
3. **Search** - Boolean and X-ray produce a longlist of 15 to 25 profiles
4. **Outreach** - Personalized first touch plus 2 to 3 sequenced follow-ups
5. **Screen** - Every responder scored against the agreed criteria
6. **Calibrate** - Hiring manager narrows to a final shortlist of 4 to 8

*Seven stages in order, from a signed brief to a shortlist you can defend.*

## Size the pool before you write a single message

The single most expensive mistake in sourcing is committing a channel and a week of outreach to a pool that is too small or wrongly located to yield. Size it first. Scarcity and geography are both quantifiable before you send anything, and they change the whole plan.

Two variables move the pool by orders of magnitude: the skill and the country. A req that swaps one skill for a rarer one, or moves from one market to another, is not the same job with different words - it is a different sourcing problem that needs a different plan.

**89.5x - How much rarer Rust engineers are than Python engineers in the US**

In Refolk's index, 614 US Software Engineers list Rust against 54,976 who list Python.

**Block A: same title, two skills, one market.** The skill line on the brief silently sets the difficulty.

| Role (United States) | Count | Share of Python pool | Scarcity vs Python |
|---|---|---|---|
| Software Engineer, Python | 54,976 | 100% | 1.0x (baseline) |
| Software Engineer, Rust | 614 | 1.1% | 89.5x rarer |

**Block B: same title and skill, two countries.** The location line does the same thing.

| Software Engineer, Python | Count | Relative pool size |
|---|---|---|
| United States | 54,976 | 20.2x Germany |
| Germany | 2,716 | 1.0x (baseline) |

The US Python pool is 20.2x the size of Germany's. A "same role, EU version" brief changes the math without anyone deciding it should. When the number is large, volume outreach works and you can afford a tighter must-have list. When the number is small - a Rust req, a niche market - stop planning for volume. You need precision, patience, and probably a wider net on adjacent skills or a longer timeline agreed with the hiring manager up front.

> **Watch out:** A scarce pool will not fill by volume
>
> A 614-person pool does not respond to more messages the way a 54,976-person pool does. Treating the two identically wastes the week you would have spent sending outreach that was never going to land.
> </callout>

## Read the brief and the funnel before you commit

The brief is the contract, and the funnel is the physics. Write three to five must-haves with the hiring manager before you open any CV, and know the conversion rates you are working against so you do not mistake a normal funnel for a broken one.

Order matters more than most people admit. Define the must-haves first, written down and agreed, then score. Teams that read CVs first and write criteria later reverse-engineer a list of people they already liked. That is not selection; it is confirmation. The timestamp on your criteria file should predate the first CV you opened.

The funnel is brutal and getting tighter. In 2026 it converts at roughly 0.5% applicant-to-hire - one offer per 200 applicants - with only about 8% of applicants passing initial screening and offer acceptance around 82%. That is why sourcing quality, not volume, decides the result: referrals and direct sourcing convert 4 to 10 times higher than job board applications. And your best pipeline may already exist - 46% of 2026 hires came from candidates already in the ATS.

```funnel
title: The 2026 sourcing funnel, per Gem
caption: Roughly one offer per 200 applicants, so quality at the top decides the yield.
stage: Applicants :: 200 :: baseline volume for one hire
stage: Pass initial screen :: 16 :: about 8% of applicants
stage: Offer :: 1 :: ~0.5% applicant-to-hire
stage: Accept :: 0.82 :: ~82% offer acceptance
```

> **Rule:** Criteria before CVs, always
>
> Write and agree three to five must-haves before you open a single profile. If the criteria file is younger than your first CV, you built a rationale for people you already liked, not a filter.
> </callout>

## The seven-step procedure

Run these in order for one role. Owners and rough durations are noted so you can staff it. Done-conditions are what a good result looks like at each stage, not just "moved on".

#### Run one full sourcing cycle

1. **Write the brief and candidate profile** - Recruiter and hiring manager, ~1 to 2 hours in an intake meeting. Translate the job description into 3 to 5 written must-haves plus preferred backgrounds, seniority signals, and deal-breakers. Done = a one-page brief both parties signed off.
2. **Research and segment the talent pool** - Sourcer, ~2 to 4 hours. Estimate pool size and where it lives before writing outreach. Done = a target list of companies, titles, and locations plus a realistic count.
3. **Build search strings and select channels** - Sourcer, ~1 to 2 hours. Boolean and X-ray on LinkedIn, GitHub, and Google; pick channels by role type. Done = saved searches producing a longlist of ~15 to 25 qualified profiles per opening.
4. **Send outreach in sequences, not single messages** - Sourcer or recruiter, ongoing over 1 to 2 weeks. Personalize the first touch and queue 2 to 3 follow-ups. Done = sequence live, reply rate tracked by channel.
5. **Screen replies against the scorecard** - Recruiter, rolling. Score on the agreed criteria, not intuition. Done = every responder scored on the agreed must-haves.
6. **Calibrate the shortlist with the hiring manager** - Hiring manager with recruiter, ~1 hour. Present the scored longlist and narrow to a final 4 to 8 with scorecard data attached. Done = agreed shortlist of 4 to 8.
7. **Measure and optimize** - Recruiter or TA lead, quarterly. Track sourced-to-hire ratio, engagement, and response by channel. Done = channel yields logged and next cycle's mix adjusted.

## Build searches and pick channels by role type

Channel choice is a bigger lever than message polish on first contact, and search-string discipline is what turns a raw skill mention into a qualified longlist. Get both right and you land a longlist of 15 to 25 profiles per opening that are worth outreach.

On the exact same candidates, 16.6% replied via LinkedIn against 4.4% via email - nearly four to one. The mechanism is trust: an InMail carries a platform signal a cold email lacks. So for most white-collar roles, lead on LinkedIn. For engineers with a public footprint, the GitHub graph adds signal a profile cannot: real contributions to real projects.

Boolean and X-ray search remain the core discovery skill. Two constraints shape how you write queries. First, GitHub search returns at most 1,000 results per query, so a search that "finds 100,000 profiles" is lying to you and mostly surfacing people who mention the skill in passing. Add must-have qualifiers and exclusions before you review anything. GitHub's advanced search supports qualifiers like `language:`, `location:`, and `followers:` to tighten the net.

This is exactly where a plain-English search removes the friction the two paragraphs above just described. Instead of hand-tuning Boolean strings and hitting the 1,000-result cap, you describe the person and get a scoped list back.

I ran this search: `Senior software engineers in Berlin who use Python and have shipped backend systems at Series A-C startups.` - [see the full result list](https://www.refolk.ai/s/73cqdbwg1q).

*Returns named engineers matching skill, seniority, city, and company stage, drawn from public LinkedIn and the GitHub graph - a ready longlist rather than a raw keyword dump.*

[Refolk](/) is where I fold that discovery into one step: you ask for the pool in the brief's own language and get the segment back, already scoped to skill, seniority, and place, so step two and step three collapse into a single pass.

**Boolean starting strings for a longlist**

```
LinkedIn X-ray via Google:
site:linkedin.com/in ("senior software engineer" OR "backend engineer") "Python" ("Series A" OR "Series B" OR "Series C") Berlin

GitHub advanced search:
language:Python location:Berlin followers:>20

Exclusions to add once results are noisy:
-intern -junior -"recruiting" -"student"
```

*Swap the skill, title, and location; add exclusions for junior titles or wrong industries before you review.*

## Write outreach that gets replies, in sequences

A single message is not a campaign, and treating it like one is why a good pool can look like a bad one. Sequence every send with two to three follow-ups, keep the first message short, and track reply-to-screen rather than raw reply rate.

Most replies land on the follow-ups, not on message one. Failing to follow up leaves results on the table. Gem's analysis of 4 million recruiter emails found a 76.6% average open rate and a 32 to 35% reply rate across a four-stage sequence - against the 4.4% you get from a single cold email. Best practice is two to three follow-ups spaced over a week or two.

Length matters on LinkedIn. InMails under 400 characters earn 22% higher response than average; messages over 1,200 characters run 11% below it. The InMail body limit is around 1,900 to 2,000 characters, and LinkedIn refunds the credit if the recipient replies within 90 days, so short and specific costs you nothing to test.

Judge the result against the right benchmark. For cold outreach to passive candidates - and roughly 70% of the global workforce is passive - 20 to 30% is solid, above 35% suggests strong personalization, and below 15% is a signal to revisit message quality and targeting.

**Short first-touch InMail (under 400 characters)**

```
Hi [first name] - I saw you built the backend for [specific system/project]. I'm hiring a senior backend engineer at [company], Python-heavy, Series B, remote-friendly across EU. Your work on [specific thing] is exactly the shape of the problem. Worth a 15-minute call this week?
```

*Name one specific thing about their work. Delete anything that could be sent to a hundred people unchanged.*

> Picking the channel is a bigger lever than polishing the message. LinkedIn out-replies email nearly four to one on the same people.

## Screen against a scorecard and calibrate the shortlist

Score every responder against the written must-haves, then hand the hiring manager a scored longlist and narrow together to four to eight. The scorecard is what stops "this feels right" from quietly re-hiring the last person who looked like this one.

The feeling that a resume is right is usually pattern recognition biased toward candidates who resemble past hires. Score on the agreed criteria instead. Typically the recruiter builds a longlist of 15 to 25, and the hiring manager narrows it to the final shortlist. Set a maximum shortlist size - 4 to 6 for most roles, 8 to 10 for high-volume positions - to prevent interview overload.

Shortlist quality caps interview ROI more than round count does. Google's people analytics team found four interviews is enough to predict role success, with each additional round adding less than 1% accuracy. Meanwhile technical roles already run 17.6 interviews per hire and business roles 11.7. Effort spent narrowing to four to eight well-matched people returns more than adding rounds.

**Block C: dilution up the ladder.** The senior pool is smaller and slower, so calibrate expectations before you promise a shortlist date.

| Band (US, Python) | Count | Senior-to-mid ratio | Typical time to fill |
|---|---|---|---|
| Software Engineer | 54,976 | - | ~44 days all-industry benchmark |
| Senior Software Engineer | 36,657 | 0.67x the mid pool | 55 to 62 days for tech roles |

Present the shortlist with scorecard data so the manager sees why each person is on it. That conversation is where a defensible shortlist is made or a reverse-engineered one is exposed.

## How this goes wrong: failure modes and false positives

Most sourcing cycles fail in one of eight predictable ways, and each has a cheap local check. This is the part worth reading twice, because every failure below costs at least the week you were trying to save.

#### Where the drop-off sits tells you what to fix

Horizontal axis runs from Low reply rate to High reply rate. Vertical axis runs from Low reach-screen to High reach-screen.

| Quadrant | What it means |
| --- | --- |
| Wrong channel or weak sequence | Fix outreach: sequence and channel |
| Vanity replies from wrong pool | Fix targeting: reply-to-screen is the tell |
| Dead pool or dead process | Re-check pool size and process speed |
| Healthy funnel | Protect it: keep speed and criteria steady |

*Diagnose by comparing reply rate against reply-to-screen, not by staring at one number.*

- **Reverse-engineered shortlist.** You read CVs before writing criteria and "discover" people you already liked. Check: the criteria file timestamp predates the first CV opened.
- **Single-message outreach masquerading as a campaign.** A 4 to 8% reply rate looks like a bad pool but is usually a missing sequence. Check: replies almost always come on follow-ups, not message one.
- **Vanity reply rate.** A high reply rate from people who will never move is worse than a lower rate from the right audience. False positive: 35% reply, 5% reach screen. Check: track reply-to-screen, not just reply rate.
- **Averages hiding leaks.** A company-wide 40-day average can hide an engineering pipeline at 62 days while admin roles close in 18. Check: segment time-to-fill by role and seniority before acting.
- **Mistaking a sourcing problem for a messaging problem.** High screening volume with low interview conversion is usually a sourcing quality problem, not a copy problem. Check: where the drop-off sits in the stage ratios.
- **Scarce-skill searches that never scale.** A 614-person Rust pool will not fill by volume the way a 54,976-person Python pool does. Check: estimate pool size in step two before committing a channel.
- **Over-collection on GitHub or Boolean.** A query returning 100,000 "profiles" mostly surfaces passing mentions, and GitHub caps at 1,000 results anyway. Check: add must-have qualifiers and exclusions before reviewing.
- **Slow process selecting from leftovers.** 57% of candidates lose interest when the process feels slow. Check: days-in-stage weekly, especially hiring-manager feedback latency.

> **Tip:** Speed is a selection mechanism
>
> Top candidates leave the market in about 10 days and a vacant role costs $4,000 to $9,000 a month in lost productivity. A 44-day average process is structurally choosing from whoever is still around.
> </callout>

## Keep the cycle honest: what to check before you close

Before you call a sourcing cycle done, verify the work against the criteria you agreed rather than against how the shortlist feels. This checklist catches the failure modes above while there is still time to act on them.

#### Before you hand over the shortlist

- [ ] The must-have criteria file was written and agreed before any CV was opened.
- [ ] You estimated pool size in step two and chose volume or precision accordingly.
- [ ] Search strings carry must-have qualifiers and exclusions, not raw skill mentions.
- [ ] Outreach ran as a sequence with 2 to 3 follow-ups, not a single message.
- [ ] You are tracking reply-to-screen by channel, not just raw reply rate.
- [ ] Every responder on the longlist has a scorecard result on the agreed criteria.
- [ ] Time-to-fill is segmented by role and seniority, not read off a company average.
- [ ] The final shortlist is 4 to 8 people, each with scorecard data attached.

## Measure the right things and adjust the next cycle

Measure sourced-to-hire ratio, pipeline engagement, and response rate by channel, then review quarterly which sources yield the best hires - not the most candidates. That last distinction is what turns measurement into better sourcing rather than a dashboard nobody acts on.

Three numbers earn their place. Sourced-to-hire ratio tells you whether your top-of-funnel quality is improving. Reply-to-screen by channel tells you which channels bring people who actually match the brief, which is why referrals - only 2% of applicants but 11% of hires, and referred candidates ten times more likely to be hired - keep beating job boards. And days-in-stage tells you whether speed is quietly costing you the candidates you already reached.

To keep this playbook current, do not chase the headline benchmark numbers, which drift year to year. Re-check the mechanism instead: pull your own reply-to-screen and time-to-fill by role each quarter, re-size the pool for any req whose skill or geography changed, and reallocate channel effort toward whatever produced hires rather than volume last quarter. The loop stays the same; the mix you feed it should move with your own data.

## Frequently asked questions

### How many candidates should be on a shortlist?

Four to eight for most roles, drawn from a longer longlist of 15 to 25. Set a hard maximum before you start - 4 to 6 for standard roles, 8 to 10 for high-volume positions - to prevent interview overload. The recruiter builds the longlist and the hiring manager narrows it to the final shortlist using scorecard data, not gut feel. Anything above eight usually means the criteria were never tight enough to discriminate.

### How long does a full sourcing cycle take?

The most cited benchmark is 44 days average time to fill across all industries, per SHRM. Technology and engineering roles run longer, 55 to 62 days per LinkedIn data. There is no authoritative fixed SLA per sourcing stage - sources give ranges, not standards. Segment your own time-to-fill by role and seniority, because a company-wide 40-day average can hide an engineering pipeline running at 62 days while admin roles close in 18.

### Is LinkedIn or email better for candidate outreach?

LinkedIn, on first contact. On the exact same candidates, 16.6% replied via LinkedIn versus 4.4% via email - nearly 4 to 1 - because an InMail carries a platform trust signal a cold email lacks. Email still works inside a sequence: Gem's analysis of 4 million recruiter emails found a 76.6% open rate and a 32 to 35% reply rate across a 4-stage sequence. Pick the channel by role type and always sequence.

### What reply rate should I expect from cold outreach?

For cold outreach to passive candidates, 20 to 30% is solid, above 35% suggests strong personalization, and below 15% signals you should revisit message quality and targeting. But watch the trap: a high reply rate from the wrong pool is worse than a lower rate from the right one. Track reply-to-screen, not just reply rate, so vanity replies do not flatter a bad list.

### How do I know if a role's talent pool is too small to source by volume?

Size it in step two before committing a channel. Scarcity is quantifiable: in Refolk's index there are 54,976 US Software Engineers with Python but only 614 with Rust, making Rust 89.5x rarer. A pool that small will never fill by volume outreach the way a 54,976-person pool does. When the number is low, plan for precision and patience, not more messages.

### What is the difference between sourcing, screening, and shortlisting?

Sourcing is the proactive identification and engagement of potential candidates to generate a flow of applicants, done before any opening is even flooded. Screening comes after outreach or application and removes the clearly unqualified. Shortlisting picks the best of who is left - typically 4 to 8 - against written criteria. They run in that order, and collapsing them is how bias creeps in.

---

*From the Refolk guide library. I revise these guides rather than replacing them, so the current version is always at https://www.refolk.ai/guides/talent-sourcing-playbook-2*