Source Effectiveness

Which channels produce hires, not applications. The metric is only as good as the attribution behind it, and attribution is where nearly every source report quietly fails.

RuleOne source of record per application, assigned at first entry, on a documented taxonomy, with the attribution method landed beside it.Neverrank channels on candidate self-selected source alone. Candidates pick it wrong most of the time; it is the lowest rung of the attribution ladder.

◆ Getting attribution honest— the taxonomy, the ladder, the first-touch rule, and the vendor-claim problem.

Decision	What goes wrong without it
A frozen taxonomy	Source categories drift, and this quarter's labels stop matching last quarter's. Pick the list, document it, and when it changes, remap history so trends survive. dim_source is small; keeping it clean is cheap.
The attribution ladder	System-captured attribution, a posting link or a referral link, outranks recruiter-tagged at entry, which outranks candidate self-select. Land which rung produced each row; a source report is only as trustworthy as its worst rung, and now you can measure the mix.
Tag at first entry	Source filled in at offer stage is reconstructed from memory. The application's source is set when the application lands, then left alone.
The first-touch rule	A candidate who applied through a job board but was already in your pipeline from earlier outreach was sourced by the outreach. Decide relationship-start versus last-touch once, and state it on the report.
One hire, many claimants	Every vendor's database contains your hire, and every vendor claims them. Claims reconcile against your source of record; the record does not move to match the invoice.

◆ The owned model, and the SQL— the funnel per source, from the same fact the funnel report reads.

No new fact. The stage-event fact gains a source key, and the funnel groups by it. Volume, quality, and conversion per source in one query:

-- May cohort by source: applied, reached interview, hired
WITH first_entry AS (
  SELECT e.application_key, e.source_key, e.stage,
         MIN(e.event_date) AS entered_on
  FROM fct_application_event e
  JOIN dim_requisition r ON e.requisition_key = r.requisition_key
  WHERE r.requisition_type <> 'Evergreen' AND e.cohort_month = '2026-05'
  GROUP BY 1, 2, 3
)
SELECT s.source_category,
       COUNT(DISTINCT CASE WHEN stage = 'Applied'        THEN application_key END) AS applied,
       COUNT(DISTINCT CASE WHEN stage = 'Interview'      THEN application_key END) AS interviewed,
       COUNT(DISTINCT CASE WHEN stage = 'Offer Accepted' THEN application_key END) AS hired
FROM first_entry f
JOIN dim_source s ON f.source_key = s.source_key
GROUP BY 1
ORDER BY hired DESC, interviewed DESC

Sample output, the same May cohort as the funnel report, now split by source:

source_category	applied	interviewed	quality	hired	conversion
Referral	18	6	33.3%	2	11.1%
Job boards	95	6	6.3%	1	1.1%
Direct outreach	27	3	11.1%	0	0.0%

The rows sum to the funnel report's 140 applications, 15 interviews, and 3 hires; this page is that funnel, split. The reading: job boards deliver two thirds of the volume and one of the three hires; referrals deliver a third of the interviews from 18 applications. Quality, the share reaching interview, is the earlier and steadier signal; conversion swings hard at small hire counts, so read it over quarters, not months. Sample values are illustrative, never client data.

Want source data worth reallocating budget on?

The attribution ladder landed on every application, a frozen taxonomy, and the funnel per source, reconciled against Workday, and you own every line.

Talk to us

Terms on this page

source of record: The one source on the application, set at first entry, never rewritten to match an invoice.
attribution ladder: System-captured over recruiter-tagged over self-selected. Landed per row.
taxonomy: The documented source category list. Frozen, and remapped when it changes.
first-touch rule: Relationship start versus last application. Decided once, stated on the report.
quality: Share of a source's candidates reaching interview. The early, steady signal.
conversion: Hires over applications per source. Swings at small counts; read over quarters.
vendor claim: A source asserting credit for a hire. Reconciled against the record, never the reverse.