Workable's Recruiting Agents Need a Per-Candidate Budget • The Weighted Average

Workable’s September 14 release makes four recruiting agents generally available to paid accounts worldwide, with action-based credit charges. At its smallest published bundle rate, 930 initial candidate screens represent $111.60 in consumed credits—a useful workload comparison, but neither a reported customer invoice nor the full cost of choosing whom to hire.

The first screen is not the whole funnel

The new lineup covers job briefs, sourcing, screening, and engagement. Workable’s detailed announcement describes agents that can attach to existing jobs, update candidate profiles, and show processing status. Existing customers can therefore test a bounded part of their current recruiting process rather than reconstructing a job merely to try the software. That lowers the procedural barrier to a pilot; it does not establish its accuracy.

The current pricing page starts paid accounts with 3,000 free AI credits. Initial screening consumes one credit per candidate, sourcing consumes two, and the chat meter consumes ten. These are different actions with different business outputs. A low screening price should not be extended to every candidate interaction, and starter credits should not be treated as a recurring monthly allowance: the current offer does not promise that reset.

The published paid bundles are 5,000 credits for $600, 10,000 for $1,000, and 50,000 for $4,750. Their respective unit prices are $0.12, $0.10, and $0.095. Credits are non-refundable and expire one year after purchase. The purchasing decision therefore has two costs: the credits a workflow consumes and the money committed to inventory that may remain unused when the year ends.

For a workload anchor, use Workable’s original CacheFly customer account. It describes more than 930 applications arriving in 36 hours on the company’s previous applicant-tracking system. That chronology matters. The story provides a real scale of inbound demand; it does not report that Workable’s newly priced agents processed those applications or produced a particular invoice.

Take 930 as a rounded scenario, not the exact historical count. Combining that separate customer account with the current tariff gives 930 × one credit × $0.12 = $111.60 in purchased-credit consumption value. The smallest cash purchase remains $600, not $111.60. Starter credits could cover those initial screens, and the calculation excludes the Recruiting subscription, sourcing, engagement, repeat actions, recruiter review, and expired balances. It is a budget building block, not a return-on-investment claim.

Engagement changes that budget quickly. After 930 initial screens, a 3,000-credit starter allocation would leave 2,070 credits, enough for 207 ten-credit chat units with no sourcing included. Do not translate that into 207 fully handled people without confirming the meter. The announcement describes a completed candidate chat, while pricing language refers to each candidate interaction with the chat. The retrieved pages do not resolve whether those expressions identify the same chargeable event.

That ambiguity is commercially material even when each individual charge looks small. Ask whether abandoned conversations, follow-up exchanges, and re-screening after a profile change create new billable actions. A forecast based on one chat per person can fail if the system counts a different boundary. Confirm it before enabling engagement across an entire applicant pool, then reconcile a small observed run against the credits report.

Keep hiring judgment outside the consumption chart

Existing paid customers with high applicant volume and explicit role requirements are the clearest pilot candidates. Start with screening on a current job, retain human control over progression, and review a representative sample of low-scored candidates. The question is not only whether the ranking saves reading time, but whether it misses qualified people or rewards irrelevant signals. No retrieved source provides an independently measured false-negative rate that could justify waiving that check.

The CacheFly account is a testimonial, not a controlled evaluation. It describes a custom question filtering applications and a workflow that helps focus review, but those are not a general accuracy score for the new agents. Avoid turning reduced manual attention into proof that every excluded candidate was unsuitable. The outcome worth measuring is a defensible shortlist with manageable review effort, not a dramatic reduction in the number of résumés somebody opens.

Workable’s MCP help page adds another adoption condition. The connector uses existing user permissions and supports both reading and changing records, including candidate movement, comments, and other authorized HR actions. Review the connecting role before giving an external assistant access. Advice to confirm changes is not evidence that every client enforces a universal approval barrier. The absence of an extra Workable MCP charge also does not make the external assistant’s usage free.

The new AI Credits Report can track consumption by job, action, and period, according to the announcement. Use it to explain where spending goes, then add subscription allocation and recruiter effort separately. A report labeled cost per hire is not automatically a fully loaded labor model. This is the hiring version of our procurement analysis of automation and final-review capacity: faster preparation can leave the consequential judgment as the scarce resource.

Today’s Cornelis lead asks whether improved utilization produces useful work. Workable deserves the same test at human scale. Expand when observed credits, review time, shortlist quality, and candidate handling improve together. Stop if engagement charges surprise the team, qualified applicants disappear below the cutoff, or broad permissions create avoidable risk. The cheap first screen earns a pilot; an accountable hiring process earns continued spending.