Workable Priced Every Bad Applicant at $0.10. Now What?

On September 14, 2026, Workable moved its four Recruiting Agents (Job Brief, Sourcing, Screening, Engagement) to general availability on a credit meter, and shipped MCP integration for Claude and ChatGPT the same day. Four days earlier it released an AI Credits Report that puts cost per hire, by agent stage, on a dashboard your CFO can read. If you run TA, the sourcing line on your invoice is about to look nothing like it did in August.

What actually changed on September 14

Workable became the first mainstream ATS to price AI sourcing per action, with one credit per candidate processed and a public rate card. Every plan includes 3,000 free AI credits, then top-ups run $0.12, $0.10, or $0.095 per credit depending on pack size.

The four agents that now burn against that meter:

• Job Brief Agent turns a manager conversation into a structured Ideal Candidate Profile.
• Sourcing Agent finds candidates against that profile.
• Screening Agent scores every applicant across 14 dimensions.
• Engagement Agent opens a personalized chat to fill in missing data, then re-scores.

Two supporting releases matter as much as the agents. The MCP Server, available at no additional cost on all subscription plans, connects the agents to Claude and ChatGPT and respects existing Workable user permissions. And the AI Credits Report, released September 10, gives enterprises a single view of credit consumption across every recruiting stage with cost per hire as a core metric.

Plan gating is the piece nobody is talking about. AI sourcing is locked to Standard and above, so Pay-per-Job and Starter customers are stuck with applicant tracking only. The entry point for proactive candidate finding on Workable is now $299 per month plus whatever credits you burn on top.

The unit economics of $0.10 per candidate

At the mid-tier rate, every candidate the Screening Agent touches costs roughly $0.10, which sounds trivial until you multiply it by the size of a modern inbound wave. Workable's own release cites CacheFly, a 30-person CDN, receiving 930 applications for a single HR role in 36 hours. That is about $93 of agent spend to triage one requisition, or 1.7% of the $5,475 average non-executive cost per hire from SHRM's 2025 Benchmarking Report.

That number looks small in isolation and enormous at portfolio scale. A team running 200 open reqs at CacheFly-like inbound volumes is metering $18,600 in a week and a half, before a single interview happens.

930

Here is the table your finance partner will want on Monday.

The mechanism to watch is not the headline rate. It is what the Screening Agent does per credit. It evaluates each applicant across 14 dimensions, and where information is missing the Engagement Agent opens a personalized conversation, asks targeted follow-ups, then re-scores. Every unqualified applicant still burns a credit, and often a follow-up cycle on top. Your JD quality, your channel mix, and your ICP tightness just became P&L levers.

Why credit pricing punishes low-precision top of funnel

Credit pricing does not just tax volume, it taxes imprecision. In a fixed-subscription world an over-broad job ad was free to leave up; under a meter, every unqualified applicant it attracts is a receipt.

Three behaviours that were harmless in August are expensive in September:

1. Job boards on autopilot. If Indeed is dumping 500 unqualified applicants a week into a req, the Screening Agent is billing you to reject them.
2. Copy-paste JDs across similar roles. Ambiguous requirements produce ambiguous scoring, which triggers Engagement Agent follow-ups, which burn more credits.
3. Keeping stale reqs live for "pipeline." Inbound keeps arriving, credits keep ticking.

The counter-move is outbound, which is where the math flips. A tightly sourced list of 40 named engineers costs $4 to score if you route it through Workable at all, and most of that work should not happen inside an ATS meter to begin with. Describing the person you want in plain English and getting a ranked shortlist back is the exact gap Refolk closes; you spend credits on candidates you already believe in, not on a firehose you have to filter.

The AI Credits Report is a finance weapon

The AI Credits Report is the first ATS dashboard that gives procurement CFO-legible cost-per-hire attribution by agent stage. That single design decision changes the political dynamic of every renewal conversation from here forward.

Until now, TA has negotiated software the way marketing negotiates HubSpot: seats, tiers, features. Consumption reporting drags recruiting into the same conversation cloud infrastructure has had with finance for a decade. Expect three follow-on effects inside enterprise buyers:

• Agent-level ROI at renewal. Procurement will ask which of the four agents actually moved cost per hire, and the ones that did not will get turned off.
• Chargeback to hiring departments. Once the report exists, someone in FP&A will notice engineering is burning 60% of the credits and start allocating cost by cost center.
• Kill-switch thresholds. TA leaders will set per-req credit caps, because "unlimited AI screening" is a phrase no CFO signs off on twice.

Consumption reporting drags recruiting into the conversation cloud infrastructure has had with finance for a decade.

The buyer set that will feel this first is small and concentrated. In Refolk's index, there are 2,761 Heads, Directors and VPs of Talent Acquisition across the US and UK. Those are the roughly 2,700 inboxes about to receive a credit invoice with their name on it.

MCP plus Claude is a shadow-sourcing Trojan horse

The MCP Server is the release nobody costed properly. It lets any hiring manager instruct Claude or ChatGPT to source candidates, update pipelines, or manage records without leaving their chat interface, with no developer involvement in setup.

Read that carefully. The recruiting team owns the credit budget. The hiring managers now have a natural-language interface to spend it. Governance is downstream of billing, which is not where you want governance to live.

The three failure modes to design against before Q4:

• Duplicate sourcing. A hiring manager and a sourcer run overlapping Claude prompts against the same req and double-bill the meter.
• Prompt sprawl. Every manager writes their own ICP in their own chat window, and the Job Brief Agent's structured profile gets bypassed.
• Silent scope creep. Managers ask Claude to "just take a look" at 500 candidates in the ATS, unaware each look is a credit.

Workable's meter is cleaner than Salesforce's, and buyers know it

Workable shipped one pricing model plus a reporting dashboard on day one; Salesforce shipped three Agentforce pricing models in 18 months and still runs all three simultaneously. In a market where 43% of enterprise buyers prefer consumption-based pricing (Futurum's 1H 2026 Enterprise Software Decision Makers survey), Workable is the first ATS aligned to how buyers actually want to buy.

The competitive read for the rest of the category:

• Greenhouse, Ashby, iCIMS will be asked to disclose comparable meters at every RFP from October onward. "We charge per seat and the AI is included" stops being a differentiator and starts sounding like an accounting dodge.
• HighRadius already went further, ditching per-seat entirely at its February 2026 Radiance conference with zero implementation fees and payment only as a percentage of achieved savings. That is the outcome-based extreme Workable stopped short of, and the direction procurement will push next.
• Salesforce Agentforce is the cautionary tale: $2 per conversation at launch, Flex Credits at $0.10 per action in May 2025, per-user licensing from $125/month by late 2025, all three co-existing and confusing every buyer.

The scarce resource just flipped from seats to credits

Credit pricing means one TA leader with a tight ICP and a good prompt can out-produce a five-sourcer team, because the constraint moves from headcount to prompt quality. The numbers underneath that claim are stark.

0.59

Where those 1,634 US sourcers actually work skews to a familiar shortlist of employers: Pinterest, Rippling, MongoDB, Anthropic, Verkada, Zoox and EvolutionIQ, concentrated in the SF Bay Area. Those are the accounts most likely to model credit unit economics first, because they already staff the analytical muscle to do it.

For teams that do not have five sourcers to redeploy, the practical move is to shrink the top of funnel before it hits the meter. Sourcing tools that operate outside the ATS earn their keep here: describe the person in plain English across GitHub, LinkedIn and the open web, and the ranked shortlist is what enters Workable, not the raw wave.

What to do in the next 30 days

Audit your inbound, tighten your ICPs, and move as much scoring as possible outside the meter. Everything else follows from those three moves.

Concrete actions for the next four weeks:

1. Pull baseline credit burn per req for the last 14 days from the AI Credits Report and rank reqs worst-to-best on credits per hire.
2. Kill or rewrite the top three offenders. These are almost always over-broad senior IC roles on job boards.
3. Set a per-req credit cap and wire it into your ATS notifications so managers see it before they hit it.
4. Move outbound sourcing off the meter. Run named-candidate sourcing in a tool that is not billed per candidate processed, then hand the shortlist to Workable for engagement.
5. Rewrite MCP access rules. Decide which hiring managers get Claude-driven ATS access, and log every session against a budget owner.
6. Recompute your CPH. Add agent spend as a line and share it with finance before they surface it themselves.

The teams that treat September 14 as an accounting event, not a product update, will be the ones still in budget in Q1. The rest will spend the winter explaining a variable line item they did not know existed in August.

FAQ

How much does Workable's new AI actually cost per candidate?

At the mid-tier rate, one credit per candidate processed works out to about $0.10, with cheaper rates ($0.095) on the largest top-up packs and $0.12 on the smallest. Every plan includes 3,000 free credits per month, which for a small team covers a normal inbound month, and for a high-volume enterprise disappears in a single well-publicised requisition. The meter counts every candidate the Screening Agent touches, qualified or not, which is why the effective cost per hire is a function of your funnel precision rather than the sticker rate.

Which Workable plans include the Recruiting Agents?

AI sourcing is locked to Standard and above. Pay-per-Job and Starter customers get applicant tracking only, so $299 per month is the effective floor for anyone who wants proactive candidate finding on Workable, before any credit top-ups. If you are currently on a lower tier and were planning to test the agents, that is a plan change plus a variable-cost commitment, not a feature toggle.

Does the MCP integration cost extra?

No. The Workable MCP Server is available at no additional cost on all subscription plans and respects existing user permissions, but the actions Claude and ChatGPT take through it still consume credits from your Workable balance. The governance risk is not the connector fee, it is that hiring managers with chat access can now spend the recruiting team's credit budget without touching the ATS UI. Set access rules before you set adoption targets.

How does this compare to Greenhouse or Ashby's AI pricing?

Neither has shipped an equivalent public per-action meter yet, which is exactly the pressure Workable's release creates. Greenhouse and Ashby currently bundle AI features into tiered subscriptions, which reads as customer-friendly until a buyer asks what the marginal cost of an extra 10,000 screenings is and gets no straight answer. Expect comparable meters, or at minimum comparable consumption dashboards, in the next two RFP cycles as procurement teams cite Workable's AI Credits Report as the disclosure baseline.