The Database Tax.

A typical Japan agency placement made through one of the major per-placement-fee databases — BizReach, Recruit Direct Scout, doda X, AMBI, and similar agency-side scout platforms — carries a database usage fee of ¥700K–¥1.05M, 20–30% of the agency’s recruiting fee. That same yen, deployed through Headhunt.AI credits at the enterprise rate of ¥63.75 per qualified match, surfaces between 10,980 and 16,470 qualified candidates from a 4M+ Japan-focused database — ranked, scored, ready to evaluate.

That is a five-figure spread on the most expensive cost line in your sourcing operation. More importantly: those 13,000+ candidates have not been touched on a per-placement-fee platform. Outreach goes out cold through your own LinkedIn, your CRM, your existing infrastructure. No tracked-contact record. No new contingent fee exposure.

Math: ¥875,000 ÷ ¥63.75 per match = 13,725 candidates. At 30%, the spread rises to 16,470. At 20%, it is 10,980. The floor is 10,000+ — case-independent.

Put Headhunt.AI in front of every search. Watch per-placement spend fall.

02Where the agency fee pool went.

The dominant Japan recruiting database is BizReach, operated by Visional Inc. (TSE: 4194). Visional’s own TSE filings describe the scale of the per-placement-fee category at the platform side. The numbers below come from the company’s March 2026 disclosure of H1 FY7/2026 results — read them as scale, not characterization.

Source: Visional H1 FY7/2026 results, TSE filing dated 17 March 2026. BizReach segment H1 revenue ¥38.3B (+19.2% YoY), 42.7% adjusted operating margin before corporate allocation. Standard headhunter plan: ¥600K per six months, plus per-placement performance fees. BizReach’s public materials report 41,800+ direct employer subscribers and 9,700+ active headhunter accounts.

The platform side compounded for over a decade on a structurally similar input: agency and corporate fees collected on placements of candidates the platforms themselves do not employ, train, or place. The candidates are public. The agencies do the work. Each agency placement made through these platforms is, by definition, a placement that could have been made another way. Until very recently, the throughput cost of working those alternatives made the database channel the rational primary path for many agencies. Section 06 explains what changed.

The platform’s margin is not the issue. The structural cut from your P&L is.

03Three layers of cost, not one.

Most agency principals price their per-placement-fee database costs as two line items: the subscription fee and the per-placement usage fee. There is a third layer — and it is the one that compounds quietly across thousands of outreach actions over a contract year.

1. Subscription — ~¥1.2M / yr

   The standard headhunter plan on the dominant Japan platform is ¥600,000 per six months, paid lump-sum in advance regardless of placements. Smaller platforms run lower; enterprise tiers run higher.

2. Per-placement usage fee — ~¥700K–1.05M

   20–30% of the agency’s recruiting fee, paid to the operator on every placement of a candidate sourced through the platform. Roughly ¥700K–¥1.05M on a typical ¥10M base role at a 35% agency fee.

3. Tracked-contact contingent exposure — every scout mail sent

   Most standard Japan recruiting database agreements include audit and attribution provisions under which a placement traceable to outreach activity on the platform may incur per-placement fees, regardless of the channel through which the candidate ultimately responded. Terms vary by operator; the category-level structure is consistent.

4. Annual category spend, mid-size desk — ¥8M–22M+

   Subscription + ~10–20 placements/year × per-placement fee, plus residual exposure on under-attributed placements.

The first two layers are predictable. The third — the ongoing claim against placements that ever trace back to a tracked outreach — is not. That third layer does not exist on Headhunt.AI.

Sourcing cost is paid once. Or paid every time the candidate moves.

04The cost of getting the contract math wrong.

On 20 January 2023, For Startups, Inc. (TSE: 7089) — a publicly listed Japan recruiting agency — disclosed a retroactive restatement covering FY3/2018 onward, recognizing ¥402 million in additional cost of sales owed to multiple human resource database operators. ¥118M was unpaid principal. ¥283M was contractual damages — a 2.4× penalty.

The cause, in the company’s own words, was a 規約の誤認 — a misinterpretation of fee-attribution terms in cases where a single career-changer was simultaneously registered on multiple databases. Under several operators’ contracts, fees could be owed to each operator the candidate had touched, not only to the one through which the placement was finalized. The company resolved it through discussions with various database operators.

Source: For Startups, Inc. timely disclosure (TDnet), 20 January 2023; FY3/2023 Q3 earnings call. Counterparties described as "各求人データベースの企業" (each of the job database companies, plural) and resolved through "各社と相談した結果" (consultation with each operating company).

This is the only public, on-the-record example of what an agency’s contingent exposure to per-placement-fee databases looks like once audited and exercised. Most peer firms have similar exposure profiles. They are simply not yet under securities-disclosure pressure to surface them.

What changed: throughput, not access.

Per-placement-fee databases became one of the dominant Japan agency sourcing channels through the 2010s for a defensible reason. Public profile data on LinkedIn was always the largest pool, but working it at the throughput an active desk requires cost more recruiter labor than the database alternative cost in fees, for many roles. The cost-per-qualified-candidate math made these databases the rational primary channel for a lot of agencies.

In the last 24–36 months that calculation has shifted. AI scoring against full profiles at scale is now viable. The episode above is no longer a cautionary tale about a specific counterparty — it is a cautionary tale about over-concentration in any single sourcing channel during a period when the AI-first alternative did not operate at this throughput. That period has ended.

The exposure is the cost of channel concentration. The fix is channel diversification.

05¥608K more margin, per placement.

Forget the platform. Walk through one ¥10M placement, two ways. Same client. Same fee from the client. The only thing that changes is which sourcing layer surfaced the candidate. Take BizReach — the dominant per-placement-fee platform — at the typical 25% rate, against a Headhunt.AI-first path.

At 25% — the midpoint of the 20–30% range — the spread is +¥608K per placement. A 12-biller desk doing 90 placements converts that into approximately ¥55M of annual margin recovered.

Same fee. Different margin.

06What changed. And what didn’t.

If qualified candidates are the question, most principals have one answer: the per-placement-fee database’s registered pool. But qualified candidates have always sat in three places. What changed in the last 24–36 months is not the pool — it is the throughput of working it.

Public profile data isn’t new.

LinkedIn has been the largest pool of working professionals in Japan for years. Every senior recruiter knows this. What was hard, until recently, was the throughput of actually working that data — Boolean strings, manual longlisting, hand evaluation of tenure and company-tier patterns. Working that pool by hand cost more in recruiter labor than the per-placement-fee database alternative cost in fees, for many roles.

That is the honest version of why per-placement-fee databases became one of the dominant Japan agency sourcing channels through the 2010s. It was not that the candidates weren’t elsewhere — they always were. It was that working them elsewhere was more expensive in recruiter labor than the database fee was in cash.

AI scoring is what is new.

Production-quality scoring against full profiles, in Japan, with bilingual signal evaluation, tenure-pattern analysis, and company-tier sequencing has only become viable at scale in the last 24–36 months. The same researcher who used to evaluate a handful of qualified profiles per week of effort can now evaluate thousands in a few minutes, with consistent JD-relative scoring. The pool was always there. The throughput is what is new. This is what changes the cost equation against per-placement-fee databases that were once cheaper per qualified candidate.

The pool was always there. The throughput is what changed.

Three layers, all now surfaceable.

Headhunt.AI works against three layers of candidate data. All three were surfaceable in theory before. The economics of surfacing them are what shifted.

Layer one. Public profile data. Public professional profile data — LinkedIn primary, with public signals from X (formerly Twitter), GitHub, Facebook, and Instagram layered in where candidates are active there — is the largest, most current dataset of working professionals ever assembled. Headhunt.AI sits on 4M+ Japan-focused profiles, refreshed continuously. The competitive question is not whether the data exists; it is whether your search method actually surfaces it. Most senior Japan candidates do not write profiles in keywords Boolean queries match. They are in the data. They never come up in a Boolean search.

Layer two. AI enrichment on top. Headhunt.AI scores tenure pattern, company-tier sequence, career-trajectory inflection, bilingual signal in surrounding context, and adjacent-industry relevance — fields the underlying profile doesn’t expose. The AI evaluates each candidate against the job in full sentences. The Boolean query evaluates against a string.

Layer three. Your own ATS. Connect Headhunt.AI to your ATS and the same scoring runs across every candidate your firm has ever sourced. The placements you paid per-placement-fee databases to surface over the past decade are still in the system — never re-evaluated. Headhunt.AI re-scores them against the new role. Candidates you already paid to source become re-placeable on a new search. No new database contact. No new contingent fee.

The candidates didn’t disappear. They are sitting in your ATS — paid for once, never re-evaluated since.

07Flag the exposure. Before the placement.

Most per-placement-fee databases include an ownership window — varying by operator and contract — during which a candidate one of your recruiters has touched on the platform carries a contingent fee, regardless of which channel ultimately produces the placement. The third cost layer in section 03 only becomes a surprise invoice because no one in your firm can see, candidate by candidate, who is still inside which window.

ATS integration, every shortlist.

When Headhunt.AI integrates with your ATS, every candidate already in the system gets re-scored against the new role. The integration also surfaces whether you or your team has logged any prior contact through a per-placement-fee database — which platforms, when, and whether the candidate falls inside that operator’s typical ownership window. The output is a column on the shortlist: clean, exposed to one operator, or exposed to multiple — the For Startups scenario in section 04, made visible in advance.

Build shortlists deliberately.

With the exposure column visible, your recruiters build every shortlist with full information. Include a single-operator-exposed candidate when the fit justifies the contingent cost. Filter out multi-exposed candidates entirely on roles where the math doesn’t work. Either way, the decision is deliberate. The third-layer cost stops compounding the moment your team has visibility into it — not three months later, when an invoice arrives.

Audit defense, built in.

When a database operator’s account team eventually sends a fee adjustment notice covering the past 24 months — and at the scale of this category, that is a question of when, not if — your finance function has line-by-line attribution. Every prior contact, every platform, every timestamp. The For Startups outcome in section 04 doesn’t happen to firms operating with this visibility. It happens to firms operating in the dark.

What For Startups disclosed cost ¥402M to surface. The same exposure now shows up as a column on your shortlist.

08AI sourcing first. Per-placement databases as fallback.

The argument is not that you should cancel your per-placement-fee database contracts tomorrow. For most desks, that would be premature — these platforms have uses, particularly in active candidate flow and segments where AI scoring genuinely struggles. The argument is sequencing.

1. Run Headhunt.AI on every search, first.

   Paste the JD, get the ranked list, surface candidates that Boolean queries on per-placement-fee databases would never raise. Reach out cold through your own LinkedIn account, your CRM, or whatever outreach infrastructure you currently use — nobody on the per-placement platforms has touched them.

2. Reach the top-decile candidates before anyone else does.

   The AI-surfaced candidates are, by definition, the ones not currently being approached by every other agency hammering the same Boolean queries on the same databases. Reply rates run structurally higher.

3. Fall back to your per-placement-fee databases only on the residual.

   If the AI list does not contain enough qualified candidates — which happens on niche searches and narrow technical specialties — your existing platform contracts are still there. You just use them less.

Desks running this sequence should expect platform-attributed placements to drop materially within two to three quarters. The subscription becomes a smaller, justifiable line item. The third-layer contingent exposure stops compounding.

Tools earn their keep, or they don’t. Use them when they do.

09Where the recovered margin goes.

The spend that comes off the per-placement-fee database line does not disappear — it becomes deployable capital. Three places it tends to go, and one place it should not.

Where it should go.

1. Senior recruiter compensation.

   The highest-leverage of the three. A senior biller at typical Japan agency comp shares (35–45%) takes a meaningful slice of every additional placement; the firm captures the rest. Reinvesting recovered database spend into senior comp produces near-term retention and output gains.

2. Researcher and associate hires.

   The AI layer absorbs the candidate-identification work that used to consume roughly 50% of researcher time. The same researcher headcount can now support 2–3× the billing volume. Adding researchers — instead of replacing them — is the cleanest way to compound the productivity gain.

3. The agency’s own candidate database.

   Every search you run through Headhunt.AI scores and structures candidates into a workspace your firm controls. Over 6–12 months, that workspace becomes a structured, AI-scored asset that per-placement-fee databases never give you access to. The data is yours.

Where it shouldn’t go.

It should not become invisible. Recovered spend absorbed into general operating budget without a tracked line item disappears in the quarterly noise. Treat it as a separate budget category for the first 12 months and report against it monthly — otherwise the saving compounds for the firm but not for any individual scorecard.

The worst place to send your firm’s margin. The best place to be visibly pulling it out of.

10Common objections we hear.

Seven questions we have heard from peer agency principals across the last twelve months. Each gets a direct answer, not a deflection.

On the limits.

AI sourcing does not make per-placement-fee databases obsolete tomorrow. There are roles where active-intent filtering on a registered candidate base genuinely produces a better list — particularly volume mid-market roles. Use the platforms there. The argument is structural advantage in specific segments, not a universal claim.

Lead with AI first. Fall back to per-placement databases on the residual.

11A test you can run on a stuck role.

Everything in this brief is theory until your team puts it in front of a real role on your desk. Two minutes. Same-day answer.

1. Buy a ¥75,000 credit pack.

   500 credits = up to 500 qualified candidate matches scoring 50+ on the ESAI Score. No subscription. No contract. Credits never expire.

2. Pick the hardest open role on the desk.

   The honest test is the role you have already exhausted on your existing database. Mid-market and contingent roles in segments where AI scoring works well — bilingual finance, IT, sales, HR, marketing, GTM, ops, most engineering — produce the cleanest signal.

3. Compare top 100 to your existing pipeline.

   The single binary question: are there candidates on this list my team has not already approached? If yes — even one of one hundred — you have your answer. ¥75,000 spent. Decision made.

¥75K spent. Fresh new candidates returned.

12Seven diagnostic questions.

The right starting question is not "should we cancel our database contract." It is a set of operational diagnostics on database concentration risk. The point is to find which questions you can answer in numbers and which you can only answer in generalities.

1. Do you know what percentage of your placements last quarter were attributable to candidates first sourced through your per-placement-fee databases, vs. candidates sourced through other channels — measured per recruiter, not estimated firm-wide?
2. Do you know your annualized total per-placement-fee database cost across subscriptions and per-placement fees, expressed as a percentage of recruiting revenue? Most principals quote 10–15%; the actual figure, once subscription, per-placement fees, and contingent exposure are added together, is often higher.
3. Have you mapped your contingent fee exposure on candidates messaged through any of your platforms but placed via another channel? If yes, what is the dollar range? If no, this is the third cost layer your P&L is not currently pricing.
4. If a database operator’s account team called tomorrow with a fee adjustment notice covering the last 24 months, would your finance function be able to defend your reported attributions line by line — or would you settle?
5. What fraction of your shortlists last quarter were sourced from passive candidates not actively registered on any job platform? If below 50%, your sourcing is competing with every other agency hammering the same active pool.
6. Have you run a structured AI sourcing test on a real open requisition in the last 12 months — or is your current view based on vendor demos and conversations with peers?
7. If a peer firm announced 15% lower per-placement-fee database spend with the same number of placements tomorrow, what is your concrete response?

13The honest take.

Japan’s third-party recruiting databases — of which BizReach is the largest — built a real market. They created the infrastructure that allowed the country’s mid-to-high-end placement business to scale through the 2010s. Some of that infrastructure remains genuinely valuable. Some of it is a tax that compounds on agency P&Ls because, until recently, the AI-first alternative did not operate at this throughput.

The alternatives now exist. The structural cost gap between AI-first sourcing and per-placement-fee platforms is wide enough that even partial migration produces meaningful margin recovery. The contingent exposure layer — the one most agencies have not priced — stops compounding the moment you stop sending tracked outreach through these platforms on candidates you could have reached another way.

None of this is a moral argument about any one operator. The operators are businesses doing what businesses do — they capture margin where the structure lets them. The question is whether your agency is going to keep being a line item on the category’s revenue statement, or whether you are going to test the alternative this week and decide for yourself.

This is uncomfortable to read. It is more uncomfortable to act on. Doing nothing is a decision, the same as any other. It just looks more like the present, which makes it feel safer than it is.