Real-Time Bidding in Recruitment Advertising — Lévis

Real-time bidding as a commercial infrastructure emerged from display advertising between 2008 and 2012, when the major advertising exchanges (Google's DoubleClick, AppNexus, OpenX) built standardised auction protocols that allowed any advertiser to bid for any impression in real time, at the moment it was available. The technical standard that emerged — the OpenRTB specification, maintained by the IAB Tech Lab — defined how bid requests and responses should be structured, enabling interoperability between advertisers and publishers that had previously operated in siloed direct relationships.

Recruitment advertising adopted RTB infrastructure more slowly and with significant modifications. The first programmatic recruitment platforms — emerged between 2013 and 2016. They applied RTB concepts to job advertising but encountered structural differences that required adaptation: job postings are not ephemeral impression opportunities but persistent content objects; candidate intent signals are richer and more specific than general browsing behaviour; and the relationship between a click and a downstream outcome (hire) is measured in weeks rather than milliseconds. These differences shaped how recruitment RTB infrastructure evolved relative to its display advertising origins.

The full auction cycle — from job seeker query to winning bid selection — typically completes in under 100 milliseconds. This is enabled by the fact that bid decisions use pre-computed relevance scores and predicted CTR values, not live calculations. The AI bidding agent's real computation happens between auction cycles, when it updates its relevance models and bid strategies based on accumulated performance data.

The bid request contains contextual signals about the job seeker's query: the search terms used, their location (at whatever granularity the publisher provides — which may be city, region, or country depending on the platform), the device type, and in some cases (particularly on Microsoft/LinkedIn integrated inventory) professional profile attributes. The bidding agent evaluates these signals against the employer's job attributes and targeting parameters to compute a relevance score, which it combines with the employer's maximum CPC to determine its actual bid.

Recruitment RTB diverges from general display RTB in four structurally important ways that employers should understand when evaluating platform claims.

The conversion event is distant and complex: In display advertising, the conversion event (purchase, sign-up) can occur within seconds of a click. In recruitment, the conversion chain — click → application → screening → interview → offer → acceptance → start date — spans weeks to months. This makes true outcome-based bidding optimisation (bidding to maximise hires, not clicks) extremely difficult in real time. Most recruitment RTB AI optimises toward application completion as a proxy for hire, which is a reasonable approximation but not identical to the employer's true objective.

The "creative" is structured data, not an image: In display RTB, the creative (banner image, video) is separate from the targeting parameters. In recruitment RTB, the "creative" is the job listing itself — a structured text object with defined attributes. This means that creative quality and targeting relevance are tightly coupled: a well-written job description both improves the relevance matching score (how often the AI wins auctions for relevant queries) and the click-to-apply conversion rate (how many won auctions result in applications). The AI cannot compensate for a poor job description with smarter bidding.

Candidate intent signals are richer: Job seekers' search queries carry explicit intent signals that most display advertising audiences lack. "Senior Python Engineer London £90,000" is a highly specific intent signal that allows recruitment RTB systems to target with far higher precision than demographic or behavioural targeting in display. This precision makes recruitment RTB more efficient per impression than display RTB for employers who configure their campaigns with specific targeting parameters.

Publisher inventory is concentrated: General display RTB operates across millions of publisher sites. Recruitment RTB operates across a much smaller set of job boards, aggregators, and professional networks. This concentration means that a handful of high-quality publishers (Indeed, LinkedIn, major sector-specific job boards) dominate conversion quality, while the long tail of programmatic publishers (smaller job boards, content sites with job widgets) produces significantly lower application conversion rates. Bid allocation across this concentrated publisher landscape is a primary lever for AI optimisation.

The shift from second-price to first-price auctions in programmatic advertising has significant implications for employers setting max CPC bids manually rather than using AI bidding agents. In a first-price environment, a naive strategy of setting your max CPC as your actual bid results in systematically overpaying relative to what a bid-shading agent would achieve. This is one practical argument for using programmatic platforms with AI bidding agents rather than managing bids manually — the AI's bid shading capability can meaningfully reduce actual CPC below the employer's stated maximum.

"AI optimises your spend" is a phrase that appears in virtually every programmatic recruitment advertising platform's marketing. Its practical meaning varies significantly between platforms and is rarely explained in detail. Understanding what a bidding AI actually does — and does not do — helps employers evaluate platform claims more critically.

What programmatic bidding AI does: A recruitment programmatic bidding agent typically performs some combination of: (1) bid shading — bidding below the employer's maximum to reduce clearing price in first-price auctions; (2) publisher allocation — shifting budget toward distribution channels (partner networks, specific publishers within those networks) that historically produce higher application conversion rates; (3) time-of-day and day-of-week optimisation — concentrating impressions during periods when job seekers are more likely to apply (typically weekday evenings); (4) audience segment prioritisation — bidding more aggressively for job seeker signals that correlate with application completion (specific search query patterns, career stage indicators); and (5) budget pacing — distributing spend evenly across a campaign's duration rather than front-loading, to avoid campaign exhaustion before the end of the contracted period.

What programmatic bidding AI does not do: It does not write or improve job descriptions. It cannot compensate for a poor click-to-apply conversion rate caused by a poorly structured application form. It cannot verify candidate quality — it optimises toward application events, not hire quality. And it cannot guarantee a specific application volume — it delivers clicks (in CPC models) or application events (in CPA models) at or below bid levels, but the total application volume depends on the matching between the job's attributes and the available candidate audience, which varies by role type, market conditions, and competition from other employers bidding for the same audience.

The demand-supply stack in recruitment RTB maps the commercial relationships between employers, programmatic platforms, and publishers.

Demand side: Employers (buyers of candidate attention) define their requirements and fund campaigns. Programmatic platforms act as demand-side platforms (DSPs) — they aggregate employer demand, manage bidding strategies, and interface with supply-side infrastructure. In Expertini's case, the platform itself is both a publisher (its own 251-country job seeker network) and a DSP (bidding on behalf of employers in external networks).

Supply side: Publishers (owners of job seeker attention inventory) make their placement opportunities available. For recruitment RTB, supply-side platforms (SSPs) include job board networks, search engine inventory (Google, Microsoft/Bing), and social media platforms. These SSPs aggregate publisher inventory and broadcast bid requests to connected DSPs when a job seeker query triggers a monetisation opportunity.

The exchange layer: In general display RTB, independent ad exchanges (Google Ad Exchange, Xandr, Index Exchange) sit between DSPs and SSPs. In recruitment RTB, this intermediate layer is often collapsed — major platforms like Indeed operate both the supply (their own job seeker audience) and the demand (their own sponsored jobs auction) without an independent exchange. While, programmatic platforms similarly operate integrated SSP/exchange functions within their publisher networks. This vertical integration is common in recruitment RTB and means that the "auction" employers experience is often a platform-internal process rather than an open market auction.

A recruitment RTB bidding agent's core function is to compute the expected value of winning a specific impression — and bid accordingly. The expected value is typically computed as: P(click | impression) × P(application | click) × employer's value of an application. Each component is estimated from available signals:

P(click | impression) — predicted click-through rate: Estimated from: the job seeker's search query (keyword overlap with job title and description), the job seeker's location relative to the job location, the publisher's historical CTR for similar jobs, the time of day and device type, and the employer's historical CTR on the platform. This estimate is computed in milliseconds using pre-trained models, not live calculation.

P(application | click) — predicted application conversion rate: Estimated from: the job category and sector (baseline conversion rates by category), the presence of a salary range (positive signal), the job description word count (positive signal above 300 words), the employer's historical application rates on the platform, and the match quality between the job seeker's inferred profile and the job's experience/education requirements. Some platforms additionally incorporate job seekers' historical application completion rates — candidates who typically complete applications quickly are treated as higher-value impressions.

Employer's value of an application: In a CPC model, the employer's max CPC bid is the explicit signal of value. The AI interprets a higher max bid as authorisation to bid more aggressively across auctions, increasing win rate and impression volume at the cost of higher per-click spend. Setting the max CPC too low relative to competitors for the same candidate audience results in low win rates and potentially insufficient click delivery before the campaign expires.

Expertini occupies a dual role in the recruitment RTB stack: it is simultaneously a publisher (operating its own 251-country job seeker network, making 700,000+ monthly user sessions available as advertising inventory) and a DSP-equivalent (bidding on behalf of employers in external partner networks).

Within Expertini's own platform, the "auction" for sponsored placement is simplified relative to external RTB: sponsored jobs are flagged in the Elasticsearch index with a Expertini_Sponsored_Active marker and receive priority placement in search results. This is not a real-time auction — it is a priority ranking modification. The programmatic RTB model operates in the external partner networks where Expertini bids as a demand-side participant using employer campaign data.

The AI optimisation layer monitors click-to-application conversion rates across all active partner channels for each campaign, dynamically adjusting budget allocation toward networks delivering higher application conversion rates. This reallocation operates at the partner network level rather than at the individual publisher level within those networks — a distinction that matters for performance ceiling. Enterprise platforms with direct API integrations can optimise at the individual publisher level (e.g. allocating more to specific Programmatic Partners member sites that historically convert well for specific job categories). Feed-based integrations optimise at the network level, which is a coarser but still meaningful control.

One important transparency note: because Expertini's click tracker is the single counter of record for all partner network clicks, the employer's dashboard provides a single, consistent view of click delivery regardless of which partner network delivered each click. This contrasts with some programmatic approaches where each partner network reports independently, requiring the employer to manually reconcile click counts across platforms — a process prone to discrepancy due to different definitions of "billable click."