HIHB vs. AI recruiting - why the briefing comes before the AI.

1The thesis: AI needs a briefing

AI recruiting in 2026 is not the future anymore - it's the present. Sourcing tools, matching engines, automated pre-filters, AI-assisted interview scoring: the toolset is mature enough to relieve recruiting teams in measurable ways. Ignore it and you lose speed. Apply it blind and you lose top candidates while accumulating regulatory exposure.

The simple truth lost in every second marketing deck: AI optimises against the goal it is given. If the goal is a half-built requirements list, a keyword filter on CVs, and an unspecified persona picture, AI will optimise against exactly that - very efficiently, to the wrong result.

Across more than 200 HIHB workshops with management boards, executive teams, and hiring managers, the same finding has repeated: critical hires don't fail at the tool, they fail at the briefing. With AI in the loop, that mechanism doesn't weaken - it accelerates and becomes harder to correct.

2What AI recruiting does excellently

Before drawing the line, an honest appreciation. AI recruiting is not replaceable when it comes to three things:

• Volume. Reviewing thousands of applications in hours, aggregating CV data points, removing duplicate profiles, checking formal minimum requirements - AI does this better and cheaper than any human.
• Matching. Cross-referencing profiles from internal talent pools and external databases against a defined search profile - AI finds candidates a recruiter misses after three keyword-search iterations.
• Sourcing scale. Personalised LinkedIn outreach at scale, A/B-tested message variants, sentiment analysis on replies - operations that are uneconomic by hand.

For this work, AI is an accelerator. Recruiting teams that apply it consistently gain time for the work AI cannot do: personal conversations, expectations negotiation, cultural judgement. Hiring-manager enablement benefits when AI takes the routine and the human concentrates on the decision.

3Where AI goes blind: four verified data points

If AI scales this precisely, hit rates should look correspondingly strong. They don't. Four verified data points draw the picture - and deliver the sharpest argument of this whole comparison series.

Those 88% are not an algorithm failure. They are a briefing failure, automated. If the briefing contains no real persona and substitutes keyword lists instead, the ATS filters exactly against those keyword lists - and rejects qualified candidates who name the keyword differently or have a six-month parental leave.

Those 38% are the tip of the iceberg. Top candidates with multiple options decide early - and they decide against the firm that runs the pre-filter, the interview scoring, and the offer phase as a purely algorithmic experience. Recruiting gains scale and loses the applicants it wanted to win.

Amazon is the canonical story. It is not closed - it repeats in smaller form whenever a recruiting team trains an AI on its own historical hires. The system learns what was chosen before and reproduces the biases that went unspoken in the briefing process. Where diversity starts as a briefing gap, it ends as an algorithmic norm.

Anyone using AI in recruiting in 2026 should mark August on the calendar. Three months after the publication date of this article, the enforcement phase is live. Companies that have not documented by then how their AI optimises against which briefing using which data risk fines on a scale that wipes out any efficiency gain from the tool.

Alongside this, the ongoing US case Mobley v. Workday: five plaintiffs over 40 were rejected by Workday's AI within minutes across hundreds of applications. In July 2024 the second motion to dismiss was denied; conditional class certification followed. The opt-in window for the ADEA class action ran until 7 March 2026 - potentially millions of class members.⁵ This is the largest AI-recruiting lawsuit in motion. Workday's own audit found "no disparate impact"; the plaintiffs' analysis of the same numbers found a statistically significant disparity.

4Three axes: Quality of Hire, time-to-fill, risk

HIHB and AI recruiting are not in competition. But on three axes, their mechanisms compare cleanly.

Axis 1: Quality of Hire

Quality of Hire measures whether the person delivers against expectations after 12 to 24 months. AI can optimise against any skill list. It cannot decide whether the skill list is the right one. Persona instead of a requirements list is a briefing decision, not an algorithm output. With the persona missing, AI optimises for a candidate type that fits formally and fails culturally.

HIHB acts here before the AI: Step C-3 (Calibration) defines a scorecard with 3-, 6-, and 12-month criteria. That scorecard is the only briefing AI can sensibly take as an optimisation target.

Axis 2: Time-to-fill

AI shortens time-to-fill at every step where volume is the bottleneck: sourcing, pre-filter, scheduling, interview transcription. In steps where information is the bottleneck - expectations alignment, stakeholder mapping, political reading - it remains ineffective. Driving time-to-fill down with AI alone simply pushes the problem into the later performance phase.

Axis 3: Risk distribution

Classic AI tools distribute risk between vendor, company, and applicant. With the EU AI Act, the structural balance shifts back to the company: without a documented briefing, without bias testing against a defined target, without human oversight, liability anchors with the firm using the system. A weak briefing base is no longer just a quality problem from August 2026 - it's a regulatory one.

HIHB tackles the risk before the AI. A documented briefing with explicit persona, breakpoints, and evaluation criteria is precisely what bias testing and human oversight need to satisfy the AI Act. The briefing becomes a compliance document.

5The HIHB reading: before the AI

HIHB is not an anti-AI stance. HIHB is the step before. The AI gets a precise briefing through HIHB - and therefore a meaningful optimisation function. The practical difference:

• Persona input, not skill keywords. The AI doesn't get "SAP experience, 10+ years" as a filter criterion but a persona definition with life phase, motivation, risk appetite. From that you derive a sensible search profile rather than a keyword filter that rejects 88% of qualified candidates.
• Scorecard as objective function. The AI optimises against the 3-, 6-, and 12-month criteria signed off by the hiring manager - not against historical hires, which may carry cultural bias (as Amazon learned in 2018).
• Breakpoints as filters. The breakpoint definition from the workshop hands the AI explicit risk filters that don't have to come from training data - something a pure data-driven AI cannot derive on its own.
• Human oversight by default. In the HIHB process there are pre-defined points at which humans override the AI decision. That is not only AI Act-compliant; it is also the signal the 60% of human-seeking applicants are looking for.
• Briefing as compliance document. The HIHB briefing documents which criteria were searched against, why, with what bias filters and which stakeholder distribution. Exactly what the AI Act requires from August 2026.

Work in this sequence and you gain speed from AI and quality from briefing. Reverse the sequence - AI first, briefing retroactive - and you gain only speed to the wrong hire. A longer discussion of that dynamic in the AI recruiting backlash.

6What to do: three micro-steps before the AI runs

If an AI tool is going live in the recruiting process tomorrow, three steps first.

Step 1: persona before keyword filter (15 minutes)

In one sentence, define who the person is you want to win - life phase, motivation, conflict style, energy source. Hand that persona description to the AI as context, not just the skill list. If your tool doesn't support this context format, it is the wrong tool for critical hires.

Step 2: three breakpoint sentences as AI filters (15 minutes)

Write three sentences that begin with "If this person does …, the hire has failed." Bring those three sentences into the AI process as explicit filter criteria - as pass/fail questions in the pre-filter, not as scoring weights running in the background. Top candidates recognise honest filters; the AI documents the decision for later AI Act compliance.

Step 3: a human in the path (10 minutes)

Mark two points in the AI workflow where a human can override the decision - pre-filter and short-list. Communicate those two human checkpoints actively to applicants. The 60% who want human contact see them early enough not to walk.

Together, the three steps cost 40 minutes. They lift AI use out of pure scale-tooling into a documented recruiting system - robust against the data points above, robust against the AI Act.

AI is here to stay. Briefings are the lever that decides what it optimises for.

Michael von Hirschfeld

Co-Managing Director, HireWorks GmbH · HIHB Workshop facilitator · 200+ workshops

Leads the HIHB methodology at HireWorks. Has facilitated workshops since 2018 with founders, management boards, recruiting teams, and hiring managers across DACH mid-market firms, large corporates, and start-ups.