Services / S2 · Test

AI Ethics Audit & Bias Testing.

Where the Readiness Assessment asks “where do we stand?”, the audit asks “how do our models actually behave?”. We measure fairness quantitatively with established statistical methods, and explain where and why your models decide systematically differently — in evidence you can put in front of an EU customer or a regulator.

Investment

from €8,000

Duration

3–6 weeks

Effort

60–120 hours

Who it’s for

You operate a system that decides about people.

The audit is the right step when you run a concrete high-risk system — recruiting, credit scoring, recommendation, automated decisions — and either face a conformity step, an EU customer’s due diligence, or an internal suspicion that a production model is treating groups differently.

01

The system is in production

It’s making real decisions about real people — not a prototype. That’s where measurable fairness, explainability, and data quality stop being academic.

02

Evidence is required

A buyer, a board, or a conformity step needs documented proof — not an assurance. The audit produces an artifact built to withstand that scrutiny.

03

No internal testing capacity

You lack the statistical methodology or the tooling in-house. We bring both; you bring data and domain knowledge.

How it runs

Three phases, reproducible by design.

Every claim in the audit is traceable to a calculation. Method beats charisma — in consulting as in science.

A

Week 1 · 15–25 h

Scoping & data access

We define the system under test, the protected attributes that matter for your context, and the fairness definitions that fit the use case — because there is no single “fair”, only fairness against a stated criterion. We set up secure, reproducible access to the data and model outputs.

What you’ll have

  • Audit scope and fairness-criteria definition
  • Protected-attribute and subgroup plan
  • Reproducible data/model access setup

B

Weeks 2–4 · 30–70 h

Measurement & explainability

The core technical work: fairness metrics across protected attributes (statistical parity, equal error rates, equalized odds), an intersectional analysis across combined attributes, and a SHAP-based explainability study of which features drive which decisions. We assess data quality in parallel, since most bias originates upstream in the data.

What you’ll have

  • Fairness metrics across protected attributes
  • Intersectional bias analysis
  • Explainability report (SHAP-based)
  • Data quality assessment

C

Weeks 5–6 · 15–25 h

Reporting & remediation

Two reports — one executive, one technical — stating findings, severity, and concrete remediation options. We hand over a reusable fairness monitoring dashboard so your team can re-run the checks, and we discuss whether Continuous Monitoring is the right next step for the post-market obligation.

What you’ll have

  • Bias detection report (executive + technical)
  • Prioritized remediation recommendations
  • Fairness monitoring dashboard (Python / Streamlit)
  • Handover and next-step recommendation

Investment

Scoped to the system under test.

A single-system audit typically runs 3 to 6 weeks. Effort scales with model complexity, the number of protected attributes, data accessibility, and the depth of explainability required.

Portfolio · custom

Multiple systems

Several high-risk systems, complex pipelines, or regulated-industry depth. Priced after a scoping call; often combined with a Monitoring retainer.

Custom estimate

120+ hours

Bundle discounts apply with the Readiness Assessment (–15%) or a full engagement (–20%). All prices net of VAT.

If we claim a model is fair, we show the calculation. If we claim a gap is closed, we show the documented artifact.

— Dr. Valentin José Mayr · Founder

Related services

Before and after the audit.

S1 · Locate

Readiness Assessment

Establishes which system to audit and why — the right starting point if you’re unsure where the risk sits.

from €15,000

Learn more →

S3 · Sustain

Continuous Monitoring

Turns the one-time audit into ongoing post-market monitoring under Art. 72 — catching bias creep as it happens.

from €6,000 setup

Learn more →

S4 · Understand

Workshops

Technical Team Training to run fairness tests in-house after the audit.

from €2,000

Learn more →

Get concrete

In 30 minutes we’ll scope whether and how to audit your system.