Architecture

How Kettrion turns structured observation into a hiring asset

14 Jun 2026

Standard hiring software treats each search as a transaction. A role opens, candidates flow through and a decision is made. The data left behind is a log of activity — who applied, when, what stage. The substance of the decision — what the interviewers actually observed, what evidence drove the call — sits in unstructured notes, scattered emails, and the hiring manager’s memory.

Kettrion is built on a different assumption. The substance of the decision is the asset.

Observation-layer data, not activity logs

Most executive hiring systems record what happened. Kettrion records what was observed.

Every assessment — interview, reference call, work-sample exercise — is broken into discrete observations. Each observation is timestamped. Each observation is attributed to a source: which interviewer, which question, which behavioral anchor. Each observation is tagged to the competency it speaks to and the evidence level it represents.

The unit of data is not “Candidate X scored 4 out of 5 on leadership.” The unit of data is “Interviewer A observed behavior B in response to question C, on date D, scored against rubric E.”

This is what the literature calls observation-level data. It is the granularity industrial and organizational psychology has called for since the 1990s. It is also what makes everything downstream possible.

A corpus that compounds with use

A flat CRM is full at the end of every search. Records are filed. The next search starts from zero.

A corpus does not.

Every observation captured in Kettrion adds to a structured body of evidence that compounds across mandates. Patterns become visible that no single search reveals. Which interview questions actually predict outcomes 18 months out. Which behavioral anchors are scored consistently between interviewers and which produce noise. Where the corpus is dense enough to draw inferences, and where it is too thin and any conclusion is founder judgement.

The corpus is the user’s asset.

How modern technology fits in

The technologies that make this practical did not exist 10 years ago.

Structured capture from unstructured input. Modern language models can take an interviewer’s natural prose — typed or spoken — and decompose it into structured observations against a predefined rubric. The interviewer writes the way they always have. The system extracts the discrete observations and asks the interviewer to confirm each one. The human stays in charge.

This matters because the friction of structured note-taking is the reason most organizations don’t do it. We are removing the friction so that the methodology becomes usable at scale.

Semantic retrieval across the corpus. Embedding models make it possible to ask the corpus questions that would have been impractical with a relational database. “Show me every observation on adaptability under pressure across the last five CFO searches.” “Surface every interviewer disagreement on a single behavioral anchor in the last year.” The retrieval is fast because the data is structured and the meaning is preserved.

Pattern detection at the rubric level. When every observation is anchored to a rubric, statistical work becomes possible. Which questions produce high inter-rater agreement and which do not. Which competencies show drift in scoring over time and need recalibration. These are not features. They are the by-products of a properly structured corpus.

Explainability built in, not bolted on. Under Article 86 of the EU AI Act, any high-risk hiring system must be able to explain its outputs to the affected person. The observation-layer architecture makes this structural rather than aspirational. Every recommendation surfaces the underlying observations, the source attribution, and the evidence chain. There is no black-box stage because the evidence record is human-anchored throughout. (EU AI Act, Reg. (EU) 2024/1689, Art. 86 [EN/DE/FR/NL])

What AI collaboration means here

Kettrion is AI-forward. The wording matters here. AI does not make the hiring decision, or score the participant.

AI does three things in Kettrion. It removes the friction between how interviewers actually write and how the methodology requires the data to be structured. It surfaces patterns in the corpus that a human analyst would need a lot of time to find. It supports the explainability layer required for compliance under the EU AI Act.

The methodology was validated on 12 years of executive search mandates before any model was involved. The technology serves the methodology, not the other way round.

What the CHRO actually receives

A recommendation, with the full evidence chain visible beneath it. Every observation, every source, every score, every disagreement between interviewers and how it was resolved. Audit-ready under the EU AI Act. Defensible in a board conversation. Robust enough to revisit 18 months later, when the hire either succeeds or doesn’t.

The hiring decision belongs to the hiring manager. Kettrion supplies the evidence that makes the decision defensible.

Why this changes the outcome

The Center for Creative Leadership’s 40-year derailment study found that failed executives are largely indistinguishable from successful ones at the point of selection — when selection relies on the methods most organizations use. The signals that predict failure are detectable. They are just not detectable in CV review and unstructured conversation.

Structured observation, properly captured and stored as a corpus, is how those signals become detectable at scale. The science is not new. The architecture is.

Confidence delivered.

Sources

← All insights