Research

Field research and applied science on AI in HR

We pair our own field research with HR and technology leaders across Europe with an academic research programme, and publish what we learn as it stands, open questions included.

Peer-reviewed research

We publish our research openly. Each paper below is archived on Zenodo with a permanent DOI and released under CC BY 4.0.

Egress-Bounded Retrieval: A Privacy-Preserving Enterprise Memory Architecture with No External Model in the Retrieval Path

Alessio Biancheri · Preprint · 12 July 2026 · Zenodo

CognitiveCore is an enterprise memory layer built around a single architectural invariant, egress-bounded retrieval: no corpus content leaves the customer's trust boundary during retrieval. Organising, indexing, searching and ranking are performed exclusively by small, local, in-process models. An external large language model is invoked only at a final synthesis step, and sees only the passages already selected for the question.

DOI: 10.5281/zenodo.21322415 Read the PDF on Zenodo

When the Model Writes the Questionnaire: An Item-Level Defect Taxonomy for LLM-Generated 360-Degree Feedback Instruments

Alessio Biancheri · Preprint · 12 July 2026 · Zenodo

A field audit of LLM-generated 360-degree feedback items across two production pilots. 37 of 48 generated items (77%) carried at least one item-writing defect, organised into eight classes. A real-time assistant intended to improve feedback quality systematically steered assessors to rewrite legitimate negative feedback in positive terms, producing data that is biased in a direction its own outputs cannot reveal.

DOI: 10.5281/zenodo.21323410 Read the PDF on Zenodo

Research programme

Preserving human expertise in AI-augmented performance management

With the University of Milan, we use Pulse360 as the empirical context to study whether AI can improve efficiency without weakening human judgment, accountability or learning. The design mechanisms under study include Coach Mode, which asks questions instead of writing the assessor's answer; confidence indicators that communicate evidence strength; source traceability that supports verification; development outputs reviewed before delivery; a human who decides; and systems designed for failure and override.

These are research questions and design principles under evaluation, not final academic findings.

Field reports

The State of 360 Feedback 2026

What 80+ HR and technology leaders across six European markets told us about why 360 feedback so often disappoints, and what a version that works would need.

Read the field report

The State of People Pairing 2026

Mentoring, buddy and cross-functional programs across six markets: where people pairing breaks down on matching, structure and measurement.

Read the field report