Decision Traces: What Multi-System Data Fusion Reveals About Hiring
A decision trace is a structured record that connects three hiring systems that almost never talk to each other. The applicant tracking system (ATS) stores what a candidate looked like on paper. The HRIS stores how they actually performed. The behavioral assessment stores a score. Linked through one candidate ID, those three systems answer a question none of them can answer alone: did the signals you screened on predict the outcomes you cared about? In a study of 10,765 hires, the answer was no.
Source: "Decision Traces: What Multi-System Data Fusion Reveals About Institutional Knowledge in Enterprise Hiring," Saad Bin Shafiq, AI Synapse Inc. (NODES), 2026. The study covers a deployment at a Fortune 500 insurance carrier, N=10,765 agents hired 2022 to 2025, run entirely inside the carrier's VPC with zero data egress. Read the paper on arXiv.
What a decision trace actually is
A decision trace links three categories of information about a single hire:
- Screening inputs (from the ATS): keywords, source channel, prior industry, license status. These are the signals traditional hiring technology screens on.
- Assessment signals (from behavioral scoring): a composite fit score and behavioral dimensions like adaptability, communication, and resilience.
- Production outcomes (from the HRIS): whether the person hit the production milestone, how many days it took, and how much premium credit they generated.
The value is not in any single field. It is in the connection between them. An ATS keyword means nothing until you can test it against who produced. A behavioral score is uninterpretable until you know what the ATS screened on and what the HRIS recorded.
The question no single system can answer
Each platform holds one piece of the story. The ATS knows who was hired. The HRIS knows how they did. Neither system knows about the other, so the reasoning behind a hiring decision lives in Slack threads and hallway conversations and then leaves when a manager retires.
A decision trace makes a query like this possible: show every agent we hired who had no insurance experience but scored above 75 on behavioral assessment, and how they performed. No single system can answer that. Only the fusion of all three can.
Three findings that were invisible inside any one system
- Resume keywords did not predict production. Of 3,597 testable keywords, zero predicted production after Bonferroni correction. Thirty were significantly anti-predictive, and the median keyword was associated with about 25% lower odds of producing. See the keyword findings.
- Personality assessment was the strongest single signal. Predictive Index type reached an AUC of 0.647 on its own and 0.735 when fused with ATS and behavioral data on the study sample. See the fusion results.
- Speed to production followed an economic constant. Each day faster to the production milestone was worth about $54 per agent, or about $35 after controlling for source channel and tenure. See the speed-to-production findings.
Why this matters for talent leaders
Most enterprises already hold all three datasets. They have simply never connected them. The cost of a decision trace is not data collection. It is data connection. Once the systems are joined, the screening rules a team has trusted for years become testable against outcomes for the first time, and the results are often the opposite of what everyone assumed.
Frequently asked questions
What is a decision trace in hiring? A decision trace is a structured record that connects screening inputs from the ATS, assessment signals from behavioral scoring, and production outcomes from the HRIS for a single hire. It lets an organization see whether what it screened on predicted what actually happened.
How is a decision trace different from a prediction? A prediction is a score about the future. A decision trace is a queryable chain of evidence about the past that makes institutional knowledge explicit, auditable, and transferable.
What data does a decision trace require? ATS records, HRIS production data, and behavioral assessment scores, linked by a common candidate identifier. Most enterprises already hold all three inside their own systems.
Where was this studied? At a Fortune 500 insurance carrier across 10,765 hires from 2022 to 2025, inside the carrier's own VPC, with no candidate data leaving its infrastructure at any point.
Related reading
- What 10,765 hires revealed about resume keywords
- Why ATS keywords fail to predict performance
- What is talent intelligence infrastructure?
See what a decision trace surfaces in your own hiring data. Book a 30-minute walkthrough.