Outcomes Inventory
A model is a method. An outcome is proof it changed something. Every figure below is a measured result Ryan Brush owned across five organizations, stated as the delta, not the activity. Where a claim is directional rather than causally clean, it says so.
The five Ryan leads with
Chat with an assistant trained on Ryan's background. Probe any number on this page: how it was measured, whether it is causally defensible, and the model behind it.
Predicting who will leave is table stakes. Proving a flagged person stayed because someone acted is the rarer claim.
Attrition-risk model built in XGBoost and surfaced to HRBPs through a Power BI drill-down, with a companion regretted / high-performer model so retention effort goes to the people whose loss actually hurts.
Critical-role attrition-risk predictions delivered to HRBPs to trigger stay interviews before the exit, concentrating effort on scientific and clinical roles that sit on the critical path.
A composite risk score across onboarding progress, manager feedback, training completion, social engagement, and ramp speed flagged at-risk new hires early and drove targeted intervention.
Overall attrition fell about 5% the quarter following model rollout across three organizations.
Stated as a measured delta, not a proven cause. The labor market moves underneath any intervention, and there was no control group.
A slow scientific hire is schedule risk, not an HR inconvenience. These numbers compress the funnel.
Funnel delay isolated to requisition approval, then the process was rebuilt, cutting time-to-post 35% and accelerating pipeline starts on critical roles.
Application flow, candidate communications, structured interviews, scheduling, and the offer process redesigned, lifting offer acceptance where the company was losing candidates to competing offers.
The strongest story here is not a number. It is a decision leadership had half-made, reversed by the analysis.
Leadership was headed toward shifting cost to employees on a medical carrier transition. A full Kaiser-to-Anthem analysis (carrier RFP, broker negotiation, cost modeling) moved them off it, avoiding over $2M in premiums while lowering employee cost and expanding access. Company and workforce both won, which is rare.
Owned the comp cycle across four public subsidiaries, authored the comp philosophy, and supported Board Compensation Committee decisions with benchmarking and scenario models, all under SOX.
Improved plan design, renegotiated vendor fees, and automated HCM-to-vendor data feeds, raising 401(k) participation across multiple organizations.
This is what justifies a seat at a company that will not grow the People team.
NLP and AI HR assistant deflected 42% of tickets across 160 sites, a 30% volume reduction versus the pre-launch baseline.
UiPath automation across the workers' comp pipeline, benefits enrollment, onboarding, report distribution, and HRIS validation eliminated 600 labor hours annually.
Rebuilt the Jira-based HR service desk, cutting response time 40% and time-to-resolution 25%, reinforcing the self-service thesis.
Cut training-content development time 80% using AI generation, the same muscle behind LLM-drafted job descriptions.
UKG Pro and WFM delivered on time: 7 modules, 1,600 employees, 160 sites, in 6 months, with carrier, Active Directory, and GL integrations.
Replaced spreadsheet reporting with a Snowflake and Power BI foundation, then a Microsoft Fabric medallion lakehouse serving six divisions, compressing leadership insight from weeks to minutes.
Zero is a very hard number to hold. These are records, not estimates.
Served as 401(k) Plan Sponsor across 5500 filings, non-discrimination testing, and fiduciary oversight, with 15 consecutive audits and no major or minor findings.
Owned 1094-C and 1095-C reporting and series-code determinations for thousands of employees across ten years with no IRS error notices.
Authored a data governance program: policy, DPIAs across 38 Tier 1 systems, master data management, and a RACI matrix, closing 8 critical audit gap areas.
Maintained SOX controls with clean audit adherence across four public subsidiaries, then productized the AI governance framework as provisional patents and a published book on defensible, AI-driven people decisions.
The results above are measured history. These are the quantified target ranges the DecisionLedger model library is engineered to produce: the outcome each model is designed to move. They are model design targets, not results claimed from any one employer.