---
id: "appraisal-metrics-provenance"
type: "appraisal"
source_timestamps: ["Enrichment Overlay"]
tags: ["evidence-quality", "provenance", "critical-reading"]
related: ["claim-vr-training-efficacy", "claim-vr-cost-at-scale", "entity-strivr", "entity-pwc", "entity-boeing"]
sources: ["reskilling"]
sourceVaultSlug: "hbr-seg-reskilling"
originDay: 10
articleStem: "hbr-edu-33-new-tools-workforce-training"
sourceUrl: "https://hbr.org/2025/12/the-new-tools-that-can-improve-workforce-training"
sourceTitle: "The New Tools That Can Improve Workforce Training"
---
# Appraisal: Which Numbers Are Peer-Reviewed vs. Marketing-Derived

## Appraisal: Which Numbers Are Peer-Reviewed vs. Marketing-Derived

The single most useful thing external research adds is a **provenance map** for the article's headline statistics. A downstream agent should cite these with the right confidence label.

| Metric | Source's number | Provenance | How to cite |
|---|---|---|---|
| VR completes training 4× faster; +275% confidence | [[claim-vr-training-efficacy]] | **PwC** enterprise report | Documented **vendor/corporate research**, not peer-reviewed |
| VR cheaper per employee at scale | [[claim-vr-cost-at-scale]] | PwC 3,000-person ROI model | **Conditional** on scale + reusability |
| [[entity-bank-of-america|BoA]] 97% / 2,000 / 200,000 | VR case | **[[entity-strivr|Strivr]] marketing case study** | Directional; specific figures uncorroborated |
| [[entity-walmart-d10|Walmart]] 15% turnover drop; 1.6M / 4,900 | VR case | **Strivr/vendor narrative** | Directional; treat 15% cautiously |
| [[entity-boeing|Boeing]] 90% first-time-quality / 30% time | AR case | **Independently corroborated** (wire-harness) | **Strongest** evidence in the piece |
| 50% / 70% / 90% forgetting | [[concept-forgetting-curve]] | Ebbinghaus paraphrase | **Stylized approximation**, not exact |
| $1.5T→$2T AI spend; "most will fail" | [[claim-ai-roi-failure]] | Gartner/analyst projection | **Forward-looking projection** |

**Rule of thumb:** the **Boeing/AR** case and the *direction* of every claim are well-supported; the crisp VR percentages are largely **vendor-produced**; the forgetting-curve percentages and AI-spend totals are **stylized/extrapolative**. Present the phenomena as real and the exact numbers as illustrative.