# Canada's federal pay centre tested an AI assistant on real pay-correction cases and found no measurable productivity gain, with half of advisors faster without it

Page: https://flyvolo.ai/en/changes/ev-20241204-payroll-specialist-7
Date: 2024-12-04 · Stage: Labour impact
Can this record move a task judgement? yes
Occupation: https://flyvolo.ai/en/careers/payroll-specialist

## Scope

Canada, federal government. The pay centre's own study of 28 compensation advisors working real acting-pay cases between 24 October and 4 December 2024 says the only statistically significant conclusion is that the virtual assistant, at that stage of development, did not yield a dramatic productivity improvement; that half the participants appear to have been more productive without it; and that most advisors (60%) felt correcting the tool's errors sometimes hurt their productivity, even though participants generally felt the tool had a positive impact on their workflow. One programme, one case type, an early version of the tool.

## Source

- Public Services and Procurement Canada, Data Science and Analytics Directorate — 'AI Virtual Assistant: Productivity Study Methodology and Findings Report – Non-EX Actings' (December 2024) — https://open.canada.ca/data/dataset/3d660210-da7f-4223-938a-fdc2e691f52b/resource/ee70ae5b-f8fd-4e29-ac4b-5c7d12f826ab/download/2024-12-methodology-and-findings-report-aiva-productivity-study-on-non-ex-actings.docx (primary source)

## What this means

In real pay corrections, an AI assistant has not yet made advisors faster, and checking it costs time.

## What it does not show yet

One early version of one tool on one case type; later versions may differ.

## How to verify it yourself

Open PSPC's AI Virtual Assistant productivity study report (December 2024) and find "did not yield a dramatic productivity improvement".


More: https://flyvolo.ai/llms.txt