CAN-001

Employment Insurance (EI) Machine Learning Workload

Download PDF
Canada North America High income Operational Deployment (Limited Rollout) Confirmed

Employment and Social Development Canada (ESDC)

At a Glance

What it does Prediction (including forecasting) — Operational and process automation
Who runs it Employment and Social Development Canada (ESDC)
Programme Employment Insurance (EI) Machine Learning Workload
Confidence Confirmed
Deployment Status Operational Deployment (Limited Rollout)
Key Risks Model-related risks
Key Outcomes Over 40,000 EI claims processed via ML triage as of July 2023; model accurately identifies no-change claims at 90% rate; improved case prioritisation and staff allocation efficiency; project ongoing under renamed Employment Insurance Machine Learning Workload.
Source Quality 6 sources — Government website / press release, Report (government / official), Dataset / database, +2 more

Employment and Social Development Canada (ESDC) developed and deployed a machine learning system to forecast and triage Employment Insurance (EI) recalculation workloads within its Benefits and Integrated Services Branch (BISB). The system, known as the Machine Learning Workload project, uses a supervised Random Forest classification model trained on historical EI production data to predict the likely outcome of benefit recalculations and prioritise associated work items for processing by officers.

The recalculation of Employment Insurance benefits involves conducting a comprehensive review and reassessment of a claim in response to new information or changes in circumstances. The primary objective of this review is to ensure accurate benefit calculations, determining whether they result in overpayment, underpayment, or no change to the claimant's benefit rate. Since the introduction of Web and Electronic Records of Employment (ROEs) in 2013, approximately 300,000 recalculation work items have been generated annually. During the COVID-19 pandemic, this number surged to over 1.7 million due to the implementation of simplification measures associated with the Emergency Response Benefit (ERB), simplified EI, and the subsequent return to regular EI processing. This created a substantial backlog of claim reviews, many of them for claims established before March 2020, which competed for resources with more current and pressing work.

The ML model was developed by the Employment Insurance Program Performance unit within ESDC in consultation with several internal stakeholders across the EI programme. A Random Forest algorithm was applied to data drawn from EI production systems. The model specifically focuses on terminated or dormant claims and classifies claim recalculations into three categories of predicted outcomes: increase in benefit rate (indicating underpayment), decrease in benefit rate (indicating overpayment), and no change in benefit rate. This classification enables the programme to close recalculations that are predicted to have no impact on the claimant and to prioritise those with a greater likelihood of resulting in a change of benefit, thereby directing officer attention to cases that matter most.

The Algorithmic Impact Assessment (AIA) completed for this system under Canada's Directive on Automated Decision-Making yielded an Impact Level of 1, the lowest level on the scale, with a current score of 30, a raw impact score of 30, and a mitigation score of 29. The AIA classified the system as partial automation, meaning it contributes to administrative decision-making by supporting an officer through assessments, recommendations, intermediate decisions, or other outputs, rather than making fully autonomous decisions. The assessment noted that the model's decisions are reversible and that impacts are most likely to be brief. The system was assessed as having little to no impact on the equality, dignity, privacy, autonomy, health, and economic interests of individuals.

The system uses personal information from EI production systems as input data, with the highest security classification being Protected A. Data is controlled by the federal government, collected by ESDC itself, and the system interfaces with other IT systems while using data from multiple different sources. The system operates within a closed environment with no connections to the Internet, Intranet, or any other external system. De-identification is achieved through the creation of a primary key. The system does not require the analysis of unstructured data.

Before deployment, strict adherence to the Treasury Board of Canada Secretariat (TBS) guidelines for Automated Decision Systems was required. The TBS has been regulating the use of ML within the Government of Canada since April 1, 2020, and compliance with the TBS Directive for ADS requires a rigorous approval process to ensure adherence to established standards. Risk mitigation measures include randomised manual spot checks conducted by agents to review claims with no anticipated change in benefits. If a client makes an inquiry about their file, officers proceed with the full recalculation process. The Integrity Service Branch (ISB) is responsible for reviewing undeclared contentious issues to ensure accuracy and fairness. Additionally, the workload process is modified to prioritise active claims, allowing for efficient allocation of resources.

Internal stakeholder consultations were conducted with Data Governance, Programme Policy, Legal Services, Communications Services, and the Access to Information and Privacy Office. Accountability for the design, development, maintenance, and improvement of the system has been formally assigned within ESDC. The system provides an audit trail that records all recommendations and decisions, identifies all key decision points, maintains a current log of changes to the model and system, identifies the system version used for each decision, and enables human override of system decisions with logging of override instances. A recourse process is established for clients who wish to challenge decisions. A concept case was prepared and presented to the Government of Canada Enterprise Architecture Review Board.

The model was reported to accurately identify claims with no change in benefits at a rate of 90 percent. The worst-case scenario identified in the AIA occurs when a claim eligible for a higher benefit rate is erroneously classified as having no change. By July 2023, the implementation of the Machine Learning model proved successful, resulting in the completion of over 40,000 claims. Although the original backlog reduction project page is no longer updated, the project continues under the name Employment Insurance Machine Learning Workload. The system is part of ESDC's broader 2023-2026 Data Strategy, which emphasises data-driven approaches to improving programme delivery and operational efficiency.

The Integrity Services Branch at ESDC has also developed broader analytics capabilities including predictive models, anomaly detection algorithms, and the Integrity Technology-Reinforced Anomaly Program (iTRAP), which has identified over 70,000 instances of potential identity theft and prevented approximately one billion dollars in benefits from being paid fraudulently since its deployment in late 2021. The ML Workload project represents one component of this wider data-driven transformation of integrity and operational functions within Canada's Employment Insurance programme.

Classifications follow the DCI AI Hub Taxonomy. Hover over field labels for definitions.

Social Protection Functions

Implementation/delivery chain
Case management primaryAccountability mechanisms
SP Pillar (Primary) The social protection branch: social assistance, social insurance, or labour market programmes. Social insurance
Programme Name Employment Insurance (EI) Machine Learning Workload
Programme Type The type of social protection programme, classified under social assistance, social insurance, or labour market programmes. View in glossary Unemployment Insurance
System Level Where in the social protection system the AI is applied: policy level, programme design, or implementation/delivery chain. View in glossary Implementation/delivery chain
Automation Subtype For operational automation cases: (a) document processing and generative staff assistance, or (b) workload and resource forecasting. (b) Workload and resource forecasting
Programme Description Machine learning system using a Random Forest model to predict the outcome of Employment Insurance recalculation work items, triaging them to prioritise cases likely to result in a change of benefit and closing those with no anticipated impact on claimants.
Implementation Type How the AI output is produced: Classical ML, Deep learning, Foundation model, or Hybrid. Affects validation, compute requirements, and governance profile. View in glossary Classical ML
Lifecycle Stage Current stage in the AI lifecycle, from problem identification through to monitoring, maintenance and decommissioning. View in glossary Monitoring, Maintenance and Decommissioning
Model Provenance Origin of the AI model: developed in-house, adapted from open-source, commercial/proprietary, or accessed via third-party API. View in glossary Developed in-house
Compute Environment Where the AI system runs: on-premise, government cloud, commercial cloud, or edge/device. View in glossary Not documented
Sovereignty Quadrant Classification of data and compute sovereignty: I (Sovereign), II (Federated/Hybrid), III (Cloud with safeguards), or IV (Shared Innovation Zone). View in glossary Not assessed
Data Residency Where the data used by the AI system is stored: domestic, regional, or international. View in glossary Not documented
Cross-Border Transfer Whether data crosses national borders, and if so, whether documented safeguards are in place. View in glossary Not documented
Decision Criticality The rights impact of the decision the AI supports. High criticality requires HITL oversight; moderate requires HOTL; low may operate HOOTL. View in glossary Moderate
Human Oversight Type Level of human involvement: Human-in-the-Loop (active review), Human-on-the-Loop (monitoring), or Human-out-of-the-Loop (periodic audit). View in glossary HOTL
Development Process Whether the AI system was developed fully in-house, through a mix of in-house and third-party, or fully by an external provider. View in glossary Fully in-house
Highest Risk Category The most significant structural risk source identified: data, model, operational, governance, or market/sovereignty risks. View in glossary Model-related risks
Risk Assessment Status Whether a formal risk assessment, informal assessment, or independent audit has been conducted for this system. Formal assessment

Risk Dimensions

Governance and institutional oversight risks
Operational and system integration risks

Impact Dimensions

Autonomy, human dignity and due process
Equality, non-discrimination, fairness and inclusion
  • DPIA/AIA conducted
  • Data minimisation controls
  • Grievance mechanism
  • Human oversight protocol
CategorySensitivityCross-System LinkageAvailabilityKey Constraints
Administrative data from other sectorsPersonalLinks data across multiple systemsCurrently available and usedElectronic Records of Employment (ROEs) from employers; system interfaces with other IT systems as confirmed in AIA
Beneficiary registries and MISPersonalLinks data across multiple systemsCurrently available and usedEI production system data including claim and recalculation records; Protected A classification; data collected and controlled by ESDC (federal government)

Canada School of Public Service (2023) Government of Canada Data Conference 2023: ESDC Integrity Services Branch's Use of Quantitative and Qualitative Data (Transcript). Available at: https://www.csps-efpc.gc.ca/video/data-conference2023/integrity-services-eng.aspx (Accessed: 31 October 2025).

View source Government website / press release

Employment and Social Development Canada (2024) Algorithmic Impact Assessment - Machine Learning Workload (Employment Insurance). Open Government Portal. Available at: https://open.canada.ca/data/dataset/6b429c8e-ee5b-451a-883f-b6180ada9286/resource/2694367e-babf-4d87-a852-827e4178141d/download/1-aia_ei_ml_workload_en.pdf (Accessed: 31 October 2025).

View source Report (government / official)

Employment and Social Development Canada (2024) Reducing Employment Insurance Backlog: A Machine Learning Approach. Open Government Portal. Available at: https://open.canada.ca/data/en/info/24d2cab2-6a0d-4234-9239-b6ce102ebabd (Accessed: 31 October 2025).

View source Dataset / database

Government of Canada (2025) Amendments to the Directive on Automated Decision-Making. Available at: https://www.canada.ca/en/government/system/digital-government/policies-standards/policy-service-digital-announcements/amendments-directive-automated-decision-making.html (Accessed: 31 October 2025).

View source Legal document / regulation

Kaye, K. (2024) 'AI Governance on the Ground: Canada's Algorithmic Impact Assessment Process and Algorithm has evolved', World Privacy Forum, 14 August. Available at: https://www.worldprivacyforum.org/2024/08/ai-governance-on-the-ground-series-canada/ (Accessed: 30 March 2026).

View source Report (multilateral / development partner)

Treasury Board of Canada Secretariat (2024) Guide on the Scope of the Directive on Automated Decision-Making. Ottawa: Government of Canada. Available at: https://publications.gc.ca/collections/collection_2024/sct-tbs/BT48-46-2024-eng.pdf (Accessed: 31 October 2025).

View source Report (government / official)
Deployment Status How far the system has progressed into real-world operational use, from concept/exploration through to scaled and institutionalised. View in glossary Operational Deployment (Limited Rollout)
Year Initiated The year the AI system was first initiated or development began. 2023
Scale / Coverage The scale and geographic or population coverage of the deployment. Over 40,000 EI claims processed via ML triage as of July 2023; ongoing under Employment Insurance Machine Learning Workload programme
Funding Source The source(s) of funding for the AI system development and deployment. Government of Canada (federal budget)
Technical Partners External technology vendors, academic partners, or development partners involved. Developed in-house by ESDC Employment Insurance Program Performance unit; no external vendor disclosed
Outcomes / Results Over 40,000 EI claims processed via ML triage as of July 2023; model accurately identifies no-change claims at 90% rate; improved case prioritisation and staff allocation efficiency; project ongoing under renamed Employment Insurance Machine Learning Workload
Challenges Worst-case scenario: claim eligible for higher benefit rate erroneously classified as no change; no Gender Based Analysis Plus (GBA+) conducted on the data; data quality processes not publicly available; no external stakeholder consultation conducted

How to Cite

DCI AI Hub (2026). 'Employment Insurance (EI) Machine Learning Workload', AI Hub AI Tracker, case CAN-001. Digital Convergence Initiative. Available at: https://socialprotectionai.org/use-case/CAN-001 [Accessed: 1 April 2026].

Change History

Created 30 Mar 2026, 08:38
by v2-import (import)