
Job evaluation is slow and manual due to inconsistent agency submissions, subjective assessments, and labour-intensive precedent checks by COMPOD.
PSD COMPOD team and Ministry HR Administrators responsible for maintaining organisation structures across the Singapore Public Service.
Ministries submit job evaluation requests to PSD COMPOD via email whenever they need to create a new position or amend the job scope of an existing one. COMPOD evaluates each Job Description (JD) using the Korn Ferry Hay Methodology (KFHM) to propose an appropriate score, profile, and grade, before identifying two comparator jobs from their database to ensure alignment. PSD COMPOD's approval is required before Ministries can proceed with subsequent processes such as recruitment or placing officers into positions. These requests come in during position creation, agencies' 5-yearly review exercises, or staff structure renewals.
Two key pain points slow the process down. First, JDs submitted by agencies are often lengthy and inconsistently structured, making it difficult for COMPOD officers to quickly understand the true nature of a position — resulting in more than 7 man-days spent seeking clarification per evaluation. Second, comparator job data is stored in an unstructured manner across different locations and formats in SharePoint, making it time-consuming to locate relevant precedents for benchmarking.
The current process takes approximately 2 weeks per evaluation, creating a significant bottleneck for downstream HR processes such as recruitment and officer placement. The manual, judgement-heavy nature of the work also introduces inconsistency across evaluations, as assessments can vary between officers. If left unsolved, this will continue to strain COMPOD's capacity and delay agencies' HR operations.
This is an operational process that affects all agencies in WOG.
The team adopted a human-centred research approach, combining qualitative and quantitative methods to validate the solution. This included direct stakeholder engagement with PSD COMPOD officers to understand their workflows, pain points, and bottlenecks within the job evaluation process.
In parallel, a process walkthrough (contextual inquiry) was conducted to map the end-to-end evaluation journey, which revealed two key friction points. (1) the effort required to comprehend lengthy job descriptions and (2) the challenge of locating relevant comparator roles.
To validate the solution, the team conducted prototype testing using 18 real job descriptions, benchmarking AI-generated outputs against historical COMPOD evaluations to assess accuracy and practical usability.
We identified two key product risks grounded in core assumptions and addressed them through targeted validation and design strategies.
Mitigation: We grounded the solution in the Korn Ferry Hay methodology and benchmarked outputs against historical COMPOD evaluations. Through iterative testing on real job descriptions, the model achieved accuracy within acceptable HRL scoring ranges for ~70% of cases, demonstrating practical viability.
Mitigation: We designed the product as a decision-support tool rather than a replacement, ensuring humans remain accountable for final approvals. In addition, we embedded the solution within COMPOD’s existing data ecosystem (e.g. comparator job database for precedent evaluations) to increase transparency and user trust.
The tool was developed in close collaboration with PSD's COMPOD team, who are the primary users and have been involved in scoping, testing, and refining the product. Eventually, this may be rolled out to all agencies in WOG which are required to submit to PSD COMPOD for revaluations.
Job Evaluation AI Tool directly addresses both pain points. It automatically summarises lengthy JDs into structured, digestible formats. It then applies the Korn Ferry Hay Methodology to propose KH, PS, and ACC scores and a job grade, while surfacing the two most relevant comparator jobs from COMPOD's historical database within a fraction of the time previously required.
The key user flow is as follows. The COMPOD officer logs into the Job Evaluation AI Tool and uploads the JD document (in Word format, up to 10 files at a time). The tool summarizes the JD and extracts key information such as job title, key skills, seniority, and job family. It then retrieves relevant methodology guidelines and searches for semantically similar historical evaluations to use as reference. Using these inputs, the tool applies the Korn Ferry Hay Methodology to propose scores and a job grade, and surfaces two comparator jobs for benchmarking. The officer reviews the proposed evaluation and comparators before proceeding with the approval process. Results can be exported in Word or PDF format.
The tool is designed to fit into COMPOD's existing workflow with minimal friction. Officers can upload JD documents directly in their existing file formats without reformatting. The interface presents evaluation results and comparator jobs in a clear, structured table. Multiple files can be evaluated in parallel, and results can be exported easily for downstream use.
(To be updated after user testing with COMPOD officers. Initial testing on 18 JDs showed that the tool's proposed grades fell within the acceptable HRL score range for 70% of cases. The KB will be uploaded with bigger datasets of historical evaluations to validate and increase accuracies)
The primary success metrics are accuracy of job evaluation results (targeting at least 70% of evaluations falling within the acceptable HRL score range compared to past COMPOD manual evaluations) and reduction in evaluation turnaround time (targeting a 25% reduction from the current 2–4 week process).
In the short term, the tool aims to reduce the time COMPOD officers spend on each job evaluation by at least 25%, freeing up capacity for higher-value work and reducing the bottleneck for agencies' HR processes. In the medium term, as the tool's knowledge base grows with more historical evaluations, accuracy and consistency of proposed grades are expected to improve further. In the long term, a more scalable and consistent job evaluation process supports the broader goal of maintaining a well-structured and fairly graded Singapore Public Service.