From Job Postings to Curriculum Decisions: Using AI to Generate Workforce Intelligence for MSW Program Planning

This paper presents a case study demonstrating how a Master of Social Work program utilized a locally deployed AI model to analyze over 40,000 job postings, generating actionable workforce intelligence on practice specializations and required competencies to inform strategic curriculum planning.

The Gist

Imagine you are the head chef of a very popular cooking school. Your goal is to teach students how to cook so they can get hired in top restaurants.

For years, you've asked your former students, "What did your bosses want?" and you've asked your advisory board, "What should we teach?" But these answers are like hearing about the menu from people who ate the meal last year. You don't know what the restaurants are ordering right now.

This paper is about a group of social work professors who decided to stop guessing and start looking at the actual "order tickets" from the job market. They used a smart computer assistant (Artificial Intelligence) to read over 41,000 job postings to see exactly what employers are asking for today.

Here is the story of what they found, explained simply:

1. The Tool: A Super-Reader Robot

Instead of hiring a team of humans to read 41,000 job ads (which would take forever), they used a "local" AI robot. Think of this robot as a super-fast librarian who can read a whole library in a minute.

• The Mission: The robot had to sort these ads into two buckets: "Is this a job for a Social Worker?" and "What specific skills does this job need?"
• The Specializations: The school teaches eight different "flavors" of social work (like Clinical Therapy, Child Welfare, Policy, and Management). The robot sorted every job into these flavors.

2. The Big Picture: What's Hot?

After the robot finished its work, a clear picture emerged:

• The "Therapist" is King: The biggest chunk of jobs (nearly 70%) was for Interpersonal Practice (basically, therapists and counselors). It's the most in-demand skill right now.
• Kids and Families are a Close Second: Jobs helping children and families were the next biggest group.
• The "Big Picture" Jobs: Jobs in management, policy, and community change were smaller in number, but they are still very important.

3. The "Mix-and-Match" Surprise

Here is where it gets interesting. The professors wondered: Do employers want specialists (just one skill) or generalists (a little bit of everything)?

• The "Swiss Army Knife" Effect: For jobs that might accept a social worker but also accept other degrees (like a counselor or a public health worker), employers wanted a "Swiss Army Knife." They wanted people who could do management, policy, and community work all at once.
• The "Specialist" Effect: However, for jobs that strictly required a Social Work degree, employers wanted a specialist. They didn't want a generalist; they wanted someone deep in one specific area.
  • Analogy: If you are hiring a "Human Services" person, you want a Swiss Army Knife. If you are hiring a "Licensed Social Worker," you want a master chef who specializes in one specific cuisine.

4. The Hidden Skills: What's Actually on the Menu?

The robot looked at the specific skills listed in the ads.

• The Universal Tools: Two skills showed up everywhere, no matter what the job was: Clinical Assessment (figuring out what's wrong) and Case Management (organizing help). It's like knowing how to chop onions; every chef needs to know how to do it, whether they are making soup or steak.
• The "Trauma" Trend: A fascinating discovery was Trauma-Informed Care. Usually, this is taught to therapists. But the robot found it appearing in manager and policy jobs too!
  • What this means: Employers don't just want managers to know how to run a budget; they want them to understand how trauma affects the whole organization. It's like a restaurant manager needing to know not just how to count money, but how to create a safe, welcoming atmosphere for everyone.
• The "Silent" Skills: Surprisingly, specific therapy techniques (like CBT) weren't mentioned in every ad. The professors realized this doesn't mean they aren't important. It's like a job posting for a driver saying "Must have a license" but not listing "Must know how to turn left." Employers assume you already know the basics; they only list the special stuff they need.

5. The Tech Factor

The robot also found that technology is huge.

• Clinical jobs need to know how to use Electronic Health Records (digital patient files).
• Research jobs need to know complex data software (like SPSS or R).
• Everyone needs to know basic tools like Excel.

The Bottom Line: Why This Matters

The professors aren't saying, "We must change our entire school based on this." Instead, they are saying, "This is a new, powerful flashlight we can use to see the market clearly."

Before, they were driving at night with a dim flashlight (surveys and old memories). Now, they have a high-powered spotlight (AI analysis of real-time job ads).

The Takeaway:
This paper shows that schools can use AI to listen to the "voice of the market." It helps them decide:

1. Do we need more training in assessment for everyone? (Yes, the data says so).
2. Should we teach trauma care to our future managers, not just therapists? (Yes, the data says so).
3. Do we need to keep our "macro" classes separate, or should we mix them? (The data suggests mixing them might be better for some jobs).

It's not about letting a robot decide the curriculum; it's about giving the human teachers a better map so they can guide their students to success.

Technical Summary

Here is a detailed technical summary of the paper "From Job Postings to Curriculum Decisions: Using AI to Generate Workforce Intelligence for MSW Program Planning."

1. Problem Statement

Social work Master of Social Work (MSW) programs face a persistent challenge in aligning their curricula with current employer expectations. Traditionally, curriculum decisions rely on:

• Advisory board input.
• Alumni surveys (retrospective self-reports).
• Anecdotal faculty/employer feedback.

These sources suffer from limitations: they are often small samples, retrospective, and lack direct analysis of explicit employer requirements. While the Council on Social Work Education (CSWE) and the Association of Social Work Boards (ASWB) provide competency frameworks (e.g., 2022 EPAS, 2024 Practice Analysis), these aggregate data capture what practitioners report doing over careers, not what employers explicitly require at the point of hire. There is a lack of systematic, scalable methods to analyze unstructured job posting data to generate "organizational intelligence" for specific institutional planning.

2. Methodology

The authors employed a three-stage pipeline using a locally deployed Large Language Model (LLM) to process a massive dataset of job postings.

A. Data Collection

• Source: 41,584 unique job postings collected in December 2025 from three major platforms: Indeed, LinkedIn, and Glassdoor.
• Scope: National (all 50 U.S. states and territories) with no geographic filtering.
• Search Strategy: 79 distinct search terms covering clinical practice, child welfare, program evaluation, and administrative roles.
• Tool: Octoparse (no-code web scraper).

B. AI Model & Infrastructure

• Model: GPT-OSS:20B (a 20-billion-parameter open-source language model).
• Deployment: Locally hosted on custom hardware using the llama.cpp inference engine.
• Rationale: Selected for computational efficiency, reduced environmental impact, and privacy preservation compared to commercial models (e.g., ChatGPT, Claude). It allows for "zero-shot" classification without requiring large manually labeled training datasets.

C. Classification & Extraction Pipeline

1. Relevance Screening: Each posting was classified into three categories:
   • Strong Alignment: Explicitly requires/prefer MSW or LCSW/LMSW; core responsibilities match social work domains.
   • Partial Alignment: Human services roles where MSW is applicable but not central (e.g., care coordination).
   • No Alignment: Distinct professions or non-human services.
   • Result: 23,732 postings (57.1%) retained for analysis.
2. Specialization Classification: Retained postings were mapped to the institution's eight specific practice specializations (e.g., Interpersonal Practice, Children/Youth/Families, Management/Leadership, Policy/Political, etc.) using multi-label classification.
3. Skills Extraction: The model extracted and normalized four categories of skills:
   • Therapeutic modalities (e.g., CBT, Trauma-Informed Care).
   • Technical skills (e.g., Clinical Assessment, Grant Writing).
   • Soft skills (e.g., Communication).
   • Technology skills (e.g., EHR, SPSS, Excel).
4. Quality Assurance: A subject matter expert reviewed a sample of model outputs. Disagreements were rare and mostly limited to edge cases. The model was found to be conservative (under-counting rather than misclassifying), which was deemed acceptable for aggregate pattern analysis.

D. Statistical Analysis

• Descriptive Statistics: Market share percentages calculated for specialization demand.
• Co-occurrence Analysis: Binary alignment matrices were constructed to calculate Phi coefficients ($\phi$), measuring the association between specializations (i.e., how often two specializations appear in the same job posting).
• Differentiation: Analysis was conducted separately for "All Aligned" positions and "Strongly Aligned" positions to distinguish between general human services roles and those specifically requiring MSW credentials.

3. Key Results

A. Specialization Demand Distribution

• Interpersonal Practice (Clinical): Dominated the market with 69.9% market share (16,597 positions). Crucially, 4,811 of these were "strongly aligned," explicitly requiring clinical credentials.
• Children, Youth, and Families: Second largest (42.8%), with a high rate of strong alignment (4,263 positions), indicating specific demand for MSW credentials in child welfare.
• Macro Specializations: Management/Leadership (22.6%), Older Adults (22.4%), Community Change (11.3%), Policy/Political (3.6%), and Global Social Work (3.3%).

B. Specialization Co-occurrence Patterns

• Macro-Level Clustering: Among all aligned positions, macro specializations (Management, Evaluation, Community Change, Policy) showed moderate-to-strong positive correlations ($\phi$ = 0.25–0.44), suggesting employers often seek integrated generalist macro skills.
• Divergence in Strong Alignment: Among strongly aligned (MSW-required) positions, these macro correlations dropped to near zero. This suggests that while generalist macro roles exist, employers requiring specific MSW credentials seek specialized depth rather than generalist bundling.
• Population Segregation: A persistent negative correlation existed between "Children, Youth, and Families" and "Older Adults" ($\phi$ = -0.27), indicating distinct, non-transferable skill sets for these populations.

C. Skills & Competencies

• Cross-Cutting Technical Skills: Clinical Assessment and Case Management were the most prevalent skills across almost all specializations, validating their status as foundational competencies.
• Therapeutic Modalities: Explicit mentions were rare. However, when specified, they were often "required" rather than "preferred."
  • Trauma-Informed Care (TIC): Appeared not just in clinical roles but significantly in Management (11.1%) and Program Evaluation (5.7%), indicating a shift from a clinical modality to an organizational framework.
  • CBT: Surprisingly low frequency (2.6% of postings), likely because employers assume it as a baseline competency for clinical roles.
• Technology Skills:
  • Clinical: EHR, HIPAA compliance, Clinical Documentation.
  • Macro/Research: High demand for statistical software (SPSS, SAS, R) and data visualization (Power BI, Tableau) in Program Evaluation roles.

4. Key Contributions

1. Methodological Innovation: Demonstrates a transferable, low-cost pipeline for generating workforce intelligence using locally deployed, smaller LLMs (20B parameters). This democratizes access to big-data labor market analysis for academic programs without needing commercial contracts or data science teams.
2. Organizational Intelligence vs. Scientific Generalization: The paper explicitly frames its output as "organizational intelligence" (actionable data for a specific institution) rather than generalizable scientific truth. It provides a model for how programs can use data to inform internal deliberation.
3. Granular Skill Extraction: Moves beyond broad job titles to extract specific, normalized skills (modalities, software, technical competencies) from unstructured text at a scale (23k+ postings) impossible with manual coding.
4. Insight into "Strong" vs. "Partial" Alignment: The distinction between jobs that accept an MSW among other credentials vs. those that require it revealed distinct hiring patterns, particularly the specialization of macro roles when MSW credentials are mandatory.

5. Significance and Implications

• Curriculum Planning: The data suggests a need to strengthen Clinical Assessment and Case Management as foundational courses for all students, not just clinical tracks.
• Specialization Structure: The findings challenge rigid specialization tracks. While macro roles show integration in the general market, MSW-specific roles demand specialization. Programs might consider hybrid tracks (e.g., Management + Evaluation) while maintaining distinct population-focused training (e.g., Child Welfare vs. Gerontology).
• Trauma-Informed Care (TIC): The expansion of TIC into management and policy roles suggests curricula should integrate TIC principles across the entire program, not just in clinical courses.
• Technology Integration: The high demand for EHR in clinical roles and statistical software in research roles indicates a need for explicit technology training integrated into field preparation and coursework.
• Strategic Decision Making: The paper argues that AI-generated workforce data should serve as one input among many (alongside advisory boards and alumni data) to inform curriculum deliberation, helping to surface systematic signals that might otherwise remain invisible.

In conclusion, the paper successfully bridges the gap between unstructured labor market data and structured curriculum planning, offering a scalable, AI-driven framework for social work education to remain responsive to evolving workforce demands.