GovTech SaaS 2024 – 2025 AI/ML · Product Lead

Classification &
Comp Study Intelligence

How I pitched and led product development for an AI-powered classification and compensation study platform — moving government agencies from 6–12 month consultant engagements to a self-service, always-on intelligence engine.

$2.4B

US public sector TAM

<4 wks

target study time vs. 4–6 months avg

85%

target AI suggestion acceptance rate

PM Lead

pitched, scoped, led eng team

Classification & comp studies are broken at every level

Government agencies run classification and compensation studies to answer a deceptively hard question: are our jobs classified correctly, and are we paying people fairly relative to the market?

The current answer takes 6–12 months to arrive. An external consultant reads hundreds of job descriptions, manually matches them to market survey positions, builds a compensation model in Excel, and delivers a static PDF. By the time the report is approved, the market data is outdated.

The deeper problem: there's no unified taxonomy. Every agency calls their jobs something slightly different. "Senior Budget Analyst" in one city doesn't mean the same thing as the same title in another. Before any market comparison can happen, someone has to standardize the job taxonomy — and that work is manual, inconsistent, and never auditable.

The status quo

  • Studies take 6–12 months; market data outdated by delivery
  • Results delivered as static PDFs — no live iteration or what-if modeling
  • Job descriptions, benchmarks, and cost models siloed across Word/Excel/ERP
  • Opaque comparator logic — unions distrust methodology, agencies can't defend it
  • 85% of agencies report significant friction translating study results into budget line items

The opportunity

  • Build a unified taxonomy and let AI do the job matching
  • Replace the static PDF with live cost modeling and scenario analysis
  • Make every comparator decision auditable and transparent
  • Shrink study cycles from months to weeks — repeatable every fiscal year

Why this problem is worth solving now

62%

budget forecast accuracy for agencies using traditional study methods

Internal customer data

3,500+

target US public sector agencies in the addressable market

SAM: $850M

50–70%

projected cost savings vs. external consultants for agencies on the platform

Business case model

Strategic timing

Rising pay equity and transparency legislation is forcing agencies to adopt defensible compensation structures. Tightening fiscal budgets require precise cost modeling, not six-figure consultant engagements. The window to own this market is now.

The platform advantage

The platform was already serving 83+ agencies for benchmarking. Connecting classification intelligence to benchmarking and cost modeling on a single data model was a moat no legacy consultant could replicate.

Who we were building for

Maria

HR Director · Mid-Size City

Owns classification and compensation studies end-to-end. Currently hires external consultants at $40–80K per engagement. Accountable to the city manager for defensible results. Frustrated that the process takes nearly a year and the output is a PDF she can't iterate on.

Pain points

→ Consultant cost and timeline is unsustainable

→ Can't answer "what if" questions mid-study

→ Hard to explain methodology to unions or council

David

Compensation Analyst · County Gov

Runs classification work in-house. Spends the majority of each study on the matching step — reading descriptions, cross-referencing survey databases, making judgment calls. His decisions live in spreadsheets with no audit trail.

Pain points

→ Manual taxonomy matching is slow and inconsistent

→ No way to document or defend matching logic

→ Cost modeling requires starting Excel from scratch

Kevin

City Manager · Budget Approver

Approves study results and presents to city council and unions. Not deep in the methodology — needs results to be defensible, transparent, and fast. Budget forecast accuracy of 62% isn't good enough when the council is asking hard questions.

Pain points

→ Can't model different budget scenarios before approving

→ Results aren't ready until decisions are already made

→ No visibility into how comparators were selected

Key insight from discovery: David (the analyst) was the daily user. But Kevin's approval was the bottleneck. The product had to serve David's workflow speed while generating outputs that could survive Kevin's budget presentation — and the inevitable union challenge.

What to build — and how

Approach
Accuracy
Speed
Auditability

Status quo — consultant-led studies

External firm runs study; delivers static PDF

Medium
6–12 months
None

Keyword / rule-based job matching

Exact/fuzzy match job titles to taxonomy lookup

Low
Fast
Partial

Digitized point-factor scoring

Structured questionnaire scores job complexity factors

High
Medium
High

AI taxonomy matching + unified platform

Selected

AI reads job descriptions, matches to internal taxonomy with confidence scores; connected to benchmarking and live cost modeling

High
<4 wks
Full

The key insight: the value wasn't just in faster job matching — it was in connecting classification to benchmarking to cost modeling in a single data model. The internal taxonomy was already the foundation; the AI layer was the intelligence engine that made it scalable.

Three pillars, one connected platform

Organized around the three phases of every compensation study: classify the jobs, model the cost, position against the market.

CLASSIFY

AI-Powered Job Taxonomy & Matching

  • AI reads job descriptions, matches to internal taxonomy
  • Confidence scores on every recommendation
  • Bias detection and hierarchy consistency checks
  • Full audit trail — every match is logged and explainable

MODEL

Real-Time Cost & Financial Modeling

  • Live salary schedule modeling with org roll-ups
  • Scenario comparison — baseline vs. multiple proposals
  • Equity and compression analysis
  • Alerts: below-market positions, compression risks

POSITION

Market Intelligence & Benchmarking

  • Before/after market positioning comparisons
  • Configurable comparators per job family
  • Compensation philosophy targets (min/mid/max)
  • Auto-flag outliers and below-market alerts

What success looked like

North Star Metric

Time-to-complete a classification & comp study

Target: Under 4 weeks vs. industry average of 4–6 months. Every other metric cascades from this one.

Leading Indicators

  • AI taxonomy suggestion acceptance rate
  • Jobs classified per analyst session
  • Data ingestion completion rate
  • Scenarios modeled per study (depth of use)

Lagging Indicators

  • Study completion time (target: <4 weeks)
  • Attach rate to new Benchmarking deals (target: 40%)
  • ARR uplift for bundled customers (target: 15%)
  • Renewal rate and NPS for the module

Counter-Metrics

  • AI classification override rate (low confidence proxy)
  • Union or council challenges to methodology
  • Support escalations tied to data ingestion errors
  • Budget forecast accuracy post-study (should improve from 62%)

The analyst experience

From data import to final report — an eight-step workflow where AI does the heavy lifting at every stage.

Screen 1 — AI Classification Interface

Classification Intelligence · FY2025 Study
CLASSIFY | MODEL | POSITION
City of Pasadena · FY2025 62 / 147 classified
Step 3 of 8: AI Classification 42% complete
Current Position
CIVIL ENGINEER I
Public Works Grade: E-3
Job Description (excerpt)

"Performs entry-level civil engineering work including site inspections, plan review support, CAD drafting, and coordination with contractors on infrastructure projects..."

Position Data
Job FamilyEngineering & Technical
DepartmentPublic Works
Current RankLevel 3 of 8
FTEs4 incumbents
AI Taxonomy Match
High Confidence
Recommended Match
Junior Civil Engineer
Match Score 94%
AI Reasoning

"Duties align with entry-level civil engineering scope — site inspections, plan review, CAD support. Direct match to regional 'Junior' taxonomy benchmark positions. 14 peer agencies hold this exact classification."

Taxonomy Details
Job FamilyEngineering & Technical
LevelEntry-Professional (Rank 3)
Peer Agencies14 matches
Accept (94%)
Override
Flag

Screen 2 — Cost Modeling & Scenario Analysis

Cost Modeling — Step 7 of 8
CLASSIFY BENCHMARK MODEL
+ Add Scenario
Generate Report
$12.45M
Current Total Comp
$12.90M
Proposed Total Comp
+3.6%
Org Cost Impact
8
Alerts Pending
Scenarios
Baseline
Merit Adj.
Full Adj.
Salary Schedule — Proposed
Grade Min Mid Max Δ
E-3 $58K $68K $78K +3.5%
M-4 $82K $97K $112K +2.8%
P-6 $96K $113K $130K +4.1%
Action Required
Accountant II 12.4% Below Market

Current: $68.4K · Market P50: $77.4K

View & Adjust
Sr. Civil Engineer Compression Risk 5%

Supervisor/subordinate pay gap narrowing

Review Equity
6 positions No Action Needed

Within target range · No changes required

How I planned to ship it

1

Phase 1 — Taxonomy Foundation + Pilot Agencies

Pilot

Finalize the internal job taxonomy. Build the AI classification engine. Run with 3–5 lighthouse agencies to validate match quality, generate before/after case studies, and calibrate confidence score thresholds. This phase is about earning trust before scaling.

Success criteria: 85%+ analyst acceptance of AI taxonomy suggestions; study completed in under 6 weeks vs. prior 6-month baseline.

2

Phase 2 — Full Platform Launch (GA)

Q3 2026

Launch CLASSIFY + MODEL + POSITION as a unified add-on to existing subscriptions. Connect the full eight-step study workflow. Package with premium Labor Costing bundle for upsell.

Revenue target: 40% attach rate on new deals; 15% ARR uplift for bundled customers.

3

Phase 3 — Continuous Intelligence

Roadmap

Move from one-time studies to always-on market monitoring. Agency jobs are continuously scored against the taxonomy as the market shifts. Budget models update automatically as labor agreements change. The study never ends — it becomes a live system.

What I'd do differently

What worked

  • Framing the pitch around the unified platform, not just the AI feature — executives cared about "classification to budget in one place," not taxonomy matching
  • The audit trail was the trust mechanism — government analysts need to defend decisions to unions; every AI match being logged and explainable was a feature, not a footnote
  • Designing for the analyst (David) while making outputs readable by the approver (Kevin)

What I'd change

  • I'd run the pilot earlier in the build cycle — not after development, but as a design validation. Pilot agencies would have surfaced data quality issues before we built the ingestion layer around bad assumptions.
  • The taxonomy itself needed more time before the AI was layered on top. If the taxonomy has gaps, the AI confidently fills them with bad matches — garbage in, garbage out at 94% confidence.
  • Change management was underweighted. Agencies have used consultants for decades. The barrier wasn't features — it was trust in an entirely new process.

The through-line: In regulated, high-stakes environments, explainability is the product. An AI that's 94% accurate and opaque is less valuable than one that's 85% accurate with a clear rationale David can cite in a council presentation and Kevin can stand behind in a union negotiation.

Next case study

Mic Check — From Idea to Game