Bench(k) v3 — Retirement Plan Benchmarking

Bench(k) v3

Loading every retirement plan in America

Bench(k) Help Center

Retirement Plan Benchmarking

Search 869,889 retirement plans from Form 5500 filings. Get instant scoring, peer comparisons across industry and geography, and AI-powered expert analysis — all in under 90 seconds.

Form 5500 Data

Plan Rank(k) Scores

AI Intelligence

15 Benchmark Charts

Export

Search for a Plan

Type a company name, plan name, or EIN in the sidebar search. Results appear after 3 characters with plan scores highlighted.

Review the Benchmark

15 charts compare the plan against peers in the same NAICS industry and participant size band. State-level peers shown when available.

Read AI Analysis

An AI-generated expert analysis highlights strengths, areas of concern, and actionable recommendations specific to the plan.

Export & Prospect

Export the full benchmark as a shareable report. Use insights for prospect meetings, plan reviews, and committee presentations.

Key Capabilities

Plan Rank(k) Scoring

Composite 1–10 score weighting employer contributions, participant deferrals, account balances, and plan maturity.

Peer Comparison

Compare against plans in the same industry, size band, and state. Decile rankings show exactly where a plan stands.

AI Expert Analysis

Senior advisor-level narrative powered by Claude AI, covering strengths, concerns, peer positioning, and next steps.

Feature Detection

Auto-enroll, QDIA, brokerage window, employer match, controlled group, and pre-approved document flags from PCC codes.

State-Level Peers

When 50+ plans exist in the same industry and state, see local comparisons alongside national benchmarks.

Compliance Flags

Late contributions, corrective distributions, bond coverage, and prohibited transaction flags all visible at a glance.

How Plan Rank(k) Scoring Works

Every plan receives a composite score from 1 to 10 based on four financial metrics, each computed as a decile ranking within the plan's participant size band.

40%

Employer Contribution

30%

Participant Contribution

20%

Avg Account Balance

10%

Plan Age

Decile Rankings

Each metric is ranked within the plan's size band (e.g., 101–250 participants). A decile of 8 means the plan is in the top 20–30% for that metric relative to similarly sized plans.

The composite score weights all four deciles using the percentages above, then maps to a 1–10 scale.

Compliance Penalty

Plans with compliance flags — late contributions, failed benefits, prohibited transactions, or corrective distributions — receive a penalty that reduces the composite score. Clean compliance records get no penalty.

Size Bands

Band	Participants	Typical Use
01	1–10	Micro plans, owner-only
02	11–25	Small business startup plans
03	26–50	Small employer core market
04	51–100	Growing companies
05	101–250	Mid-market sweet spot
06	251–500	Upper mid-market
07	501–1,000	Large plans
08	1,001–5,000	Enterprise plans
09	5,001+	Jumbo / mega plans

Form 5500 Filing Data

Bench(k) loads 869,889 plans from DOL/EFAST2 Form 5500 filings for plan year 2024. Every qualified retirement plan is legally required to file annually. This data is publicly available and attested by the plan administrator.

Plan & Sponsor Data

Plan name, sponsor name, EIN, city, state, plan effective date, TPA name, and participant count.

Financial Metrics

Total assets, employer contributions, employee contributions, distributions, average account balance, and asset growth.

Plan Characteristic Codes

PCC flags for auto-enrollment (2S), QDIA (2T), brokerage (2R), employer match (2K), controlled group (3H), pre-approved doc (3D).

Compliance Indicators

Late contributions, failed benefits, loss/fraud, prohibited transactions, bond adequacy, corrective distributions.

Peer Groups

Plans are grouped for comparison using two dimensions:

NAICS Industry Sector

20 sectors from the 2-digit NAICS code: Manufacturing, Finance, Healthcare, Technology, and more. Derived from the plan's NAICS code on the filing.

Participant Size Band

9 size bands from 1–10 up to 5,001+. Plans are ranked against peers of similar size to ensure fair comparison.

State-Level Peers

When 50+ plans share the same industry and state, a geographic peer comparison is shown alongside the national benchmark.

Decile Within Band

Each metric is ranked within the size band to show where the plan falls: 1st decile (bottom 10%) through 10th (top 10%).

AI Analysis

The optional "Expert Benchmark Analysis" uses Claude AI via a Cloudflare Worker proxy. Only plan-level summary metrics are sent — no participant data, no SSNs, no personal information. The AI narrative covers strengths, concerns, peer positioning, and recommendations.

Key Terms

Plan Rank(k) Score

Composite 1–10 rating based on four weighted decile rankings computed within the plan's size band. Higher is better.

Decile

A ranking that divides plans into 10 equal groups. Decile 7 means the plan is in the top 30–40% for that metric within its size band.

NAICS Code

North American Industry Classification System. A 6-digit code identifying the plan sponsor's industry. Bench(k) uses the first 2 digits to group plans into 20 sectors.

Form 5500

Annual filing required by DOL/IRS for every qualified retirement plan. Contains financial data, participant counts, plan features, and compliance information.

Size Band

One of 9 participant-count brackets (1–10 through 5,001+) used to ensure plans are compared against similarly sized peers.

Auto-Enrollment (PCC 2S)

Plan feature that automatically enrolls eligible employees at a default deferral rate unless they opt out. Increasingly common under SECURE 2.0.

QDIA (PCC 2T)

Qualified Default Investment Alternative — the investment option used for auto-enrolled participants. Typically a target-date fund.

Brokerage Window (PCC 2R)

Self-directed brokerage access that allows participants to invest beyond the plan's core fund lineup. Less common in smaller plans.

Employer Match (PCC 2K)

An employer matching contribution arrangement under IRC §401(m). The most common employer contribution design in 401(k) plans.

Controlled Group (PCC 3H)

Related employers that must be treated as a single employer for testing purposes. Affects coverage, nondiscrimination, and top-heavy testing.

Corrective Distribution

Distribution made to correct excess contributions or deferrals that failed nondiscrimination testing. A red flag in compliance review.

Safe Harbor

A plan design that automatically satisfies ADP/ACP nondiscrimination testing by meeting a required match or nonelective contribution formula.

3(21) vs 3(38) Advisor

ERISA fiduciary roles. 3(21) advisors recommend; plan sponsor decides. 3(38) managers have full discretion and fiduciary liability for investment selection.

Net Cash Flow

Total contributions minus total distributions for the plan year. Positive means the plan is growing; negative means more money left than entered.

Frequently Asked Questions

How is the Plan Rank(k) score calculated?

The score weights four decile rankings: Employer Contribution (40%), Participant Contribution (30%), Account Balance (20%), and Plan Age (10%). Each decile is computed within the plan's size band. A compliance penalty applies for plans with flags like late contributions or corrective distributions. Everything is calculated in your browser — no AI involved.

Where does the data come from?

From DOL/EFAST2 Form 5500 filings for the 2024 plan year. Every qualified retirement plan must file annually. The dataset contains 869,889 plans and is publicly available. The data is legally attested by each plan's administrator.

What does the AI analysis do?

The "Expert Benchmark Analysis" uses Claude AI to generate a senior advisor-level narrative covering strengths, areas of concern, peer positioning, and actionable recommendations. Only plan-level summary metrics are sent — never participant data. The AI narrative is supplemental commentary, not a fiduciary recommendation.

How are peer groups defined?

Primary peer groups use two dimensions: NAICS industry sector (20 sectors from 2-digit codes) and participant size band (9 bands from 1–10 to 5,001+). State-level peers are shown when 50+ plans share the same industry and state.

What are Plan Characteristic Codes (PCC)?

Codes reported on the Form 5500 that indicate plan features. Key codes: 2S (auto-enrollment), 2T (QDIA), 2R (brokerage window), 2K (employer match), 3D (pre-approved document), 3H (controlled group). Bench(k) detects these and shows adoption rates among peers.

Why does loading take 20–30 seconds?

Bench(k) downloads and parses the entire 299 MB Form 5500 dataset directly in your browser. This ensures no plan data is stored on any server. After the initial load, all searches and scoring happen instantly. Subsequent visits to the same session don't re-download.

Is this data secure?

Yes. All Form 5500 data is public. The dataset is parsed entirely in your browser. AI requests send only summary metrics via TLS 1.3 to a Cloudflare Worker, which proxies to Anthropic's enterprise API with zero data retention. Click the Security & Trust button in the top nav for full details.

Can I use this for prospect meetings?

Absolutely — that's a primary use case. Search for a prospect's plan, review the benchmark, then export a report you can share in a committee presentation or prospect meeting. The AI analysis provides talking points specific to that plan's strengths and gaps.

Release History

Complete redesign — sidebar-based layout with persistent search and contextual navigation.
Plan Rank(k) scoring — composite 1–10 scores computed across 4 weighted metrics within size bands.
15 benchmark charts — score gauge, contribution deciles, balance, plan age, cash flow, contribution balance, feature adoption, compliance, asset growth, and more.
Company & plan name search — search by name in addition to EIN, with live autocomplete across 869,889 plans.
State-level peers — geographic peer comparisons shown when 50+ plans share the same industry and state.
Security & Trust Center — full security disclosure modal matching (k) Suite enterprise standards.

51 PLANSPONSOR industries — expanded from 12 in v1 to the full survey universe.
Four data sources with numbered citations — PLANSPONSOR, Vanguard, Fidelity, American Century.
Form 5500 EIN lookup — 869,889 plans, pre-fills plan name, NAICS, industry, participation, balances.
NAICS auto-mapping — 108 prefix rules route any filing to its PLANSPONSOR bucket.
Claude AI analysis — expert narrative with inline citations.
Search(k) deep-link — "Benchmark this plan" button in Search(k) opens Bench(k) pre-loaded.

Vault(k) save integration — save benchmarks to your advisor vault for CRM tracking.
Browser Rendering PDF — pixel-perfect PDF export without the browser print dialog.
"Find similar plans" reverse link — jump to Search(k) pre-filtered by industry and size.
Multi-plan comparison — side-by-side benchmark of 2–5 plans in a single view.

All Systems Secure

Last verified: today

API-First Architecture

Bench(k) processes all plan data in your browser. The 299 MB Form 5500 filing dataset downloads directly from a public storage bucket and is parsed client-side. No plan data is stored on any server. When you request an AI analysis, only summary metrics are sent to the Cloudflare Worker — never raw filing data, never participant-level information.

Secure Processing Pipeline

Your Browser

TLS 1.3

Cloudflare Worker

Claude API

Auto-Delete

Core Security Features

Zero Training

Your plan data is never used to train AI models. Anthropic's enterprise API guarantees zero training on customer inputs.

End-to-End Encryption

All data in transit is encrypted via TLS 1.3. At rest, Anthropic encrypts with AES-256.

Zero Data Retention

AI analysis prompts and responses are not stored by Anthropic. Data exists only for the duration of the API call.

No Human Review

No Anthropic employee sees your data. Enterprise API calls bypass all human review pipelines.

Compliance Certifications

SOC 2 Type II

ISO 27001

HIPAA

GDPR

CCPA

Consumer vs Enterprise API

Bench(k) uses Anthropic's enterprise API — not the consumer chat product. This distinction matters for data governance:

Feature	Consumer (claude.ai)	Enterprise API
Data used for training	Possible	Never
Human review of inputs	Possible	Never
Data retention	30+ days	Zero
SOC 2 compliance	Partial	Full
HIPAA eligible	No	Yes

AI Provider: Anthropic

Claude API — Enterprise Tier

Model: Claude (via Cloudflare Worker proxy)
Certifications: SOC 2 Type II, ISO 27001, HIPAA eligible
Data handling: Zero retention, zero training, no human review
Usage in Bench(k): AI-generated expert analysis of plan benchmark data only. Plan scoring, peer comparisons, and all charts are computed entirely in-browser — no AI involved.

Plan Rank(k) scores are calculated entirely algorithmically in your browser — no AI is involved in scoring. AI is only used for the optional "Expert Benchmark Analysis" narrative section.

Secure Processing Pipeline

When you search for a plan, here's exactly what happens:

Form 5500 CSV

Browser Parse

Local Scoring

AI Analysis

Auto-Delete

What Data Is Sent to AI

Summary Metrics Only

Plan name, EIN, contribution averages, score, decile rankings, plan feature codes (yes/no flags). All of this is already public Form 5500 data filed with the Department of Labor.

Never Sent

No participant names, no SSNs, no individual account balances, no personal information of any kind. The AI sees only plan-level aggregates that are already public record.

The entire 299 MB Form 5500 dataset is parsed locally in your browser. It is never uploaded to any server. The raw CSV data stays in browser memory and is discarded when you close the tab.

Regulatory Framework

ERISA

Bench(k) uses 100% public Form 5500 data and quantitative scoring to support the prudent process for plan evaluation. No proprietary participant data is accessed or stored.

DOL Fiduciary Rule

AI-generated analysis is clearly labeled as supplemental commentary. All scoring is algorithmic and transparent — weighting is disclosed (EC 40%, PC 30%, AB 20%, PA 10%).

DOL Cybersecurity

Zero data retention, TLS 1.3 encryption in transit, AES-256 at rest, enterprise-grade API with SOC 2 Type II certification from both Cloudflare and Anthropic.

GDPR / CCPA

No personal data is collected or stored. No cookies are set beyond what the browser requires. No tracking, no analytics, no third-party data sharing.

Security Controls

No backend database or server storage
All computation runs client-side in the browser
API key stored as encrypted Cloudflare Worker secret
CORS allowlist restricts API access to authorized origins
Zero data retention on AI provider (Anthropic enterprise API)
No participant-level or personal data ever leaves the browser
Source data is public DOL/EFAST2 filings — not proprietary
Export produces a local file — nothing is uploaded

Frequently Asked Questions

Is my plan data stored on any server?

No. The Form 5500 dataset is downloaded to your browser and parsed locally. When you request an AI analysis, only summary metrics (plan name, aggregated contribution data, feature codes) are sent to the Cloudflare Worker, which proxies them to Anthropic's API. Neither the Worker nor Anthropic retains the data after the response is generated.

Does AI training use my plan data?

Never. Bench(k) uses Anthropic's enterprise API, which contractually guarantees that customer inputs are never used for model training. This is different from the free consumer chat product.

Can Anthropic employees see what I send?

No. Enterprise API calls bypass Anthropic's human review pipelines entirely. No Anthropic employee sees your prompts or the model's responses.

How is the Plan Rank(k) score calculated?

The score is computed 100% algorithmically in your browser — no AI is involved. It weights four decile rankings: Employer Contribution (40%), Participant Contribution (30%), Account Balance (20%), and Plan Age (10%). Each decile is computed within the plan's size band. A compliance penalty applies if flags are present.

Where does the Form 5500 data come from?

From the DOL's EFAST2 filing system. Every qualified retirement plan must file a Form 5500 annually. The data is publicly available and legally attested by the plan administrator. Bench(k) uses the 2024 plan year dataset covering 869,889 plans.

Is the AI analysis a fiduciary recommendation?

No. The AI-generated "Expert Benchmark Analysis" is supplemental commentary designed to help advisors identify discussion topics. It is clearly labeled as AI-generated and should not be treated as legal, tax, or investment advice. All scoring is transparent and algorithmic.

What happens when I export a report?

The export generates a file entirely in your browser using the data already displayed on screen. Nothing is uploaded to any server. The file is saved to your local device.