Bench(k) · The Evolution · A brief for Prime Capital

From a 24-row spreadsheet
to a 4-source analyst
in 90 seconds.

This is the story of how Debi's plan benchmarking workflow — one of the most reliable client-facing tools at Prime Capital — evolved from a manual Excel comparison to an AI-powered, cite-ready report that enriches from the full Form 5500 universe on every run.
2026 Q2 Prepared for Prime Capital leadership Powered by the (k) Suite
Industries
1251
Every PLANSPONSOR industry is now a benchmarkable peer cohort. 4.25× coverage.
Data Sources
14
PLANSPONSOR plus Vanguard, Fidelity, American Century — each citation numbered.
Plan Universe
869Kplans
Every Form 5500 filing for plan year 2024 enriches the report on every run.
Time per Review
30m90s
EIN pre-fill + real-time AI analysis. From one-at-a-time to portfolio scale.
The Evolution · Three Chapters

How we got here.

The original Excel workflow wasn't broken — it was a careful, professionally-maintained tool that produced real client value. Each chapter below preserves what worked and upgrades what was limiting.

Chapter 1 · The Origin

The Excel Spreadsheet

Plan_Industry_Review.xlsx
  • 12 industries digitized from PLANSPONSOR's 2026 DC Plan Industry Report.
  • 24 data points per industry — participation, deferral, match, vesting, auto-features, loans, Roth.
  • Manual data entry: Debi types each client plan's figures into the comparison columns.
  • Visual output: a clean benchmark grid the advisor could walk a sponsor through.
  • Single source: PLANSPONSOR data only, no participant-behavior context.
  • Manual synthesis: the advisor drew their own conclusions.
~30–45 minutes per plan review, start to finish.
Chapter 2 · The Digitization

The v1 SPA

prime-plan-review.html · Cloudflare Pages
  • Prime navy and gold branded report with preserved visual identity.
  • 15 SVG donut charts rendered live as the advisor types.
  • Key-metric stack with your-plan-vs-peer side-by-side comparison.
  • Template-based "AI Insights" — rule-driven strengths / concerns / opportunities.
  • One-click PDF export for the plan sponsor meeting.
  • Still single-source: PLANSPONSOR only, no cross-reference data.
~10–15 minutes — the interactive UI was faster than Excel, but the advisor still manually entered every field.
Chapter 3 · The Intelligence Layer

Bench(k) v2

bench-k.pages.dev · Claude Sonnet 4.6 · (k) Suite
  • All 51 PLANSPONSOR industries, grouped by data depth.
  • Four authoritative sources with numbered citations: PLANSPONSOR, Vanguard HAS 2025, Fidelity BFF Q4 2025, American Century 12th Annual.
  • Real Claude Sonnet 4.6 analysis — evenhanded, ranked, weighted strengths / concerns / opportunities.
  • Form 5500 enrichment: enter an EIN, the report pre-fills plan name, sponsor, city, balance — and auto-selects the industry from NAICS.
  • Dynamic peer cohort: "based on 36,188 plans" replaces "industry averages."
  • Citations auto-number as numeric footnotes in the report.
~90 seconds per plan review — including Form 5500 pre-fill + 15-second AI generation.
Side-by-Side

What changed, axis by axis.

Each row below is one dimension of the benchmarking workflow. The "Then" column captures the state under Excel and v1; the "Now" column captures v2.

Axis
Before (Excel & v1)
Now (Bench(k) v2)
Industries covered
12 industries, hand-selected
All 51 PLANSPONSOR industries +325%
Peer data sources
PLANSPONSOR only
PLANSPONSOR · Vanguard · Fidelity · American Century
Plan lookup
Manual typing, every field
EIN → Form 5500 pre-fill (869,889 plans)
Industry selection
Manual selection from dropdown
NAICS auto-mapping from Form 5500
Peer cohort framing
"Industry averages"
"Based on 36,188 peer plans in Form 5500"
AI analysis
Rule-based templates (if/else logic)
Claude Sonnet 4.6 — ranked, weighted, evenhanded
Source citations
Single "Source:" line at the bottom
Numbered footnotes inline + full citation block
Fidelity overlay
Not available
22-industry Fidelity cross-reference at year-end 2025
Participant behavior context
Not available
61 behavior norms from Vanguard, Fidelity, American Century
Generational segmentation
Not available
Boomer / Gen X / Millennial / Gen Z Fidelity data
Report output
Printed Excel or v1 print-to-PDF
Same print-to-PDF, Browser Rendering upgrade queued
Time per review
30–45 minutes (Excel) / 10–15 (v1)
~90 seconds end-to-end 20×+ faster
Enhancements · What Each One Does

Six material upgrades.

The jump from "digitized Excel" to "intelligence layer" happens along six specific axes. Each one was chosen because it either saves the advisor time or makes the report more defensible in front of a sponsor.

01 · Data Depth

From 12 to 51 industries

Every PLANSPONSOR industry cut is now a benchmarkable peer. Twelve of these — the original set — carry full plan-design detail (match structure, vesting tiers, auto-features, QDIA). The remaining 39 carry headline metrics with the "All Industries" aggregate overlaid for plan design. As additional PLANSPONSOR industry cuts are licensed, they graduate from "headline" to "rich" tier without code changes.

02 · Source Plurality

Four surveys, one report

PLANSPONSOR answers "what does this industry look like?" Vanguard's How America Saves 2025 answers "how are participants behaving nationally?" Fidelity's BFF Q4 2025 provides a 22-industry cross-reference with year-end 2025 data. American Century's 12th Annual surfaces the perception gap between participants and sponsors. Each fact in the AI analysis carries a {{cite:source}} token that renders as a numbered footnote.

03 · AI That's Actually AI

Template → Claude Sonnet 4.6

v1's "AI Insights" were well-crafted rule templates — if deferral is 1.5 points below peer, emit concern. Useful, but rigid. v2 calls Claude Sonnet 4.6 on every generation with the full plan + peer data. The model ranks material observations by weight, groups them (leads / trails / worth exploring), and writes in a consultative voice suitable for a sponsor meeting. Citations are strict — the model is instructed never to invent statistics.

04 · Form 5500 Enrichment

869,889 plans on tap

The entire Form 5500 corpus for plan year 2024 loads into the app at startup, indexed by EIN. Paste an EIN — with or without dashes — and the report pre-fills plan name, sponsor, city, state, and average account balance. NAICS auto-mapping drops the advisor straight into the right industry peer cohort. The report footer shows the cohort size: "based on 36,188 plans" rather than the abstract "industry averages."

05 · Citation-Ready Deliverable

Every stat, footnoted.

Every peer benchmark, behavior norm, and industry statistic in the AI analysis carries a numbered superscript. Below the analysis, a "Sources cited in this analysis" block auto-populates with full citations — publisher, sample size, publication date. This turns the report from "marketing collateral" into a document that can stand up to sponsor scrutiny.

06 · Scale

From one plan to a portfolio.

The Excel workflow topped out at roughly 1 plan / 30 min of advisor time. Bench(k) v2 runs at 1 plan / 90 sec — a 20× throughput improvement. For a Prime advisor with a book of 100 plans, that's the difference between a quarter-long project and a single-afternoon exercise. It opens the door to running benchmarks proactively rather than reactively.

The New Workflow

What 90 seconds looks like.

Here's the complete advisor flow from opening the app to downloading the client-ready report. The clock starts when the advisor types the EIN.

1

Open the app

Visit bench-k.pages.dev. Form 5500 universe begins loading in the background — status pill shows progress. By the time the advisor has the sponsor on the phone, the universe is loaded and ready. ~25s · one time per session

2

Enter the EIN

Type or paste the sponsor's EIN — dashes welcome or omitted. Plan name, sponsor name, city, state, participant count, and average balance auto-populate from the Form 5500 record. Industry is auto-selected from NAICS. <2s

3

Enter the current plan design

Segmented buttons for match, vesting, auto-enroll, Roth, loans, QDIA. Every input is click-through — no typing required. Report updates live as the advisor configures. ~30s for a standard plan

4

AI analysis runs

The NOX pulse-wave spinner fires. Claude Sonnet 4.6 analyzes the plan against peer benchmarks, behavior norms, and Fidelity overlays. Status messages cycle through the five phases of analysis. 15–20s

5

Review & export

Advisor reviews the generated narrative, scans the ranked strengths / concerns / opportunities, and clicks Export PDF. The report lands in a Prime-branded, citation-ready deliverable the sponsor can archive. ~15s

What we kept

Prime Capital's visual identity anchors every report — Prime navy & gold, the "Retirement & Wellness" lockup, the clean benchmark-grid layout the sponsor community already knows. The advisor's underlying workflow — identity, metrics, design, optional features — mirrors Debi's original four-stage Excel columns. Nothing that worked was discarded; everything that was slowing the advisor down was automated.

What This Enables

Beyond speed.

Faster reports are the visible improvement. The deeper change is what Prime can now do that it couldn't do before.

Proactive Benchmarking

Every plan, every quarter.

At 90 seconds per review, an advisor can run their entire book of plans in an afternoon. Quarterly benchmarking becomes a calendar event, not a project.

Prospecting Leverage

"Here's what your peers are doing."

Before a prospect meeting, the advisor runs Bench(k) on the prospect's EIN — open data — and walks in with a citation-backed benchmark of their current design. Zero sponsor effort required.

Defensible Recommendations

Every claim, sourced.

When the AI says "74% of peers offer auto-escalation," a numbered footnote ties it to the exact study and sample size. Sponsors, ERISA counsel, and plan committees can trace every data point.