Advisor operating system
Advice you cannot source is advice you cannot defend.
Retirement plan advisors are asked to justify recommendations years after they make them. RetireForge is built so the justification is a byproduct of the work — every fee, every illustration, every recommendation reproducible from the filing, the engine version, and the limit set it came from.
Example plan record. Forgewell Manufacturing is a synthetic sponsor rather than a client, and each figure below is followed by the source it came from.
Sourcing is a property of the system, not a discipline you keep.
Nothing here depends on an advisor remembering to record where a figure came from. Three mechanisms run underneath every surface in the product, on every action, and they are the reason the file is already complete by the time a committee, a plan sponsor, or a regulator asks to see it.
Deterministic by design
Regulated math — fees, allocations, nondiscrimination testing, benchmarks, funding — runs in versioned engines, reproducible from its inputs, the engine version, and the IRS limit set. The language model never computes a number.
Provenance on every field
Each figure on a plan record carries where it came from, how fresh it is, and how confident we are. When two sources disagree, the conflict is recorded beside the winning value — never quietly resolved away.
An audit trail you can hand to a committee
Append-only events on every regulated action, visible in the interface rather than buried in a log file. Automation records what triggered it, which rule version ran, what it produced, and who approved it.
Four stages, one plan record
Each stage captures structured data the next one consumes. Nothing is re-keyed, and nothing is lost between the pitch and the annual review.
Find
Search the public Form 5500 universe for plans that fit your practice, and pull the ones you want onto your book.
Diagnose
Price the incumbent against peer filings, project a decade forward, and read the adoption agreement the sponsor has been living under.
Win
Model what a better design is worth to the owners, turn it into a branded proposal, run a structured RFP, and track the conversion to the day it goes live.
Serve
Run quarterly committee prep, keep the fiduciary file complete, and let automation propose the follow-through.
Fee benchmarking
What plans like this one actually pay
Every 401(k) and profit-sharing plan files its costs with the Department of Labor. Here is the real 2023 distribution. Drag the marker to any all-in fee and see where it falls against its peers.
A plan this size paying 1.35% all-in costs more than 50% of the 247,295 comparable plans that filed in 2023.
EFASTU.S. Department of Labor Form 5500 filings, plan year 2023. All-in cost is disclosed Schedule C/H fees plus fund expense as a share of assets; plans reporting no fee data are excluded, and bars are clipped where most plans cluster at the size-band floor. Illustrative of peer ranges — not a quote or an actuarial certification.
Start from what the filings already say
Every qualified plan in the country files a Form 5500. RetireForge turns that public record into a prospecting surface and a diagnostic, so the first conversation starts with evidence instead of a questionnaire.
Prospecting
Faceted search across the national Form 5500 corpus — plan size, assets, participants, provider, state, city, ZIP — with Schedules A, C, H, R and MEP loaded behind it. The incumbent recordkeeper, the insurance carrier, and the pre-approved document sponsor resolve as three separate, confidence-tiered signals: a plan can carry all three, and they are never conflated. Searches and lead lists are saved, not re-typed.
Plan diagnostic
A branded benchmark report generated straight from a plan's own filings — fees, participation, and design, set against comparable plans. Deterministic: the same filing and the same engine version produce the same report, every time.
The plan record
Adding a plan puts it on your book and writes an audit event. From there it accumulates: filing history, realized rate of return derived from Schedule H, an assets series, and a structured profile of what you learned in the sponsor meeting — pre-filled from the filings so a broker-of-record change is not an afternoon of re-keying.
Price the plan the sponsor already has
Most retirement plan advice loses its first argument on cost, and most cost arguments are made with a spreadsheet nobody can reproduce a year later. These three surfaces share one fee taxonomy and one audit trail.
Fee benchmarking
The incumbent against provider quotes, side by side, broken out per participant and in basis points across recordkeeping, third-party administration, investment expense, revenue sharing, advisory, and custody. Peer-percentile placement, fee leakage, and revenue-share exposure mined from Schedule C and Schedule H. Compare against a curated library of standard provider pricing rather than re-keying a schedule you have already normalized once.
Ten-year projection
Not a single rate applied to today's number. Every fee component is re-evaluated against each future year's own assets and headcount — tiers crossed, minimums hit, included participants consumed — so the year a plan grows out of its pricing shows up as the step it actually is. Any number of schedules projected side by side, exportable as a branded report.
Plan design review
Drop in the adoption agreement. Document intelligence extracts the plan-design fields with per-field confidence and provenance, anything below threshold is routed to you rather than assumed, and a deterministic rule catalog — no language model anywhere in it — flags design considerations against real statutory citations.
Show the owner what a better plan is worth
One census, stacked scenarios, and a versioned actuarial engine underneath. Output is illustrative — offered for discussion, not as an actuarial certification — and says so on every page it reaches.
Plan design illustrations
Safe harbor, then profit sharing, then cash balance, stacked over a single census. Coverage, ADP/ACP, top-heavy, gateway, and cross-testing all run in the engine. Allocation groups, vesting and forfeiture treatment, SECURE 2.0 startup credits, and a cash-balance funding and deductibility estimate come with it, and scenarios are versioned so last quarter's illustration is still reproducible next year.
Design recommender
Given the census and what the owners are trying to accomplish, the engine searches design space and returns candidate structures with the testing consequences attached — and the rationale for each, in terms you can repeat to a committee. It proposes; it does not decide.
Owner-only plans
A dedicated solo 401(k) and cash balance illustration with earned income computed properly for a self-employed owner: the deferral, the catch-up, the profit sharing, and the pay credit as one constrained stack, with the combined-plan deduction limits enforced rather than annotated.
Diligence on a pooled plan, not a brochure
Pooled employer plans are sold on a summary sheet and bought on trust. Every pooled plan provider registers with the Department of Labor before it may operate one, and those registrations — plus the plans' own filings — are a public record nobody has been reading systematically. We read it.
The universe
Every registered pooled plan provider on file with the DOL and every pooled plan they have reported, ingested from Form PR and cross-walked to those plans' own Form 5500 filings. Independent certification status comes in from the CEFEX registry. Provider acquisitions and cessations are on the record, because who runs a plan today is not who ran it when the sponsor signed.
A rubric, not a rating
Seventeen deterministic criteria — trustee affiliation, who provides which seat, fee level against standalone peers, track record, governing-document terms — evaluated into conformance findings with their qualifiers and confidence attached. There is no composite score, on purpose. A single number is what makes a due-diligence file feel finished when it is not.
Head to head, and the market around it
Two providers side by side in a fixed order, so the comparison cannot be arranged to flatter one of them. Around it, the shape of the industry: how concentrated it is, how often a provider hires itself for the seats it is supposed to be monitoring, and how rare independent certification actually is.
Into the proposal, with its limits
A comparison persists as a versioned scenario and becomes a sponsor-facing section with a citation on every flag. Where a provider has not been assessed, the document says so in print — coverage language that survives the printer, because a committee packet is where an honest caveat stops being visible on most platforms.
From the diagnosis to the signed plan
The proposal, the RFP, and the conversion share one census, one fee picture, and one record of what the committee actually saw when it decided.
Proposals
Compose the prospect's facts, the fee comparison, the illustration, and the pooled-plan diligence into one document. Every figure in it passes through from the engine that computed it, and the required fee-disclosure language is generated from the arrangement rather than pasted in.
RFP and conversion
Scoring criteria and a question library, an invite-only portal where providers answer in structure rather than in prose — fund lineup mapped, fee schedule captured as a schedule — and a side-by-side aggregation that feeds the fee engine directly. Deciding freezes an immutable snapshot of exactly what was compared. Conversion milestones, from file to blackout to live, flag slippage before it becomes a problem.
Your name on the output
Set the firm's logo, colors, and name once and every client-facing artifact carries them — every report, every proposal, every externally shared document, including the pages a sponsor opens without logging in.
The quarterly meeting, prepared in advance
Committee prep is the workflow the year turns on. RetireForge organizes the platform around it, so the file is complete before anyone asks to see it.
A queue, not a dashboard
What is waiting on you, across the whole book: drafts to approve, meetings scheduled, action items past due, and the week's activity as it happened. Assets, pipeline, and plan mix are on the same screen as trends rather than as decoration.
Committee and meetings
Roster, agenda, minutes, action items, and acknowledgments as one cycle, across every plan on the book rather than one at a time. Members are deactivated, never deleted, because the question is always who was on the committee then. Minutes export to Word or PDF for the file.
Fiduciary vault
Versioned documents, each with a content hash. A new version is a new record — nothing is ever overwritten. Share externally with a PIN and an expiry, and every access, allowed or denied, is logged.
ERISA research
Search primary authority — statute, regulation, and agency guidance — and resolve a citation to the text it actually says. Results are ranked by what kind of authority they are, and every excerpt is returned verbatim from the source. Nothing is paraphrased on the way to you.
Playbooks
Declarative automation: when this event fires and these conditions hold, propose these actions. Draft the notice, flag the fund, open the task. Every action lands as a proposal a human approves — nothing executes on its own.
Data reconciliation
See which fields you actually hold on a plan, which source each came from, and how stale it is. Feeds that drift from their expected layout are quarantined whole rather than half-parsed, and disagreements between sources surface as conflicts to resolve.
For broker-dealers and aggregators
Oversight across every firm in the network
A parent organization sees its whole subtree without leaving the platform, and without a spreadsheet arriving by email every quarter.
- Plans by lifecycle stage, sponsors, and advisors by role — rolled up across every child firm and drillable down to one.
- Recent scenario and proposal activity per firm, so you can see which practices are actually working the tools.
- An advisor who sits in two places in the hierarchy is modeled as one person with two affiliations, not as two accounts that have to be reconciled by hand.
- Relationships between firms are typed as what they legally are — referral, solicitation, sub-advisory, joint servicing — with the use restrictions that attach to each, and commercial splits recorded bilaterally rather than in a side letter.
- Subtree visibility is enforced by row-level security in the database itself, not by a filter in application code that someone can forget to apply.
For pooled plan providers
Run a PEP without losing control of the disclosure
Three distribution models, one fee engine, and a hard rule that a fee can never drift out of sync with what has been disclosed to the plan sponsor.
- Open PEPs use a provider-maintained template that locks completely once a plan adopts it.
- Firm-exclusive PEPs let the advisor fee change within a ceiling the broker-dealer sets — but the change routes through your approval queue rather than applying itself.
- Advisor-owned PEPs cap the advisor fee at the plan owner's limit and apply changes directly, with a switchable fee-disclosure presentation.
- Sponsor identity stays masked to the provider until the plan is won.
- List the plan in the marketplace and control, per resource, which firms can see what — the same grant mechanism the rest of the platform uses, not a second permission system.
The model drafts. The engines decide. You approve.
Language models are good at prose and bad at arithmetic, and a retirement plan is arithmetic. RetireForge puts a wall between the two, and enforces it in code rather than in policy.
Numbers are verified, not trusted
Every regulated number comes from a deterministic engine. Before an AI-written draft is saved, each figure in its prose is verified against the structured data the model was handed. A number that cannot be traced back fails closed — the draft is rejected, not flagged.
Citations are verified, not generated
References to ERISA and the Internal Revenue Code are checked against the same corpus of statute, regulation, and agency guidance the research surface searches, before a draft is persisted. A citation the model invented never reaches your file.
Every write is a proposal
Narrating a prospect, drafting a meeting agenda, reading facts out of a provider contract, explaining a conflict between two data sources — each lands as a proposal, with the grounded data attached for you to inspect. Nothing is written to your book until a person approves it.
Ask in plain language
Press ⌘K and ask for what you need. Natural language routes to the right surface against a fixed catalog of destinations, so the assistant can navigate the product but cannot invent a place to send you.
Built for the file, not the demo
You are a fiduciary, and the platform you keep the evidence in is part of that responsibility.
- Multi-tenant isolation enforced in Postgres with row-level security, with an isolation test on every tenant-scoped table that runs on every build.
- Sensitive census fields — Social Security number, date of birth, compensation — encrypted at the field level, separately from the database at rest.
- Append-only audit events on every regulated action, surfaced as a product screen rather than an internal log.
- Every external dependency — recordkeeper data feeds, document parsing, email, billing — sits behind an adapter. Replacing a partner is configuration, not re-architecture.
- Built against the SOC 2 criteria from the first commit, with controls that emit their own evidence automatically. The independent audit is not yet complete, and we will not claim otherwise.
See it against your own book
We will walk through prospecting, a fee benchmark, and a cash balance illustration on plans you actually care about.
Sign in