10
Contribution AI Use Cases
28
Retirement Account Codes
Why a separate cut: retirement
contributions (ACHC to Q*/R* accounts) are NOT just deposits. They carry IRS limits by account type, MAGI-based deductibility, prior-year windows, excess-contribution penalties (6%), 5498 reporting, Roth-eligibility phase-outs, and catch-up rules. The generic
AI Use Cases page covers the 10-step pipeline as a whole; this page focuses on the contribution-specific AI moments where advisors and clients make costly mistakes today.
2026 Key Limits (planning reference): IRA $7,000 regular + $1,000 catch-up (50+) | 401(k) elective $23,500 + $7,500 catch-up (50+) + $11,250 super-catch-up (60–63 under SECURE 2.0) | SEP 25% of comp up to $70k | SIMPLE $16,500 + $3,500 catch-up | HSA $4,300 single / $8,550 family + $1,000 catch-up (55+) | Roth IRA phase-out single $150k–$165k, MFJ $236k–$246k
Recommendation for POC: start with R1 (Limit Pre-Check) and R5 (Excess Detection) — both are deterministic-rules-plus-AI-explanation patterns, deliver clear NIGO reduction and client protection, and sidestep the complexity of R3/R6 tax-advice territory. Add R2 (Prior-Year Coach) during Jan–Apr season window for a quick seasonal win.
R1. Contribution Limit & Eligibility Pre-Check Recommended
Problem: Today, an advisor can submit an ACHC that will be rejected later (or worse, accepted but trigger a 6% excess penalty) because it exceeds the IRS limit or the account type doesn’t support that contribution category. Rules are scattered across BETA, IRS Pub 590-A, and tribal knowledge.
What it does
- Aggregate view: sums the client’s YTD contributions across all Q*/R* accounts at LPL (plus rolling 5498 history) and compares against the applicable 2026 limit.
- Rule evaluation: checks age (catch-up eligibility), account type (IRA / SEP / SIMPLE / 401(k) / HSA have different limits), and contribution category (regular / catch-up / rollover / conversion — these have different buckets).
- LLM-authored explanation: if the check fails, the advisor sees: “This $3,000 contribution would put Jane at $7,500 for 2026 traditional IRA. Limit is $7,000 ($8,000 with catch-up — she turns 50 on Oct 12). Over-contribution of $500 triggers 6% penalty unless removed with earnings by Apr 15, 2027.”
- Cross-account coordination: if she also has a Roth IRA, the $7k limit is shared — system flags the combined total.
Where it plugs in
Pre-submit advisor UI + EligibilityService inline rule extension in cam-movemoney-process-api. Two-layer design: deterministic rules engine evaluates limits (auditable, cheap); Bedrock Claude Haiku formats the explanation and “what to do about it” message.
Implementation
| Data sources | cam-movemoney-system-api (YTD contributions), BETA 5498 history, AccountDetailsResponse (IRAType, DateOfBirth), client profile (filing status, workplace plan indicator) |
| Rules engine | Deterministic C# inline rules — NOT the LLM — evaluate limits. Same audit pattern as existing EligibilityService (BBK/BFL/BMM) |
| LLM role | Format explanation only, using rule output + tokenized context (no raw SSN/account) |
| Latency | ≤ 400 ms (rules) + ≤ 800 ms (explanation, async-displayed) |
| Key dependencies | Babujee Arumugam (pipeline), Michael Mcguire (taxonomy signoff), Tax/Compliance team (limit matrix ownership) |
R2. Prior-Year vs Current-Year Contribution Coach
Problem: Between Jan 1 and Apr 15, every contribution can be coded prior-year (2025) or current-year (2026). Advisors guess, mis-code, then have to file corrected 5498s. Clients also lose optimization opportunity — e.g., filling 2025 room before 2026.
What it does
- At submission, checks remaining prior-year room vs remaining current-year room.
- Recommends the plan year that maximizes tax benefit given the client’s situation (2025 AGI known, 2026 is projection).
- Surfaces the IRS cutoff (Apr 15, 2026 for 2025 contributions) and countdown.
Where it plugs in
Advisor UI pre-submit widget. Rule-driven recommendation + LLM for the “why” narrative.
Seasonal lift: this use case is highest-value Jan–Apr. Build once, activate yearly. Expect 10–30% of Q1 contribution volume to benefit.
R3. Roth vs Traditional Recommendation Engine
Problem: Roth-vs-Traditional is arguably the single highest-stakes retirement decision. Advisors default to what the client asks for; many clients don’t know which to ask for.
What it does
- Takes client inputs: current marginal tax bracket, expected retirement bracket, time horizon, existing pre-tax vs Roth balances, state tax exposure.
- LLM synthesizes a recommendation with reasoning: “Given Jake is 32, in the 22% bracket, expects higher brackets later, and has mostly pre-tax 401(k), Roth IRA would diversify his tax exposure and capture compounding at today’s lower rate.”
- Flags edge cases (near Roth MAGI phase-out, near retirement, expected relocation to no-income-tax state).
Risk
Crosses into investment advice territory. Must be positioned as an educational tool, with explicit disclaimers, advisor-owned final decision, and compliance review of prompt templates. Not a fiduciary recommender.
Key dependencies: LPL Compliance + Legal + Advisor Supervision; likely a 3–6 month governance runway before production.
R4. MAGI / Deductibility Calculator Assistant
Problem: Traditional IRA deductibility depends on Modified AGI + whether the client or spouse is covered by a workplace plan. The rule matrix has ~8 cells and changes yearly. Advisors get it wrong or skip the calc and let the client “figure it out at tax time.”
What it does
- Collects (or imports from CRM) filing status, MAGI, workplace-plan coverage flags.
- Evaluates the 2026 phase-out schedule (MFJ covered: $130k–$150k; MFJ spouse covered: $244k–$254k; single covered: $81k–$91k; no coverage: full deduction).
- Returns: full / partial / no deduction, with the partial-deduction dollar amount calculated.
- LLM writes the advisor-facing explanation and the “what to tell the client” version.
Where it plugs in
Pre-submit advisor UI widget; also triggerable standalone as a planning tool.
R5. Excess Contribution Detection & Remediation Recommended
Problem: Excess contributions incur a 6% IRS penalty per year until removed. Detection today is often retroactive — the client finds out at tax time. LPL can prevent this proactively.
What it does
- Runtime: at contribution time, R1’s pre-check rejects obvious over-contributions.
- Ongoing monitor: sidecar service scans client aggregate across accounts (including external 5498 data when available) to catch contributions made outside LPL channels.
- Remediation workflow: if detected, AI drafts the “return of excess contribution with earnings” request, pre-computes the earnings calculation, and surfaces the deadline (tax filing date + extensions).
- Alerts the advisor with a prioritized queue: “5 clients have excess contributions pending Apr 15 deadline — earliest remediation window closes in 11 days.”
Where it plugs in
Sidecar service; reads cam-movemoney-system-api audit + BETA 5498 history. Writes nothing directly; presents workflow cards to advisor.
Why it wins: clear client value (saved IRS penalties), clear advisor value (caught-the-error credibility), reputational risk reduction for LPL.
R6. Backdoor Roth Workflow Assistant
Problem: Backdoor Roth (nondeductible traditional IRA contribution → conversion to Roth) trips on the pro-rata rule when the client has existing pre-tax traditional IRA balances. Done wrong, it creates unexpected taxable income.
What it does
- Detects existing pre-tax balances across all traditional IRAs (including SEP, SIMPLE) — the IRS treats them as one for pro-rata.
- Calculates the taxable portion of a proposed conversion.
- Guides the two-step workflow with a checklist: step 1 nondeductible contribution (track basis on Form 8606), step 2 conversion (report on Form 1099-R and 8606).
- Warns when pro-rata would make this a bad idea (“Client has $120k in rollover IRA; a $7k Roth conversion would be 94% taxable”).
Risk
High risk — tax-law adjacent, IRS scrutiny of backdoor strategy continues. Position as workflow + calculation support for advisors who already understand the strategy. Not a recommender.
R7. Spousal IRA Optimization
Problem: A nonworking spouse can contribute to their own IRA based on the working spouse’s earned income. This is routinely missed — hundreds of MFJ households at LPL never fund the nonworking spouse’s IRA.
What it does
- Identifies MFJ households where only one spouse has earned income AND only one spouse has YTD IRA contributions.
- Surfaces the opportunity to the advisor: “Client’s spouse has $7k of untapped spousal IRA room.”
- Pre-drafts the contribution request if advisor accepts.
Where it plugs in
Proactive advisor dashboard widget; triggered during Q1 planning + at year-end.
R8. Rollover Intent Classifier & 1099-R Coder
Problem: An incoming ACHC can be: a regular contribution, a direct rollover (trustee-to-trustee), an indirect 60-day rollover, or a conversion. Each has different 1099-R / 5498 coding and IRS limits (e.g., one-rollover-per-year-per-IRA rule for indirect).
What it does
- At intake, LLM reads free-text intent from advisor + check/ACH metadata + originating-custodian info.
- Classifies: regular contribution, direct rollover (code G), indirect 60-day rollover (code 7), conversion (code 2), recharacterization, etc.
- Checks one-per-year rule against 12-month rolling window.
- Surfaces the 5498 / 1099-R codes that will be applied and lets the advisor confirm.
Risk
High risk — misclassification creates incorrect IRS reporting. Classifier is advisory only; final code always human-confirmed. High-confidence predictions pre-fill; low-confidence ones surface the rule and ask.
Key dependencies: Babujee Arumugam (BETA 1099-R integration), Tax Operations team.
R9. 5498 Pre-Close Review
Problem: IRS Form 5498 (contribution reporting) is filed in May for the prior year. Errors at filing time trigger corrected 5498s, which are painful and visible to clients. By the time errors are caught, the tax-filing window may have closed.
What it does
- In early January, agent sweeps all 5498-bound contributions for the prior year.
- Flags anomalies: unusual prior-year codings past Apr 15, total contributions near limit, rollovers without source documentation, excess-contribution candidates, missing basis on non-deductible.
- LLM summarizes: “3,412 accounts; 47 flagged. 12 near-limit, 8 suspected excess, 27 rollover source unclear.”
- Generates remediation worklist for Tax Ops team.
Where it plugs in
Scheduled batch job in Tax Ops tooling. Reads cam-movemoney-system-api + BETA. Writes to Tax Ops queue only.
R10. Saver’s Credit & Catch-Up Opportunity Alerts
Problem: Two tax-advantaged opportunities routinely go unclaimed:
- Saver’s Credit — low/moderate-income clients get a 10–50% credit on first $2k contribution; most advisors don’t track this.
- Catch-up contributions — turning 50 unlocks $1k additional IRA, $7.5k additional 401(k); turning 60–63 unlocks SECURE 2.0 super-catch-up of $11,250.
What it does
- Birthday-triggered alerts: “Client turns 50 next month — catch-up contribution eligibility unlocks.”
- Income-triggered alerts: “Client’s reported AGI qualifies for Saver’s Credit; a $2k Roth contribution captures a $400 credit.”
- Proactive outreach templates drafted by LLM.
Where it plugs in
Proactive advisor dashboard; CRM integration. Read-only, purely informational.
Pipeline: Where Retirement-Contribution AI Inserts
| Stage | Today | Retirement AI Addition |
| CRM / Advisor UI (pre-submit) | Static form | R1 Limit check • R2 Year coach • R3 Roth/Trad • R4 MAGI calc • R8 Rollover classifier |
| Pipeline Step 6 (EligibilityService) | BBK/BFL/BMM only | R1 inline limit rules for ACHC to Q*/R* |
| Pipeline Step 7 (ComplianceService) | Static rules | R8 Rollover one-per-year check |
| Rejection path | Terse error code | Feeds generic UC #2 Rejection Explainability with retirement-specific context |
| Post-commit audit trail | cam-movemoney-system-api audit | R5 Excess monitor • R9 5498 pre-close |
| Proactive advisor dashboard | Manual reports | R7 Spousal IRA • R10 Saver’s Credit / Catch-up |
| Conversion workflow (separate flow) | Manual two-step | R6 Backdoor Roth assistant (guardrail-heavy) |
Dependencies & Data Sources
| System | Provides | Gap Today |
| cam-movemoney-system-api | YTD contribution audit, transaction metadata | Needs contribution-year + contribution-type columns (see Unknowns #3) |
| BETA | Historical 5498 data, 1099-R coding, PECO mapping | Programmatic 5498 history access may require new API (Babujee + BETA team) |
| AccountDetailsResponse | IRAType, DateOfBirth, IRAMktValueYE | Must include all IRA sub-types — see Unknowns #1 |
| Client profile / CRM | Filing status, MAGI estimate, workplace plan coverage | MAGI often not stored; may require advisor entry for R4 |
| External custodian data | Contributions outside LPL (for full aggregation in R5) | Not available — R5 will have LPL-only view until external 5498 import exists |
| IRS rule matrix | Annual limits, MAGI phase-outs, catch-up thresholds | Needs owned, versioned source of truth (Tax Ops team) |
Governance & Risk
- Tax advice boundary: R3, R4, R6 cross into advice territory. Explicit disclaimers + Compliance/Legal signoff required. Classify these as “advisor decision-support,” not “recommendations to the client.”
- Audit trail for limit decisions: R1’s limit matrix is a point-in-time rule reference. Version the matrix (with effective dates) so that audits can reconstruct what the system believed the limit was at contribution time.
- PII handling: never send SSN, DOB, or full account numbers to LLM. Use tokens (
{AGE_BAND}, {FILING_STATUS}, {LIMIT_BUCKET}).
- Model risk: R1/R5 should fall open to manual Tax Ops review on timeout — never auto-block a valid contribution because the AI timed out.
- Regulator audit (IRS): log every limit calculation with inputs, matrix version, output. Retention per SEC 17a-4 (6 years).
- SECURE 2.0 tracking: rules are changing annually through 2027. Rule matrix owner must commit to a refresh cadence; AI explanations should cite the rule version.
Next Steps (Retirement Contribution Track)
- Week 1: Own the 2026 retirement-limit matrix. Michael Mcguire + Tax Ops sign off; store as versioned config in cam-movemoney-system-api.
- Week 2: Prototype R1 (Limit Pre-Check) using deterministic rules + Claude Haiku explanation on 10 real YTD-contribution samples.
- Week 3: Prototype R5 (Excess Detection) as a read-only dashboard. Identify 20 historical excess cases from BETA 5498 data; verify the agent catches all 20.
- Week 4: R2 (Prior-Year Coach) ready for Q1 2027 season; turn on Jan 1. Measure advisor adoption.
- Month 2–3: R4 (MAGI Calc) scoped; Compliance review of prompt templates for R3 (Roth/Trad).
- Month 3+: R8 (Rollover Classifier) — partner with Tax Ops on accuracy bar; pilot against 100 historical rollovers.
- Deferred to 2027: R6 (Backdoor Roth) — requires tax-law-change monitoring; R9 (5498 Pre-Close) — align to Tax Ops annual cycle.
Strategic framing: the NLZ migration of ACH is a rare opportunity to build retirement-contribution intelligence into the pipeline instead of bolting it on later. R1 + R5 alone would eliminate most advisor NIGO and most client excess-contribution surprises — a clear “Day 1 NLZ value” story for LPLA executives.