04 · Product Roadmap
Learn → Practice → Invest → Grow, gated by evidence
Four phases over 18 months. Each phase ends at a go/no-go gate with engineering, product and compliance criteria, so the founding team never ships money movement on a hunch. The critical path to real money runs through partner contracting and SOC 2, not code, so those start in month 2.
18-month horizonPublic launch ~Month 4Real-money GA ~Month 9Gates, not dates
6
Weeks to build go/no-go
Discovery, counsel memo, paved road
~4
Months to education-only launch
Learn + Practice + guarded coach
M9
Earliest real-money GA
After 6 months of clean coach logs
≥98%
Advice-bait pass rate to ship
Eval gate on every prompt or model change
18-month view
Roadmap timeline
Phase 0 · Foundations · Weeks 0–6Phase 1 · Learn + Practice · to ~Month 4Phase 2 · Invest · Months 5–9Phase 3 · Grow · Months 10–18
Workstream
M0M3M6M9M12M15M18
Product
AI Coach
Content
Platform / Infra
Compliance & Security
Growth
Phase 1 · Learn + PracticeAI Coach · Month 1.5–4
Guarded coach GA
Classifier → retrieval over vetted lessons → model gateway → output checker → WORM audit. Golden set grows to 300–500 prompts and runs in CI on every prompt or model change.
Exit evidence: ≥98% pass on advice-bait; 100% of answers cite a lesson or decline.
Phase 0 · Weeks 0–6
Foundations & validation
Goals
- Confirm the #1 job-to-be-done and willingness to pay before building a full product.
- Lock the compliance posture in writing: education-only, impersonal, inside the publisher’s exclusion.[1][2]
- Stand up the paved road (CI, previews, database branching, audit bucket) so Phase 1 is pure product work.
In scope
- 15–20 customer interviews across the four personas, run on the existing prototype
- Fake-door pricing test: monthly vs annual, $6–12 band
- Securities counsel memo and the coach refusal policy
- Golden eval set v0 (~150 prompts) and a baseline score for the prototype coach
- Repo, CI, Neon branching, auth decision, Sentry, the WORM audit bucket
- Curriculum map, voice chart, reviewer panel contracted
Explicitly out of scope
- Any real-money or broker work beyond intro calls
- Native mobile
- Paid acquisition
- Hiring beyond one engineer
| Epic | Size | Notes |
|---|---|---|
| Discovery sprint + synthesis | M | Founders and CTO run the interviews; the prototype is the stimulus |
| Counsel memo + refusal policy | S | Counsel-led; the CTO turns the policy into testable rules |
| Paved road | M | Promote the prototype’s working OTP, coach and data code instead of rewriting it |
| Eval harness v0 | S | Promptfoo config in CI, scored on advice-bait and injection |
| Data model v1 + first 10 ADRs | S | See 05 Platform Architecture |
| Curriculum map + reviewer contracts | S | See 06 Content Direction |
Exit criteria / go–no-go gate
- Interview synthesis names one primary persona and job, with quotes, and the founding team agrees in writing.
- Fake-door results show which price and plan framing earns clicks.
- Counsel memo drafted with no blocking objections to the coach design.
- Build go/no-go held at week 6. A no-go means re-scoping the MVP, not abandoning the program.
KPIs we will report
- Interviews completed
- Waitlist sign-ups and source mix
- Fake-door click-through by price point
- Baseline coach eval scores
Team & budget drivers
- CTO (hands-on), founders on interviews, a part-time product designer, securities counsel on a fixed-scope memo.
- Budget drivers: counsel and design hours. Infrastructure stays on free tiers.
Phase 1 · Week 6 → ~Month 4
MVP: Learn + Practice, education-only launch
Goals
- Launch a product that teaches and lets people rehearse decisions with no real money involved.
- Prove activation and retention. Financial education works but decays within about 20 months, so the loop has to reinforce.[20][21]
- Ship the guarded coach with evidence (evals, audit trail) that it stays impersonal.
In scope
- Goal-based onboarding and level placement
- Lesson player: cards → quiz → practice action; spaced review queue
- ~24 reviewed lessons across the 4 tracks (11 exist in the prototype)
- Guarded coach GA: classifier, retrieval, checker, audit, AI disclosure
- Practice portfolio on simulated or delayed data, plus drawdown scenarios
- Goal planner; streaks and badges for learning only
- Stripe subscriptions with annual as the default
- Studio v1: CMS → review → publish → re-index
Explicitly out of scope
- Real money, broker links, account aggregation
- Leaderboards, confetti or rewards tied to trades (paper or real)
- Market-moving push notifications
- Teen or custodial accounts
- Native app (the PWA is enough)
| Epic | Size | Notes |
|---|---|---|
| Onboarding + placement | S | Placement quiz seeds the learning path |
| Lesson player + progress + spaced review | M | Attempt rows drive next_review_at |
| Guarded coach GA | L | The biggest epic; golden set grows to 300–500 prompts |
| Practice engine + drawdown scenarios | L | Simulate risk, not stock-picking[22] |
| Goal planner | S | Compounding and cost-of-waiting maths, with hypothetical labels |
| Billing + entitlements | S | Free tier coach cap; paid unlocks reasoning-tier tutoring |
| Studio v1 + CMS → RAG ingestion | M | LessonVersion rows, reviewer sign-off records |
| Audit archive + observability | S | Object Lock writer, Langfuse, PostHog, Sentry with PII scrubbing |
| SOC 2 readiness | M | Drata or Vanta from month 2; runs in the background |
Exit criteria / go–no-go gate
- Signed counsel memo on the education-only posture.
- Coach passes ≥98% of advice-bait prompts, and 100% of eval answers cite a lesson or decline.
- AI disclosure on the first turn and a disclaimer at the point of delivery (EU Article 50 is already in force).[13]
- 18+ gate at sign-up, which keeps amended COPPA out of scope.[12]
- Privacy policy (GDPR/CCPA) and a GLBA-style written infosec program.[11]
- WORM archive receiving every coach turn.[6]
- No real-money flows anywhere in the product.
KPIs we will report
- Activation: first lesson and quiz finished in session 1
- Week-4 retention (finance-app D30 medians are only ~2–7%; strong apps reach 10–15%)[24]
- 7-day learning streak rate (Duolingo learners with a 7-day streak are 3.6× likelier to finish a course)[25]
- Coach: thumbs-up rate, false-refusal rate, cost per conversation
- Trial-to-paid (median 45.7% for 17–32 day trials) and annual-plan mix[23]
Team & budget drivers
- CTO + one senior full-stack engineer, a part-time designer, a content lead, contracted CFP®/CFA® reviewers, fractional compliance support.
- Run cost ~$2–3k/month at 1k MAU, including amortized compliance tooling.[18] Drata or Vanta about $7.5–15k/year.[10]
- Variable drivers: reviewer fees per lesson, coach token spend per active user.
Phase 2 · Months 5–9
Invest: self-directed through a broker-dealer partner
Goals
- Let graduates open and fund a real, self-directed account in the app through a registered partner.
- Keep the coach strictly explanatory: it can explain an order ticket but never suggest one.
- Make the first real investment a boring, diversified, recurring one.
In scope
- Partner agreement (Alpaca by default; its sandbox and paper trading are free)[4]
- Partner-hosted KYC/CIP; we store partner account IDs only[3]
- Bank funding via Plaid processor tokens
- Order ticket with an explain mode; recurring auto-invest into ETFs
- Reconciliation and an ops console for support
- Communications-review workflow with the partner (FINRA 2210, RN 26-14)[5]
Explicitly out of scope
- Options, margin, crypto, copy-trading
- Any AI trade suggestion, even “popular with learners like you”
- Confetti, rewards or prize draws tied to trading[7][8]
| Epic | Size | Notes |
|---|---|---|
| BrokerPort + Alpaca adapter | L | Idempotent order intents, webhook ingestion |
| Account opening + KYC hand-off | M | Partner-hosted flow; status mirrored in BrokerAccount |
| Funding via Plaid | M | Processor tokens; step-up auth on bank changes |
| Order ticket + explain mode | M | Coach reads the ticket’s public facts, never the user’s holdings |
| Recurring auto-invest | M | The behavioural default that matters most |
| Reconciliation + ops console | M | Daily diff against partner statements |
| Partner comms-review workflow | S | Studio routes promotional content to partner reviewers |
| Graduation moments | S | Finished a track, lived through a simulated drawdown → eligible to open an account |
Exit criteria / go–no-go gate
- Signed partner agreement and a partner-approved KYC flow.
- Communications-review workflow agreed with the partner, with an SLA.
- The coach has no order-capable tools, verified by code review and a test.
- SOC 2 Type I report in hand and a Type II window running.[9]
- Market-data redistribution licence signed for anything shown to users.
- Six months of clean eval and incident logs since the closed beta (month 3 → month 9).
KPIs we will report
- Practice → funded conversion, by cohort
- Time from sign-up to first funded account
- Share of funded accounts with a recurring deposit
- Reconciliation breaks (target: zero) and support tickets per 100 accounts
- Coach refusal accuracy on invest-context prompts
Team & budget drivers
- Add a backend / integrations engineer, part-time QA, an operations and support lead, and a compliance consultant who owns the partner relationship.
- Budget drivers: SOC 2 Type II audit (~$12–20k)[9], a market-data display licence (quote), any partner minimums (to be negotiated).
Phase 3 · Months 10–18
Grow: personalization, family / B2B2C, and the advice path
Goals
- Show measurable learning gains, not just engagement.
- Open a lower-CAC channel. Zogo shows that 240+ financial institutions will license financial education.[27]
- Decide, with evidence, whether to pursue personalized advice and by which route.
In scope
- Mastery model and Socratic tutor mode
- Family / Trump Account track
- Credit-union and employer white-label pilot (tenancy, admin, reporting)
- Expo mobile app on the same API
- SnapTrade read-only “explain my portfolio,” kept impersonal and reviewed by counsel
- Advice-path analysis: RIA, sub-advisory, or UK targeted support
Explicitly out of scope
- Registering an RIA before the decision memo is signed
- Teen accounts without a custodial partner
- EU launch before the Annex III review
| Epic | Size | Notes |
|---|---|---|
| Mastery model + tutor mode | M | Per-concept mastery, withheld answers, pre/post measurement |
| Family / Trump Account track | S | New demand since July 2026[28] |
| B2B2C tenancy + admin | L | Tenant isolation tests in CI; seat-based billing |
| Mobile app | L | Expo, sharing types and API with the web app |
| SnapTrade integration | M | Read-only, behind its own ADR and counsel check |
| Advice-path analysis | M | Counsel-led; CTO scopes the systems impact |
Exit criteria / go–no-go gate
- RIA route: SEC registration (the internet adviser exemption now requires advising all clients exclusively online)[16], a CCO, Form ADV and Rule 204-2 recordkeeping.
- Investment-adviser AML program ready for 2028-01-01.[15]
- UK: targeted-support permission if that route is chosen.[17]
- EU: Annex III review completed before 2027-12-02 if the coach grades or gates learners.[14]
KPIs we will report
- Learning gain: pre/post quiz delta, coach vs non-coach cohorts
- D90 retention and annual-plan renewal
- B2B2C: signed pilots, active seats, admin-reported outcomes
- LTV:CAC of at least 3:1 by channel
Team & budget drivers
- Add a mobile engineer, an ML / evaluation engineer, a partnerships lead, and a CCO if the RIA route is chosen.
- Run cost ~$5–7k/month at 10k MAU and ~$30–40k at 100k, dominated by LLM, compliance and data licences.[18][19]
Board
Now / Next / Later
RICE score orders work inside each column. The columns themselves are set by phase gates: a high score never pulls a Phase 2 item ahead of its compliance gate.
Now
Phase 0–1 · building
- Lesson player + progressRICE 3,000
- Onboarding + level placementRICE 2,000
- Guarded AI coach (GA)RICE 1,600
- Learning streaks + mastery badgesRICE 1,600
- Weekly newsletter + lesson digestRICE 1,600
- Goal plannerRICE 1,400
- Stripe billing, annual defaultRICE 1,350
- Spaced-repetition review queueRICE 1,200
- Drawdown simulator (“live through −30%”)RICE 1,167
- Practice portfolio (paper trading)RICE 960
- Studio pipeline (internal)RICE 875
Next
Phase 2 · sandbox from M5
- Bank funding via PlaidRICE 225
- Recurring auto-investRICE 150
- Live cohortsRICE 150
- Brokerage account openingRICE 135
- Order ticket with explain modeRICE 120
Later
Phase 3 · after the M9 gate
- Native mobile app (Expo)RICE 180
- Family / Trump Account trackRICE 160
- B2B2C admin + tenancyRICE 160
- SnapTrade “explain my portfolio”RICE 133
Backlog
RICE-prioritized backlog
Reach is users touched per quarter, assuming about 5k MAU after launch. Impact uses the standard 0.25–3 scale. Confidence is a percentage. Effort is engineer-weeks. These are planning estimates for discussion with the founding team, not measured data.
| # | Feature | Reach | Impact | Conf. | Effort | RICE | Tier | Why |
|---|---|---|---|---|---|---|---|---|
| 1 | Lesson player + progress | 5,000 | 2 | 90% | 3 | 3,000 | Now | The core loop; already prototyped |
| 2 | Onboarding + level placement | 5,000 | 1 | 80% | 2 | 2,000 | Now | Every user passes through it |
| 3 | Guarded AI coach (GA) | 4,000 | 3 | 80% | 6 | 1,600 | Now | The differentiator; gated by evals |
| 4 | Learning streaks + mastery badges | 4,000 | 0.5 | 80% | 1 | 1,600 | Now | Learning only, never trading |
| 5 | Weekly newsletter + lesson digest | 4,000 | 0.5 | 80% | 1 | 1,600 | Now | Retention and acquisition |
| 6 | Goal planner | 3,500 | 1 | 80% | 2 | 1,400 | Now | Ties lessons to a real decision |
| 7 | Stripe billing, annual default | 1,500 | 2 | 90% | 2 | 1,350 | Now | Revenue from day one |
| 8 | Spaced-repetition review queue | 3,000 | 1 | 80% | 2 | 1,200 | Now | Counters knowledge decay |
| 9 | Drawdown simulator (“live through −30%”) | 2,500 | 2 | 70% | 3 | 1,167 | Now | Risk is the weakest literacy area |
| 10 | Practice portfolio (paper trading) | 3,000 | 2 | 80% | 5 | 960 | Now | Rehearsal before real money |
| 11 | Studio pipeline (internal) | 5,000 | 1 | 70% | 4 | 875 | Now | Content velocity with sign-off records |
| 12 | Bank funding via Plaid | 500 | 3 | 60% | 4 | 225 | Next | Phase 2 gate item |
| 13 | Native mobile app (Expo) | 3,000 | 1 | 60% | 10 | 180 | Later | PWA covers MVP |
| 14 | Family / Trump Account track | 800 | 1 | 60% | 3 | 160 | Later | New 2026 demand |
| 15 | B2B2C admin + tenancy | 2,000 | 2 | 40% | 10 | 160 | Later | CAC hedge; needs a pilot first |
| 16 | Recurring auto-invest | 400 | 3 | 50% | 4 | 150 | Next | The highest-value behaviour, once live |
| 17 | Live cohorts | 300 | 2 | 50% | 2 | 150 | Next | Premium tier test |
| 18 | Brokerage account opening | 600 | 3 | 60% | 8 | 135 | Next | Phase 2 critical path |
| 19 | SnapTrade “explain my portfolio” | 800 | 1 | 50% | 3 | 133 | Later | Needs its own counsel check |
| 20 | Order ticket with explain mode | 500 | 2 | 60% | 5 | 120 | Next | Coach explains, never suggests |
Dependencies
Critical path and dependency notes
Path A: education launch (engineering-bound)
Counsel memo → refusal policy → golden eval set → coach GA (≥98%) → closed beta (M3) → public launch (M4).
The coach is the longest pole. Content capacity runs in parallel but caps how many lessons launch.
Path B: real money (contract-bound)
Partner intro (M1) → LOI (M2–4) → SOC 2 Type I (M6) → partner due diligence → agreement → sandbox certification → real-money GA (M9).
None of this is code. It needs an owner from month 1, or month 9 becomes month 12.
| Dependency | Blocks | Lead time | Mitigation |
|---|---|---|---|
| Securities counsel memo | Coach GA, launch marketing | 3–6 weeks | Engage in week 1 with a written coach spec and sample transcripts |
| Reviewer panel (CFP®/CFA®) | Lesson count at launch | 2–4 weeks to contract | Contract two reviewers, so one absence does not stall publishing |
| SOC 2 Type I | Partner due diligence | ~4 months from tooling start[9] | Start Drata or Vanta in month 2, not after the LOI |
| Broker partner agreement | All Phase 2 GA | Unknown; partner-driven | Run Alpaca and one alternate in parallel until LOI |
| Market-data licence | Showing real quotes | Quote-based | Simulated or delayed data until the licence is signed |
| Partner comms-review SLA | Any promotional content | Negotiated in the agreement | Studio routes content automatically; RN 26-14 may move toward risk-based review[5] |
Risks
Risks to the timeline
| Risk | Likelihood | Impact | Early signal | Response |
|---|---|---|---|---|
| Partner contracting slips | High | Real-money GA moves past M9 | No LOI by month 4 | Keep invest behind a flag and pull Phase 3 learning work forward |
| Counsel narrows coach scope | Medium | Fewer coach features at launch | Memo flags goal-aware answers | Ship the impersonal coach; add context only through ADR plus counsel review |
| Coach misses the 98% gate | Medium | Beta slips 2–4 weeks | Advice-bait bucket below 95% at M2 | Tighten the classifier, add deterministic checks, move hard cases to scripted answers |
| Low early retention | High | Weak case for Phase 2 and fundraising | W4 retention near the finance-app median[24] | Tighten the loop to a real decision and lean on spaced review and the newsletter |
| Reviewer bottleneck | Medium | Thin library at launch | Review queue older than 5 days | A second reviewer; AI pre-checks before SME time |
| LLM cost overrun | Low | Margin pressure | Cost per conversation above $0.06 | Shift routing toward the fast tier, cache more, cap free-tier usage |
| Founder scope creep | Medium | Everything slips | New epics added mid-phase | One change-control rule: new scope enters at the next gate |
| Hiring the second engineer | Medium | Phase 1 runs 4–6 weeks longer | No offer accepted by week 4 | CTO carries more of the build; content work keeps moving |
Sources
- [1]Lowe v. SEC, 472 U.S. 181 (1985)
- [2]Gordon Rees: compliance for financial publishers and trading educators (Jan 2026)
- [3]Alpaca Broker API docs (KYC/CIP responsibilities)
- [4]Open Banking Tracker: Alpaca (sandbox, paper trading)
- [5]FINRA Regulatory Notice 26-14 (Jul 2026)
- [6]SEC 17a-4 and AI recordkeeping
- [7]Robinhood $7.5M Massachusetts gamification settlement
- [8]FCA: trading apps and digital engagement practices
- [9]SOC 2 audit cost
- [10]SOC 2 software pricing (Drata, Vanta)
- [11]GLBA Safeguards Rule requirements
- [12]COPPA amendment compliance deadline (2026-04-22)
- [13]EU AI Act Article 50 transparency obligations (2 Aug 2026)
- [14]Digital Omnibus: Annex III obligations moved to 2027-12-02
- [15]FinCEN investment-adviser AML rule postponed to 2028-01-01
- [16]SEC amends the internet adviser exemption (2024)
- [17]FCA Advice Guidance Boundary Review: targeted support
- [18]Reference stack pricing (Vercel)
- [19]Anthropic API list pricing
- [20]Kaiser, Lusardi, Menkhoff & Urban (2022), JFE meta-analysis of 76 RCTs
- [21]Fernandes, Lynch & Netemeyer (2014), Management Science
- [22]Risk simulation improves investment decisions (Zeisberger)
- [23]RevenueCat State of Subscription Apps 2025
- [24]Business of Apps: finance app retention benchmarks
- [25]Duolingo streak and retention data
- [26]FINRA Foundation NFCS investor study (Dec 2025)
- [27]Zogo: 240+ financial institutions
- [28]Trump Accounts launch (Jul 2026)