Version: 2.0 Date: 2026-02-18 Owner: BizOps Status: Accepted (Token Bank model approved per DR-2026-004) Product: Legionis Related: DR-2026-001 (Pricing Model), DR-2026-004 (Token Bank Pricing), pricing-strategy.md, Legionis Agent Catalog
Legionis currently operates a BYOK (Bring Your Own Key) model where users provide their own Anthropic API keys. The platform charges $10-25/mo for team-based modules. This analysis explores adding a managed billing option where Legionis handles Claude API calls and bills users with a markup on token costs.
The core tension: BYOK gives Legionis near-zero COGS and clean margins. Managed billing introduces variable COGS, payment processing drag, and token cost risk, but removes a significant friction point for users who don't want to manage API keys.
Recommendation (v2.0): Offer managed billing via Prepaid Token Banks at 15% markup, replacing the previously proposed 30% metered model. Users purchase dollar-denominated credit packs ($10/$25/$50/$100) with tiered bonuses at higher amounts. No expiration at launch. Optional auto-top-up. Per-task cost preview before execution. BYOK remains the hero path: "Bring your own keys for zero markup, or use our credits for instant setup."
Why the change from 30% metered to 15% prepaid: Prepaid Token Banks eliminate credit risk (cash collected before service rendered), reduce Stripe fees (one charge per top-up vs. recurring metered charges), deliver better UX (no surprise bills), and are competitively differentiated at 15% vs. industry-standard 20-40% markups. See DR-2026-004 for full decision record.
| Dimension | Value |
|---|---|
| COGS per user | ~$0 (user pays Anthropic directly) |
| Platform fee | $10-25/mo (100% gross margin minus infra) |
| User effort | Must create Anthropic account, get API key, manage billing, monitor usage |
| User profile | Technical, comfortable with APIs, cost-conscious power users |
| Risk to Legionis | Zero token cost risk |
| Dimension | Value |
|---|---|
| COGS per user | Variable (Anthropic API costs + Stripe fees) |
| Platform fee | $10-25/mo + token markup revenue |
| User effort | Zero API key management. Just use the product. |
| User profile | Non-technical PMs, teams with procurement constraints, "just make it work" users |
| Risk to Legionis | Token cost pass-through, heavy user exposure, Stripe fees on gross amount |
With BYOK, Legionis is a pure SaaS platform: predictable costs, clean margins. The moment you introduce managed billing, you become a reseller of AI compute on top of being a SaaS platform. These are fundamentally different business models running in parallel.
The good news: they can coexist. The key is ensuring the managed billing option is self-sustaining (covers its own costs + contributes margin) without subsidizing it from BYOK margins.
For managed billing to work, the markup must cover:
Markup Revenue = Anthropic API Cost x Markup %Must cover:
Anthropic API cost (pass-through) = 100% of base
Stripe processing fee on token charges = ~2.9% + $0.30 per invoice
Billing system overhead (metering, invoicing) = [TBD - engineering cost]
Bad debt / failed payments = ~1-2% of billings
Margin contribution = the actual profit
Structure: User pays Anthropic rate x (1 + markup%)
| Markup % | Anthropic Cost | User Pays | Legionis Keeps | After Stripe (2.9%) | Net Margin |
|---|---|---|---|---|---|
| 15% | $10.00 | $11.50 | $1.50 | $1.17 | 10.1% |
| 20% | $10.00 | $12.00 | $2.00 | $1.65 | 13.8% |
| 30% | $10.00 | $13.00 | $3.00 | $2.62 | 20.2% |
| 40% | $10.00 | $14.00 | $4.00 | $3.59 | 25.7% |
| 50% | $10.00 | $15.00 | $5.00 | $4.57 | 30.4% |
v2.0 update: 15% markup selected over 30% when combined with prepaid Token Banks (see new Section 5A below). Prepaid model eliminates the credit risk and Stripe fee drag that justified higher markup in the metered model.
The "switch threshold" remains healthy at 15%: At what spend level does a user think "I should just get my own API key"?
| Monthly Token Spend | Markup % | Rationale |
|---|---|---|
| $0-25 | 35% | Small amounts, high convenience value |
| $25-75 | 25% | Moderate users, still convenient |
| $75+ | 15% | Heavy users, keep them from switching to BYOK |
Pros: Retains heavy users on managed billing longer. Cons: Complex billing, harder to communicate, users may game tiers. Not recommended for MVP.
Structure: Flat fee per operation (e.g., $0.02 per skill invocation) regardless of actual token cost.
Problem: Operations vary wildly in token cost. A /context-recall might use 1K tokens ($0.003) while a /qbr-deck with Opus might use 100K tokens ($7.50+). A flat per-op fee either overcharges simple operations or undercharges complex ones.
Verdict: Only works if combined with model routing and tight operation cost estimates. Too complex for now.
Replaces the v1.0 recommendation of 30% flat metered markup. The shift to prepaid Token Banks at 15% delivers better unit economics because:
See Section 5A for full Token Bank mechanics.
Should managed billing users pay a different platform fee than BYOK users?
| User Type | Platform Fee | Token Cost |
|---|---|---|
| BYOK | $10-25/mo | Pays Anthropic directly |
| Managed | $10-25/mo | Pays Legionis (Anthropic rate + 30%) |
Why same fee works:
| User Type | Platform Fee | Token Cost |
|---|---|---|
| BYOK | $10-25/mo | Pays Anthropic directly |
| Managed | $5-15/mo (lower) | Pays Legionis (Anthropic rate + 30%) |
Rationale: The token markup generates revenue, so you can afford a lower platform fee to attract more managed users.
Problem: This incentivizes users to choose managed even if they'd be better off with BYOK. You're acquiring token cost risk for a platform fee discount. Bad trade.
| User Type | Platform Fee | Token Cost |
|---|---|---|
| BYOK | $10-25/mo | Pays Anthropic directly |
| Managed | $15-30/mo (higher) | Pays Legionis (Anthropic rate + 30%) |
Rationale: Managed billing is a premium feature that requires infrastructure (metering, billing, support).
Problem: Double-dipping. Users already pay a markup on tokens. Adding a platform premium feels punitive. Creates resentment.
The token markup is the managed billing business model. The platform fee is the SaaS business model. Keep them clean and separate.
v2.0 Note: The minimum monthly spend concept from v1.0 is superseded by the Token Bank prepaid model (Section 5A). With prepaid credit packs, the Stripe micro-transaction problem is solved by design: users buy credits in $10+ increments, and Stripe's fixed fee is amortized over larger, less frequent transactions.
The v1.0 analysis below is retained for reference but is no longer the active model.
v1.0 Minimum Spend Analysis (historical reference)
Without a minimum, managed billing users could:
Stripe charges ~2.9% + $0.30 per transaction. On small token invoices:
| Token Spend | + 30% Markup | Stripe Fee | Net to Legionis | Effective Margin |
|---|---|---|---|---|
| $1.00 | $1.30 | $0.34 | -$0.04 | Negative |
| $2.00 | $2.60 | $0.38 | $0.22 | 8.5% |
| $3.00 | $3.90 | $0.41 | $0.49 | 12.6% |
| $5.00 | $6.50 | $0.49 | $1.01 | 15.5% |
| $10.00 | $13.00 | $0.68 | $2.32 | 17.9% |
| $20.00 | $26.00 | $1.05 | $4.95 | 19.0% |
Users purchase dollar-denominated credit packs upfront. Credits are real dollars: $1 credit = $1 of Anthropic API cost at current rates. No fictional "tokens," "coins," or "gems."
| Credit Pack | Price (with 15% markup) | Bonus Credits | Total Available | Effective Markup |
|---|---|---|---|---|
| $10 | $11.50 | None | $10.00 | 15.0% |
| $25 | $28.75 | None | $25.00 | 15.0% |
| $50 | $57.50 | $5.00 bonus | $55.00 | ~4.5% effective |
| $100 | $115.00 | $15.00 bonus | $115.00 | ~0% effective |
Tiered bonuses incentivize larger purchases. The $100 pack at ~0% effective markup is intentional: large prepaid amounts improve cash flow, reduce transaction frequency, and these users are likely to become long-term retained users.
| Model | Transaction Frequency | Avg Transaction Size | Stripe Fee % of Revenue |
|---|---|---|---|
| 30% Metered (v1.0) | Monthly or daily | $5-30 | 3.5-8% (high, esp. on small amounts) |
| 15% Token Banks (v2.0) | Per top-up (infrequent) | $11.50-115.00 | 2.9-3.2% (low, large transactions) |
Stripe fee on each credit pack:
| Pack Price | Stripe Fee (2.9% + $0.30) | Net to Legionis | Effective Margin on Markup |
|---|---|---|---|
| $11.50 | $0.63 | $0.87 | 7.6% |
| $28.75 | $1.13 | $2.62 | 9.1% |
| $57.50 | $1.97 | $5.53 | 9.6% |
| $115.00 | $3.64 | $11.36 (before bonus cost) | 9.9% |
Even the smallest pack ($10) is comfortably profitable. No micro-transaction problem.
Optional, user-configured:
Before executing any operation, the UI shows estimated cost:
📝 PRD: Authentication Feature
Estimated cost: ~$0.12 (Sonnet 4.5)
Your balance: $8.43[Execute] [Cancel]
This transparency is a key differentiator. Users always know what they're spending before they spend it.
At launch: Purchased credits never expire. This is a strong trust signal and removes the "use it or lose it" anxiety that competitors impose.
Revisit after 12 months: If credit breakage (purchased but unused credits) exceeds 15% of total purchases, evaluate whether expiration should be introduced (e.g., 24-month expiry). This is a re-decision trigger in DR-2026-004.
| Dimension | 30% Metered (v1.0) | 15% Token Banks (v2.0) | Winner |
|---|---|---|---|
| Credit risk | Legionis fronts API costs, bills later | Prepaid: cash before service | Token Banks |
| Stripe fees | Recurring charges, micro-txn problem | Infrequent, larger charges | Token Banks |
| User experience | Surprise bills, usage anxiety | Predictable, user-controlled | Token Banks |
| Markup perception | "30% fee" feels expensive | "15% with bonuses" feels fair | Token Banks |
| Cash flow | Delayed (bill after usage) | Immediate (prepaid) | Token Banks |
| BYOT switching pressure | High (30% = noticeable) | Lower (15% = tolerable longer) | Token Banks |
| Complexity | Metering + invoicing infrastructure | Credit balance tracking | Similar |
| Trial Type | BYOK | Managed Billing |
|---|---|---|
| Who pays for tokens? | User pays Anthropic | Legionis pays Anthropic |
| Trial cost to Legionis | $0 | $[actual token usage] per trial user |
| User friction | Must set up Anthropic account during trial | Zero friction |
| Conversion signal | Already invested (has API key) | No investment signal |
This is the biggest economic difference between BYOK and managed billing. A 30-day free trial with managed billing means Legionis eats all token costs for trial users.
Assuming a trial user does 5-15 operations per day for 30 days:
| Usage Level | Ops/Day | Monthly Ops | Avg Token Cost/Op | Monthly Token Cost |
|---|---|---|---|---|
| Light trial | 5 | 150 | $0.01 (mostly Haiku) | ~$1.50 |
| Moderate trial | 10 | 300 | $0.03 (Haiku + some Sonnet) | ~$9.00 |
| Heavy trial | 15 | 450 | $0.05 (mixed models) | ~$22.50 |
| Power trial | 30+ | 900+ | $0.05+ | ~$45.00+ |
Note: These token cost estimates are illustrative of the framework. Actual costs depend on specific operations and their token consumption, which needs to be measured in production.
| Element | Value |
|---|---|
| Duration | 14 days (shorter than BYOK trial) |
| Token cap | $5 worth of tokens (Legionis eats the cost) |
| Model restriction | Haiku only during trial (lowest COGS model) |
| After cap | Must add payment method or switch to BYOK to continue |
Max cost to Legionis: $5 per trial user. If 20% convert, CAC = $25 on tokens alone.
| User Type | Trial Duration | Token Cost During Trial | Model Access |
|---|---|---|---|
| BYOK | 30 days | User pays Anthropic (no cost to Legionis) | All models (user's API key) |
| Managed | 14 days | Legionis covers up to $5 in tokens | Haiku only |
| Managed (CC on file) | 30 days | Legionis covers up to $10 in tokens | Haiku + Sonnet |
This creates a natural funnel:
| Metric | BYOK User | Managed User ($20 token spend) | Managed User ($50 token spend) |
|---|---|---|---|
| Platform fee | $10-25 | $10-25 | $10-25 |
| Token revenue | $0 | $23.00 (= $20 x 1.15) | $57.50 (= $50 x 1.15) |
| Token COGS | $0 | -$20.00 | -$50.00 |
| Stripe on platform | -$0.59 to -$1.03 | -$0.59 to -$1.03 | -$0.59 to -$1.03 |
| Stripe on tokens | $0 | -$0.97 (single top-up charge) | -$1.97 (single top-up charge) |
| Infra cost | ~-$1.00 | ~-$1.00 (simpler than metering) | ~-$1.00 |
| Net margin | $8.41-23.97 | $10.44-26.00 | $3.94-19.50 |
| Gross margin % | ~84-96% | ~32-79% | ~5-26% (on total) |
Key insight (v2.0): At 15% markup, per-user margin is lower than 30%, but this is offset by: (1) lower Stripe fees from prepaid top-ups, (2) zero credit risk (prepaid), (3) higher user retention on managed billing (lower switching pressure), and (4) better cash flow (prepaid). The net effect on blended portfolio economics is positive because more users stay on managed billing longer.
Comparison to v1.0 (30% metered):
| Scenario | BYOK % | Managed % | Blended Gross Margin |
|---|---|---|---|
| Mostly BYOK | 80% | 20% | High (~75-85%) |
| Even split | 50% | 50% | Moderate (~55-65%) |
| Mostly managed | 20% | 80% | Lower (~40-50%) |
The healthy mix: You WANT a majority of heavy users on BYOK (zero COGS) and convenience-seeking moderate users on managed. The markup pricing naturally creates this self-selection.
For context, here's what Anthropic's current rates mean in practice:
| Operation | Model | ~Input Tokens | ~Output Tokens | ~Cost |
|---|---|---|---|---|
| Context recall | Haiku 3.5 | 5,000 | 1,000 | ~$0.008 |
| User story | Haiku 3.5 | 3,000 | 2,000 | ~$0.010 |
| PRD (standard) | Sonnet 4.5 | 10,000 | 5,000 | ~$0.105 |
| Agent spawn (@pm) | Sonnet 4.5 | 15,000 | 8,000 | ~$0.165 |
| Strategic bet | Opus 4 | 20,000 | 10,000 | ~$1.050 |
| PLT gateway | Opus 4 | 30,000 | 15,000 | ~$1.575 |
Note: These are rough estimates. Actual token counts will vary significantly by operation complexity, context size, and prompt caching effectiveness. These should be validated with production measurements.
| User Type | Daily Pattern | Monthly Token Cost (est.) | With 15% Markup (v2.0) | With 30% Markup (v1.0) |
|---|---|---|---|---|
| Light PM | 5 Haiku ops/day | ~$1.50 | ~$1.73 | ~$1.95 |
| Regular PM | 10 mixed ops/day | ~$10-20 | ~$11.50-23 | ~$13-26 |
| Power PM | 20 mixed ops/day | ~$30-60 | ~$34.50-69 | ~$39-78 |
| Strategic leader | 10 heavy ops/day (Sonnet/Opus) | ~$50-100+ | ~$57.50-115+ | ~$65-130+ |
At 15% markup (Token Banks), a regular PM pays roughly $11.50-23/mo in tokens on top of their $10-25 platform fee. Total: $21.50-48/mo. This is $1.50-3/mo cheaper than the 30% model, which reduces BYOT switching pressure and improves retention on managed billing.
Scenario: A managed user runs 50 Opus-heavy PLT sessions in a month = potentially $50-80 in token costs. Mitigation (v2.0): With Token Banks, users can only spend what they've prepaid. No credit risk. If balance hits $0, operations pause until top-up. The 15% markup covers operational overhead on all prepaid amounts. Protection: At $100+ monthly spend, surface BYOT suggestion: "You're spending $100+/mo on credits. Switch to BYOT for zero markup and save $15+/mo." This is good customer service AND good business.
Scenario: Anthropic increases API prices by 30%. Mitigation (v2.0): Token Bank credits are dollar-denominated against current Anthropic rates. If Anthropic raises prices, existing credits buy fewer operations (the credit represents dollars, not token counts). New credit purchases reflect current rates. Protection: Terms of service should state: "Credits represent dollar value applied at current Anthropic API rates plus a 15% platform fee. Underlying API rates may change."
Scenario: Anthropic drops Haiku to $0.40/MTok input. Impact: Your 30% markup now generates less absolute revenue per operation. But users are happier (cheaper) and may use more. Action: No action needed. Lower costs = more usage = more operations = more total markup revenue.
Scenario: Someone uses Legionis managed billing as a cheap API proxy. Mitigation: Rate limiting, fair use policy, monitoring for non-product-org usage patterns. Your agent/skill architecture naturally limits this -- you can't just send raw prompts.
Scenario (v1.0): User's monthly token charge is $2.00. Stripe takes $0.36 (18%). Mitigation (v2.0): Eliminated by Token Bank design. Minimum credit pack is $10 ($11.50 with markup). Stripe fee on $11.50 = $0.63 (5.5%). The prepaid model solves the micro-transaction problem by design.
| Component | Complexity | Notes |
|---|---|---|
| Token metering | Medium | Track tokens per user per operation. Vercel AI SDK may provide hooks. |
| Usage dashboard | Medium | Show users their token consumption in real-time |
| Billing aggregation | Medium | Roll up daily token usage into monthly Stripe invoices |
| Managed API key pool | Low | Legionis uses its own Anthropic API key for managed users |
| BYOK/Managed toggle | Low | User setting to choose billing mode |
| Spend alerts | Low | Notify at 50%, 80%, 100% of projected spend |
| Rate display | Low | Show current per-model rates on pricing page |
| Question | v2.0 Recommendation | v1.0 (superseded) | Rationale for Change |
|---|---|---|---|
| Markup model? | Prepaid Token Banks at 15% | 30% flat metered | Eliminates credit risk, reduces Stripe fees, better UX |
| Credit structure? | $10/$25/$50/$100 packs with bonuses at $50+ | N/A (metered) | Encourages larger purchases, improves cash flow |
| Expiration? | No expiration at launch (revisit at 12 months) | N/A | Trust signal, competitive differentiator |
| Auto-top-up? | Optional, user-configured threshold | N/A | Convenience without forcing spend |
| Cost transparency? | Per-task estimated cost before execution | N/A | Users approve before spending |
| Different platform fee? | No -- same fee for both | Same | Token markup IS the managed revenue. Don't double-dip. |
| Free trial for managed? | Asymmetric: shorter, capped, Haiku-only | Same | Limit trial cost exposure |
| BYOT positioning? | Hero path: "zero markup, or use credits for instant setup" | Neutral | BYOT is the brand promise; Token Banks are the on-ramp |
| Cap on managed spend? | No cap, but suggest BYOT at $100+/mo | Same | Natural BYOT funnel for power users |
Revisit this analysis if:
Version: 2.0 Created: 2026-02-14 Last Updated: 2026-02-18 Owner: BizOps Next Review: After managed billing launch (90-day data review) Distribution: Founder, Product Leadership
Related Documents:
decisions/DR-2026-004-token-bank-pricing.md (Token Bank pricing decision)decisions/DR-2026-001-pricing-model.md (original pricing model pivot)pricing-strategy.md (comprehensive pricing analysis, v2.0)pricing-usage-based-model.md (operation-based pricing)End of Document