How We Priced the Stack: 5 Tiers, Real Margin Math, No Demand-Curve Theater

We rebuilt /pricing on a five-tier table earlier this month. The shape: Free $0, Starter $29, Growth $99, Scale $249, Enterprise from $999. We also killed the old Pro $19 SKU. This post is the long version of how we got there: the cost-of-goods math, the demand-curve thinking we explicitly rejected, and the things we still don't know.

What we sell, in revenue-recognition terms#

The platform has four metered surfaces today: email send/receive, SMS send/receive, voice minutes (Basic and Premium tiers), and per-phone-number rental. Vault credentials and audit retention are quota-based but not metered.

Each tier prices the same product. Tiers differ in (a) what's included before metered overage kicks in, (b) which features are gated behind the tier (audit logs, SSO, dedicated support), and (c) the per-unit overage rate.

That's the surface. The interesting decisions are below it.

Cost-of-goods, line by line#

We work backward from cost. For voice the math is unforgiving because it's the most expensive thing per unit. For everything else it's about quota economics.

Voice (the hard one)#

Basic voice runs through Telnyx Kokoro for synthesis and Telnyx PSTN for the dial. Promo rate $0.07/min, steady-state $0.08/min from 2026-10-01.

Per-minute Telnyx PSTN list: ~$0.0085 outbound for US 48-state. ElevenLabs Kokoro hosted by Telnyx: ~$0.025/min synthesized. Speech-to-text via Deepgram Nova: ~$0.0073/min. Add ~$0.005/min for our LLM context window on the orchestrator. Sum: ~$0.046/min variable cost.

At $0.07/min promo: 33% gross margin per minute. At $0.08/min steady-state: 42% gross margin.

Premium voice runs through ElevenLabs (better voices). Promo $0.15/min, steady-state $0.18/min. ElevenLabs cost ~$0.10/min. Same Deepgram + LLM overhead. Sum: ~$0.12/min variable cost. Gross margin 20% at promo, 33% at steady-state.

Why we picked these numbers: voice is the loss-leader nobody else has clean compliance on. We accepted a thinner margin per minute in exchange for being the best-priced voice in the agent infrastructure space at the moment we ship the launch. Once the TCPA-class-action narrative spreads (and it will), the steady-state $0.08/$0.18 rates have room to move up; until then, we eat the smaller margin.

Email#

Outbound goes through AWS SES at ~$0.10 per 1,000 sends. Inbound parsing through our own infra. Storage on RDS + S3. All-in cost per 1,000 outbound: ~$0.30 once you load DKIM + bounce processing + storage.

We charge $1/1,000 outbound on overage past tier quota. 70% gross margin on overage; included quota is essentially loss-leader since it's covered by the platform fee.

Phone numbers#

Twilio US-local list: $1/mo. Anima list: $2/mo. Margin pays for the gates (RND check cache, time-of-day enforcement, area-code → timezone resolver).

Vault#

Per-credential cost is dominated by encryption-envelope storage (~$0.0001/cred/mo at S3 + KMS rates). The 25,000-credential cap on Scale ($249/mo) is sized so that a customer pegging the cap costs us under $3/mo to host — the platform fee covers it 80x over.

MCP#

Per-call cost is essentially zero (it's just routing). The 100 req/sec/org soft cap exists for shared-infra protection, not cost recovery.

Why we killed Pro $19#

The old Pro $19 SKU sat between Free and the original $49 (now Starter $29). It existed because we copied a Stripe-startup pattern without thinking about the actual value gradient.

The honest read: Pro $19 was indistinguishable from Free + $20 of overage in practice. Customers who hit the Free quota and converted to Pro got... slightly more email quota and one more phone number. The features didn't justify a separate SKU. They needed to either commit to Starter (real platform) or stay on Free (real exploration). Pro was middle-ground that confused buyers and complicated the SKU page.

Killed 2026-04-25. Zero customers to migrate (we checked the audit script). The LEGACY_PRICE_IDS env var maps any straggling webhooks to Starter so we don't crash on a stale subscription event — but in practice none arrived.

What we explicitly rejected#

A few patterns we considered and walked away from.

Demand-curve theater. "Set the price 30% above intuition because the customer's perceived value is higher than your cost." This works for SaaS where the marginal cost is zero. It does not work for usage-based infrastructure where the customer can do back-of-envelope math against AWS list prices in 60 seconds. We priced from cost-plus with a sanity check on competitor list rates, not from a willingness-to-pay model.

Annual-only Enterprise. Some platforms force annual on Enterprise to lock revenue. We offer monthly, quarterly, and annual at the same per-month rate (16.7% off for annual prepay). The lock-in is real but it's the customer's choice; we don't make it the only option.

Free tier with no quotas. Some competitors run Free as "use everything, hit a wall, upgrade to continue." We hate this. Our Free tier is real exploration: 3 identities, 1 phone number, 100 emails, 10 SMS, 10 voice minutes, 20 outbound calls/mo. You can build a real demo agent on Free. You cannot run production workload — and we say that on the page.

Hidden compliance fees. Twilio adds carrier fees, A2P 10DLC registration fees, country-specific surcharges that pop up on the invoice. Anima rolls these into the published rate for US-domestic. International is a separate sales conversation.

Things we still don't know#

Two open questions we'll know more about in 60 days.

Whether $99 Growth is the right shape. It's priced for "scaling startup with 5-10 agents and serious volume." If the actual customer concentration is at 3-agent / lower-volume, we'll need a more graceful Starter→Growth bridge. We're watching the upgrade conversion rate and time-on-tier distribution to find out.

Whether voice promo pricing recovers. We committed to $0.07/$0.15 through 2026-09-30 as a launch lever. The steady-state $0.08/$0.18 was set assuming Telnyx and ElevenLabs costs hold flat. If they move (likely up, based on six-month trajectory), we'll either eat more margin or move steady-state up too. We won't move promo pricing during the promo window — that's a commitment we keep.

How to read /pricing#

The /pricing page is built around a usage calculator. Plug in your expected email/SMS/voice/phone numbers and it computes the cheapest tier, plus what your overage would look like. We added a comparison table showing what an equivalent stack costs on AgentMail + AgentPhone + Composio ($160/mo for the same volume), SES + Twilio DIY ($82 + eng), and Resend + Vapi DIY (~$65 + eng) — list rates only, not loaded for the engineering time it takes to actually build the compliance and audit infrastructure.

We think Anima Growth at $99 wins for the typical mid-volume agent operator. We could be wrong. The calculator on the page is the honest answer; if your usage profile makes Starter or DIY cheaper, run it and pick.

Pricing is a contract. We'd rather show our cost math and let you decide than play the willingness-to-pay game.