Per-event metered pricing: why monthly tiers force over-commit

Update, October 2026: Mails.ai did not launch the metered plan this post describes. Our plans are Free, Pro and Scale, billed monthly or yearly — see the pricing page. The rate card and worked examples below describe the plan as it was proposed in May 2026; it is not available.

Most email pricing was built for newsletters and marketing blasts. You buy a monthly bucket and hope you use most of it. AI agents don't work that way — they're quiet most of the time and busy in bursts. This post explains why that mismatch costs you money, and how per-event pricing would fix it.

Monthly email tiers were designed for marketing-blast SaaS where send volumes are predictable and roughly the same every month. Agent traffic is not that. A support agent might send 10 emails on a quiet day and 1,000 on an incident day. A 2FA-bridge agent fires only when a human signs in. A document-parser agent spikes when a customer uploads a batch. Monthly tiers force you to over-commit for the spike and waste capacity on the quiet days.

Architecture

The over-commit math

The current incumbent pricing for a 1,000-emails-per-month workload:

• Resend: free up to 3,000 emails a month, then $20/month for 50,000. At 1,000 you pay nothing — at 3,001 you pay $20.
• Postmark: $15/month for 10,000 emails. You use 1,000. You ate $15. Effective cost: $0.015.
• SendGrid Essentials: $19.95/month for 50,000 emails. Same shape.

The opposite scenario is also bad. A 5,000-email-per-month workload on Resend Free (3,000 cap) has to upgrade to the $20 tier and instantly access 47,000 emails of unused capacity. The pricing structure does not match how agent products actually scale.

Per-event metered pricing — the alternative

Per-event metered pricing means every event — a send, a parsed inbound, an opt-in classification — posts a usage record to the billing system. At month-end, the invoice is the aggregated usage. No tier. No minimum. Mails.ai proposed exactly this in May 2026 and did not launch it.

The mechanism would be Stripe Metered Billing, the same billing primitive that OpenAI, Twilio, AWS, and most modern usage-based SaaS run on: each time an agent sends or receives an email, one usage record posts to Stripe, and the month-end invoice is the running total.

What an event would have cost

The proposed rate card (not available):

EVENT TYPE                          PRICE
─────────────────────────────────────────
send                                $0.001
inbound (structured reply event)    $0.002
classification (opt-in)            +$0.003
  (intent, entities, urgency)
suppression update                  free
identity check                      free
agent.list_threads call             free

Under the proposal, reads against your conversation history, suppression list updates, and identity checks were free; you paid only when an actual email moved — in or out. Today, every inbound event carries injection_score and sender_reputation on every plan, and classification (intent, entities, urgency) of first-contact mail is opt-in per agent on paid plans at no extra charge.

Worked examples at the proposed rates

Run the math on actual agent shapes:

• 10/day support agent: 10 sends × 30 days = 300 sends. Plus 200 parsed inbounds (replies). 300 × $0.001 + 200 × $0.002 = $0.70/month.
• Inbound-heavy triage agent: 5,000 parsed inbounds in a busy month. 5,000 × $0.002 = $10/month — before any sends.
• 1,000/day operations agent: 1,000 sends × 30 = 30,000 sends. 30,000 × $0.001 = $30/month.
• 2FA-bridge agent (50 sends/month): 50 × $0.001 = $0.05/month. The agent that exists for one specific job pays one specific amount.

Same agent shape, different volumes, proportional cost. Compare to Resend at $20 flat anywhere from 3,001 to 50,000 emails.

Why incumbents cannot match

Resend, Postmark, SendGrid, and Mailgun all built their economics around monthly tiers. The over-commit IS their gross margin — the difference between what customers pay and what those customers actually use. Switching to per-event pricing would cannibalize existing revenue at every renewal. They'd have to cut prices for their existing customers to match — that's why they won't.

Amazon SES has true per-event pricing ($0.10 per 1,000 emails). But SES is raw infrastructure — no agent primitives, no structured reply events, no injection scanning, no per-agent reputation, no MCP server. You build all of that on top yourself.

Mails.ai built the agent-native primitives but did not launch the per-event plan; its plans stay monthly or yearly, with separate allowances for emails and inbound replies.

Checking usage in your dashboard

The dashboard shows this month’s emails and inbound replies against your plan’s two allowances. The same numbers come back from GET /v1/billing/usage — useful if your agent needs to throttle itself before it reaches the cap.

Which plan to choose

The right plan depends on volume:

• Monthly Pro / Scale if your agent has steady volume. At full included volume the monthly plans are far cheaper per event than the proposed metered rates — Pro’s 50,000 emails + 50,000 inbound would have cost $150 metered versus $20 flat.
• Yearly Pro / Scale if you know you’ll stay — $200/year and $990/year, two months free.
• Free if you are still prototyping — 3,000 emails and 3,000 inbound a month at $0.

You can change plans month to month without contract penalties, and Free costs $0 in a quiet month.

Read the pricing page for the full plan table and the architecture page for how deliverability protection actually works.