By the Onsites AI team · Last updated · 9-minute read
"AI-priced support" has become a phrase that hides two very different ideas: subscription platforms that charge per seat and then per AI resolution, and prepaid utility models that charge only when the AI actually does something. The second is dramatically easier to budget once you see the per-action numbers, so this guide publishes them, walks two real monthly workloads through the math, and lists the levers that cap the bill.
One credit is roughly a cent. AI actions spend fixed, published amounts: a Copilot draft is 3 credits (~$0.03), a suggested reply is 6 (~$0.06), a translation is 3, a conversation summary is 6, a Harness question is 15 (~$0.15) and Harness with tools is 24 (~$0.24) — that last one being an AI run that reads your CRM, handbook and chat history and acts with visible sources. Credits come in prepaid packs — 1,100 for $10, 2,400 for $20, 6,500 for $50 — and never expire. Spent only when AI is used; the desk is free either way.
Take a genuinely busy small team: 300 conversations in a month, AI assisting on 60% of them. Draft 180 replies (540 credits), translate 60 messages (180), summarize 150 threads (900), ask Harness 40 questions (600) and run 20 tool-enabled answers (480). Total: 2,700 credits. One $50 XL pack (6,500) covers more than two such months; one $20 pack plus a part-month covers one. The line item for all AI assistance in a 300-conversation month is $20–50, prepaid, never expiring — and if the team leans harder on Copilot and lighter on Harness some month, the bill follows the usage down rather than the baseline up, which is the property that per-seat billing never offers.
A one-person operation leans on AI differently: 120 conversations a month, mostly drafts (say 80 drafts, 240 credits) and summaries of 60 threads (360), a dozen translations (36) and the rare Harness question (150 across ten). Total around 800 credits — the $10 Popular pack (1,100) runs a month and a half, and the unspent 300 credits roll into the next month because credits never expire. Even with the desk fully AI-assisted, the annual AI bill lands under $100, itemizable on one published price list.
Per-seat-and-per-resolution platforms commonly charge around $0.99 per AI resolution. The same 300-conversation month with, say, 300 resolutions lands near $297 in resolution fees alone — before seat fees, before the next tier's minimums. Our comparison in the FAQ is blunt about the result: a comparable 10-person workload costs roughly $255/month here versus ≈$3,100 on a per-seat-plus-per-resolution platform. Fixed credit prices are why: the meter never scales with your customers' patience, only with your actual usage. The per-resolution math post takes that argument apart in full.
For the full billing reference — packs, storage meters, the 14-day cloud refund policy and what counts as consumed — read the billing doc; and if you are weighing the self-hosted track, remember AI there runs on your own model, with no credit packs sold on that track at all. When you want the whole pricing story on one page, the pricing page carries the live calculator.
AI credits are one of three meters, and deliberately the only one that touches the support workflow by default. Seats: the first three are free forever, the fourth onward is $15/seat/month prorated daily — that meter moves when you hire, not when you help. Storage: 100 MB free, additional packs at $1 per GB per month (10GB=$10, 50GB=$50, 100GB=$100, 500GB=$500); chats and emails count too, but text is tiny, so in practice the meter tracks attachments. Neither meter interacts with AI spend: a team can spend a year on the free tier with the credit meter reading exactly zero, and plenty do — the free plan's AI is simply off until a pack makes it a choice.
The version finance sign-off accepts reads: "300 conversations a month, AI-assisted where useful, one $20–50 pack — that is the whole line." The full sentence behind it: drafts at 3 credits and summaries at 6 dominate the bill; translations at 3 are rounding error; Harness asks at 15 and tool runs at 24 are deliberate purchases, not ambient costs. Worst case, if a procurement officer wants one: the pack you chose to buy — prepaid, fixed-price, never expiring, with soft and hard limits already on in Settings → Billing → AI credits. Nothing in the structure can invoice a surprise, which is the point of pricing AI like ink instead of like seats.
Do credits expire?
No — prepaid and non-expiring; the balance waits until AI is used.
What does one action cost?
Drafts 3 (~$0.03), suggested replies 6, translations 3, summaries 6, Harness asks 15, tool runs 24 (~$0.24).
What prevents runaway spend?
AI stays off until enabled; a soft limit is on by default; hard limits are available; every action's price is fixed and published.
What if credits run out?
AI actions pause until the next pack; the desk itself works normally, and the unspent balance never expires.