By the Onsites AI team · Last updated · 5-minute read
Pricing pages are designed to be compared feature-by-feature, which is exactly how the meter hides. Put the comparison in the only currency that matters — the annual bill for the same workload — and the story becomes blunt. This guide takes one desk: ten people, 4,000 conversations a month (a mid-sized operation across email and chat), AI assists used on about half the replies (drafts, occasional summaries, some translation), attachments modest. It runs that one workload through a per-seat-plus-per-resolution stack and a flat metered one — Onsites — line by line, with every assumption stated so you can swaps your numbers into the calculator and rerun it. The published-meters summary: ≈$3,100/mo for the resolution-metered model, ≈$255/mo for the flat one. The derivation matters more than the total, because your volume will differ and the shape of the gap is what transfers.
Fix the shape first. Ten seats — eight agents, one team lead, one admin who also handles billing. 4,000 conversations monthly across email, widget chat and WhatsApp, seasonal peaks to 6,000. AI use at realistic mid-adoption: roughly 1,200 drafts (3 credits each on the credit-priced desk), 300 suggested replies (6), 200 summaries (6), 400 translations (3) — about 7,400 credits monthly at full adoption, and half that for a team still building the habit. Storage: text-first support with some PDFs, comfortably inside 100 MB. SLAs: honest ones — 1 business-hour first response. Nothing exotic: this is what a real ten-person desk looks like in most trading, agency and light-industrial businesses. Hold these numbers through both derivations; change them freely after.
The resolution-era model, which the majors moved to during the Fin era, meters two things: seats and each conversation AI (or a human, under some packs) actually resolves. Seats: ten, on a mid-suite tier, in the low-hundreds per seat per year depending on tier and negotiation — using conservative list-style arithmetic, roughly $1,200–$1,500 monthly for the desk alone, before add-ons. The resolution meter: published rates around $0.99 per AI-resolved conversation; even a modest 20% AI share on 4,000 conversations is 800 resolutions ≈ $790/mo — and AI deflection is supposed to climb, which raises this line every quarter it succeeds. Add-ons and the tier tax: the features a ten-person desk actually wants (advanced routing, surveys, quality tooling) typically live a tier up or ride add-on pricing — budget several hundred more. Total for the same workload: ≈$3,100/mo, and the line items grow with success (more AI share, more resolutions) and punish the peak season twice. Rerun with your own seat count and AI share; the cliff shape persists.
The same workload on the metered desk, line by line. Seats: three free forever, seven paid at $15 = $105/mo billed annually, prorated daily as people join — the 11th hire prices $15, not a tier jump. AI credits: 7,400 credits at published prices ($10/1,100, $20/2,400, $50/6,500 — never expiring) ≈ $100/mo on the best pack rate, a line that falls to near-zero in slow months rather than billing a subscription. Storage: 100 MB included; text-first desks fit, so $0 most months, $1/GB beyond if attachments pile. Everything else: no per-resolution meter, no automation caps, no survey add-on, and the first payment removes the watermark permanently. Total: ≈$255/mo — and the same workload in a flood month (6,000 conversations, heavier AI) still lands near $300, because volume rides only the credits, which bill at cents, not at resolution rates. That is the whole trick: no meter in the stack compounds with volume except one, and it is priced at about a cent.
The $2,845 monthly difference is not one number; it is three multipliers. Seat count multiplies the tier: every pricing-model quirk (the tier tax, the add-on) applies per seat monthly — a $50 add-on is $500 for ten people and $6,000 a year. Success multiplies the resolution meter: AI that deflects better raises the line it sits on; the desk is charged extra for its own automation working, which inverts the incentive — the credit meter prices the same drafts at cents regardless of how well they perform. Headcount multiplies everything: per-seat models reprice at every hire, and the hire that tips a tier boundary reprices retroactively at renewal. The flat meter has exactly one seat line and one credits line, both published, both scaling gently. Twelve-month view for this workload: ≈$37,200 on the resolution-metered stack, ≈$3,060 on the flat one — a full junior hire, every year, spent on the meter's shape rather than the desk's work.
The derivation is only as honest as its inputs, so stress-test both sides. What favors the resolution-mtered stack: enterprise discounts negotiated off list (though the calculator's job is modeling list, procurement negotiates); heavy seat sharing where "ten people" licenses five; AI resolution rates lower than assumed. What favors the flat meter: seasonal peaks (credits pause; tiers rarely reprice down mid-year); headcount growth (hire 5, nothing changes at the tier level); AI adoption rising (a good quarter lowers the credits line rather than raising it). The tie-breakers: the loop — quotes, orders, invoices in the thread — is native on the flat desk and an integration project on the suites; self-hosting with your own model exists on neither of the majors; and the free three seats mean a team below that size pays $0 either way, making the comparison purely about scale. Run the worked comparison at your real numbers; the shape of your curve decides which meter deserves you.
Half the difficulty of pricing audits is translation: the stacks describe the same work in different words. A "resolution" on the AI-metered desk is a conversation the assistant (or a human, under resolution packs) closed; the flat desk has no such line — its closest unit is the credit, spent per assist, not per success. A "seat" means the same thing everywhere but prices wildly differently: the suites' seats ride tiered rates that bundle features, so a seat is really "a seat plus its tier's bundle," while the flat meter's seat is just a chair at the queue. "Automation" on suite pricing can meter per action (workflow runs, bot steps) — invisible lines that compound in busy months; the credit-priced desk folds all automation cost into cents-per-assist. When finance asks for the comparison, give it this translation table rather than either vendor's page: one meter bills outcomes (which rise as you succeed), the other bills inputs (which you control by usage). Inputs budget; outcomes surprise. The full models guide generalizes this to every pricing family you will meet.
What does a 10-person support team actually cost monthly?
Per-seat plus per-resolution stacks: roughly $3,100/month (suite seats, AI resolutions at about $0.99 on a fifth of 4,000 conversations, tier and add-on taxes). Onsites' flat meter: roughly $255/month (seven paid seats = $105 after three free, ~$100 of AI credits, storage included). Both derivations are line-by-line above.
Why is the per-resolution line so expensive?
It bills success: each AI-resolved conversation costs about $0.99, so AI adoption — the thing the product markets — raises the line. 800 AI-resolved conversations monthly is ~$790 before seats and tiers. Onsites prices the same drafts as prepaid credits at about a cent, never expiring.
Does Onsites charge per resolution?
No. The only meters are seats ($15/month beyond three free, prorated daily), storage ($1/GB past 100 MB), and AI credits (prepaid, ~1 cent each, never expiring). No per-resolution or per-automation fees at any volume.
What if our team is smaller — 3 people?
All three seats ride free forever with the full product, so the metered desk costs $0 plus optional credits. Any paid per-seat desk prices three seats from month one. Under three free seats, Onsites is the strictly cheaper meter.
Are these numbers negotiable or inflated?
They model published meters at conservative list-style rates; enterprise discounts move the resolution-metered figure down. The point is the shape — seats plus a volume-priced AI line against flat seats plus cent-priced credits. Swap your numbers into the calculator and rerun.