There is a pattern spreading through business software pricing, and it deserves more scrutiny than it gets: the two-meter bill. Meter one counts your people. Meter two counts your AI usage. Individually, each looks reasonable. Together they produce something new — a bill that grows when your team grows and when your business does well, with a ceiling nobody can predict.
How we got here
Per-seat pricing was the default of the SaaS era for a good reason: it roughly tracked value when software was a tool a person operated. More operators, more value, more seats.
Then AI arrived with a real marginal cost. Every model call costs the vendor actual money, so usage-based pricing followed — per message, per credit, and in support software, per resolution: the AI answers a customer successfully, you pay for that answer.
Most vendors did not replace the seat meter with the usage meter. They stacked them. And in the support and CRM category, a third layer is common: the AI features sit in higher tiers, so you also pay a plan upgrade for the right to pay for usage.
Why stacked meters hurt more than either alone
You pay twice for the same outcome. The promise of AI in support is that software resolves what people used to. Under stacked pricing you keep paying for the seats and start paying per AI resolution. The substitution you were promised never shows up on the invoice.
Your bill peaks exactly when you are weakest. Usage meters bind to activity, and activity peaks are rarely good news: a product incident, a shipping delay, a viral complaint. Per-resolution pricing means your worst support week is also your most expensive one. A seat bill at least fails predictably.
Forecasting becomes guesswork. A CFO can budget seats. Nobody can budget “resolutions next quarter” — that number depends on your growth, your product quality and the model’s mood. When you cannot forecast a cost, you either over-provision or you ration usage of the thing you bought because it works.
Seat prices quietly gate collaboration. When every login costs money, companies start sharing accounts, leaving the warehouse team out of the system, or exporting data to spreadsheets for people who “don’t need a seat.” The software’s value depends on everyone being in it; per-seat pricing pushes people out of it.
We will not quote specific competitor prices here — those change and deserve their own dated, sourced treatment, which we maintain on our comparison pages. The structural argument does not depend on any one vendor’s numbers.
What the alternative looks like
The failure modes above suggest the design directly: price the workspace flat, meter the AI transparently, and make the cap a stop rather than a bill.
That is what we did, and since this is a pricing post, here are the real numbers rather than a “contact us”:
| Plan | Price / month | Seats included | AI credits / month |
|---|---|---|---|
| Free | $0 | 2 | 50 |
| Starter | $19 | 3 | 1,000 |
| Pro | $59 | 10 | 5,000 |
| Business | $129 | 25 | 15,000 |
Prices are flat per workspace, not per user. Extra seats are $5/month each — deliberately cheap, because we want your whole team in the system, not sharing a login. Annual billing is two months free. Full detail, including the feature ladder, is on the pricing page.
How honest AI metering works
Three mechanisms make an AI meter trustworthy rather than just present:
Weighted credits, published. Not all AI actions cost the same to run, and pretending they do hides the meter’s logic. In Oneop, a chatbot reply costs 1 credit, a copilot query or AI-drafted email 2, a document OCR 3, an agent action 5, an executive-level insight 10. You can see what any workflow will cost before you build it.
One choke point. Every AI call in the product — chatbot, agents, drafts, OCR — passes through a single metering layer that reserves credits before the model call and records usage per workspace. There is no unmetered side door, and we run an automated test that fails our own build if any agent code path tries to bypass it.
A hard cap, not overage. When credits run out, the AI feature declines the call. It does not silently continue and bill you later. Deterministic features — your flows, your tickets, your records — keep working. The worst case of a hard cap is a paused feature and an upgrade decision made by you, in daylight. The worst case of overage billing is a number you discover on an invoice.
And for teams that outgrow any allowance: on the Business plan you can bring your own Google Gemini API key and pay Google’s rates directly, with Oneop applying its guardrails and metering visibility on top. The model bill becomes a first-party cost you control, not a marked-up mystery.
Questions to ask any vendor before you sign
- “What is my worst-case monthly bill?” If the answer requires assumptions about your ticket volume, you do not have a price — you have a formula with your risk in it.
- “What happens at the usage limit?” Hard stop, throttle, or automatic overage? Get it in writing.
- “Which AI actions are metered, and at what weight?” If they cannot enumerate it, they cannot have built a clean meter.
- “Does adding a read-only user cost a full seat?” This tells you whether the pricing wants your whole team in the product or just your budget.
- “If your AI resolves more, do I pay more?” Per-resolution vendors will say yes and call it alignment. Decide whether you agree.
The principle underneath
Pricing is an architecture decision that leaks into your operations. A meter on seats keeps teammates out. A meter on success makes you ration the product. A flat price with a transparent, capped AI allowance keeps the incentive clean: we win when the platform does more for the same money, so our pressure is toward efficiency — smaller prompts, cheaper paths, deterministic flows where a model call is not needed.
If you want to test the model rather than take our word: the free plan is $0, two seats, 50 AI credits a month, no credit card — and the credit ledger is visible in the product from day one, so you can watch exactly what each action costs before a dollar changes hands.