AI credits explained: what agencies actually pay for

AI credits are a prepaid balance, priced in fractions of a cent, that you draw down per AI call based on the tokens that call actually used — not a flat per-seat “AI add-on” that charges the same $20/month whether someone uses it once or constantly. That distinction is the whole story of why agencies find AI pricing pages so confusing right now: most vendors haven’t decided which model they’re actually selling, and the label on the pricing page often doesn’t match the mechanism underneath. This post uses sSystm’s own credit wallet as a concrete, verifiable example of the token-based model — and gives you a mental model to bring to any other vendor’s pricing page, sSystm’s included.

Why AI pricing confuses agencies in the first place

Three patterns show up repeatedly when SaaS vendors bolt AI onto an existing product, and all three obscure what you’re actually paying for:

• Per-seat AI add-ons. A flat monthly fee per user, regardless of usage. A copywriter running twenty AI drafts a day and a project manager running none pay the identical fee.
• Opaque “fair use” limits. Marketing copy promises “unlimited AI,” and the fine print defines unlimited as a monthly cap that resets — until you hit it mid-project and get throttled with no warning.
• Surprise overage bills. Usage-based pricing without a prepaid ceiling means the bill arrives after the usage happened, sized by however much the team generated that month.

None of these are illegitimate business models. The problem is that the pricing page rarely tells you which one you’re buying, so you can’t predict next month’s invoice from this month’s usage pattern. A model that shows you the meter before you spend — a balance you top up and watch draw down — solves that predictability problem structurally, not with better copy.

How the credit wallet actually works

sSystm’s AI pricing is built as a prepaid wallet denominated in micro-USD — millionths of a dollar. That precision matters: a single short AI reply might genuinely cost a fraction of a cent to generate, and a billing system built on whole cents either rounds that away or forces an artificial minimum charge per call. Micro-dollar accounting means the ledger can record the exact cost of every individual call without distorting it.

The mechanics, end to end:

1. Top up once. An organisation buys credit via a hosted Stripe Checkout session, in any amount from $5 to $1,000, at 1:1 — a dollar buys a dollar of credit.
2. Every AI call is measured. Each call’s actual input and output token counts are converted to a cost in micro-USD using a per-model rate table, then a resale markup is applied.
3. The balance is drawn down per call, not per seat and not per month. A quiet afternoon costs nothing beyond what was generated; a heavy sprint costs proportionally more — and it’s visible in a running ledger, not discovered on next month’s invoice.
4. New organisations start with a small free credit grant so you can see the mechanism working before you ever reach for a card.

This is genuinely different from a seat-based AI tax: cost tracks consumption, not headcount. A five-person agency that uses AI heavily and a fifteen-person agency that barely touches it will land in opposite places on the bill — which is the outcome you’d want from something billed as “usage-based,” but which a flat per-seat fee can never produce.

One honest caveat, because overclaiming a launch status is worse than a small omission: the wallet, the top-up flow and the per-call usage ledger are built and live today. Whether a given account’s balance is actively debited for every call, versus usage simply being measured and recorded, is a rollout switch the team controls deliberately — the same way a utility company might finish installing smart meters before flipping on metered billing. The point for you as a buyer isn’t which day that switch flips; it’s that the mechanism is architected around real consumption from the ground up, which is the part that determines whether your bill will ever feel arbitrary.

What’s free versus what’s actually monetised

This is the part most all-in-one platforms blur, deliberately or not: bundling often means everything is nominally “included” until you discover the asterisk. sSystm draws the line in exactly two places, and nowhere else:

	Free, always	Monetised
Core workspace	CRM, Projects, Billing, Documents, Calendar	—
AI usage	—	Prepaid credit wallet, billed per token via micro-USD ledger
Premium modules	—	À la carte module packs (see below)
Core AI surfaces	Assistant rail, MCP configuration, Cloudflare automation	Never gated — these are platform, not product

The core BYOC workspace — the CRM, project delivery, billing and documents an agency runs day to day — carries no seat tax and never will; see why agencies need CRM, projects and billing in one tool for what that core actually includes. The two things that are monetised are the AI wallet described above, and a second, separate axis: optional premium module subscriptions.

The other paid axis: à la carte module packs

Separate from AI usage, sSystm sells access to curated bundles of premium modules as flat monthly or annual subscriptions — the “capability” axis of pricing, as opposed to the wallet’s “consumption” axis. Four suites, plus a bundle of all four:

Suite	Monthly	Annual	What it unlocks
Build Suite	$49	$490	Design System, Components, Wireframes, Build (AI code generation), Skills, Interactions, Layouts
Marketing & Content Suite	$39	$390	Content Planner, Campaigns, Newsletter, Social, Brand Assets, Tone of Voice, Leads, Approvals, Reports
Analyze Suite	$29	$290	Cognitive Flow, Rage-Click Predictor, Z-Index Exorcist, Dehalucinate
Collaboration Suite	$15	$150	Chat, Routines
All Access	$99	$990	All four suites — roughly 25% cheaper than buying them separately

Every annual price is exactly ten times the monthly price — two months free compared with paying monthly, with no odd percentage rounding. Notice what’s deliberately absent from every row: the AI assistant, MCP configuration and the underlying Cloudflare provisioning are never part of a paid pack. Those are core platform, not premium product — an agency should never have to buy a suite just to keep talking to its own AI. See the full module catalogue for what each suite contains, and pricing for the current numbers in one place.

A mental model for reading any vendor’s AI pricing page

You don’t need to be evaluating sSystm to use this. The next time a vendor’s pricing page mentions “AI-powered,” ask three questions before you compare a single dollar figure:

1. Does the price change with usage, or only with headcount? If the AI line item is a flat fee per seat, you’re paying a tax on having employees, not a price for consuming AI. If it’s metered — credits, tokens, a balance that visibly draws down — the price tracks what you actually did.
2. Can you see the meter before you’re billed, or only after? A prepaid balance you top up and watch drain is predictable by design. A monthly invoice that totals up usage after the fact is usage-based in name but surprise-based in practice.
3. What, specifically, is core versus paywalled? Vendors that gate the AI assistant itself behind a paid tier are charging you to access their product’s brain. Vendors that keep the assistant and its configuration free, and charge only for consumption or for premium feature suites on top, have drawn the line in a more defensible place.

None of this makes metered pricing automatically cheaper than a seat-based plan — a very heavy user can spend more on tokens than they would on a flat fee, and that’s the fair trade-off of paying for what you use rather than what you might use. What it does is make the bill predictable and legible: you can watch the wallet, not wait for the invoice. That’s the property worth demanding from any AI pricing page you read next, sSystm’s included. For the architecture underneath — why this all runs on infrastructure your organisation owns rather than a shared vendor database — see how sSystm works, or start directly with the AI assistant the credit model is metering.