Understanding AI credits: a metered approach to practice automation
How a prepaid AI credit wallet keeps automation costs predictable across summaries, triage and drafting, and where the spend actually pays for itself.
Most accounting principals have the same reaction to AI features inside their software: cautious interest, followed quickly by a question about the bill. The interest is fair, because summarising a long client thread or triaging an inbound request genuinely saves minutes that add up across a busy practice. The worry is fair too, because per-seat AI add-ons and open-ended usage charges make it hard to know what you have committed to. A metered credit model answers that worry directly. Instead of an unlimited promise with an unpredictable invoice, you hold a balance of credits and spend them only when an AI action runs.
What an AI credit actually is
In Finye, AI features draw from a single prepaid credit wallet attached to your practice. Every AI action consumes credits from that balance, and nothing runs once the balance reaches zero. There is no surprise overage and no metered charge that lands after the fact. You top the wallet up, the work draws it down, and you can watch it move.
Credits are consumed by the AI features the practice already relies on day to day:
- AI ticket summary condenses a long work item, with its comments and back-and-forth, into a short brief so whoever picks it up is oriented in seconds.
- AI triage reads an incoming request and suggests a priority, an assignee and a category, so new client emails do not sit unsorted.
- AI reply and compose drafts a response you can edit, rather than starting from a blank box.
- AI draft turns a one-line brief into a structured work item.
- AI knowledge-base drafting produces a first-pass article from existing material.
- Dashboard AI summary gives a plain-language read of what is happening across the practice.
Because each of these is a discrete action rather than an always-on subscription, you only pay for what your team genuinely uses. A quiet week costs less than a busy one, and the cost tracks real work rather than headcount.
Why metered beats per-seat for a practice
Per-seat AI pricing assumes every staff member uses the feature the same amount, which is rarely true in an accounting firm. A senior who triages the inbox each morning and a junior who occasionally summarises a thread are charged identically under a seat model, and you pay for the seats whether or not anyone touches the feature that month. The result is a fixed cost that drifts away from actual value.
A credit wallet inverts that. The cost follows usage, so the heavy users carry the spend and the light users cost nothing when idle. For a practice with seasonal peaks, this matters: the run-up to a BAS quarter or the end-of-financial-year crunch is exactly when AI triage and summaries earn their keep, and exactly when a metered model lets you lean on them harder without renegotiating a plan. In the quiet weeks afterwards, the balance simply sits there.
It also makes the spend legible. Because every action draws from one shared balance, you can see how quickly credits move and tie that movement back to the features driving it. That visibility is the thing principals usually want most: not unlimited AI, but AI they can budget for.
Platform credits or your own key
Finye supports two ways to power these features. You can use platform credits, where you top up a wallet and the practice handles the underlying AI provider for you, with Stripe top-ups and an optional auto-recharge so the balance refills when it runs low. Or you can bring your own provider key, in which case the AI calls run against your own account and you manage that cost directly.
The choice usually comes down to how much you want to think about it. Most practices prefer platform credits because the wallet, the top-ups and the auto-recharge remove any operational overhead. Firms that already have a provider relationship, or that want their AI usage billed entirely on their own terms, can use their own key instead. Either way the features behave the same inside Finye; only the billing path differs.
Where the spend pays for itself
The honest answer is that not every AI action is worth running, and a metered model is the right tool precisely because it forces that judgement. A few patterns tend to pay off reliably:
- Triage on a shared inbox. If inbound client emails become work items, triage suggesting a priority, assignee and category saves the daily sort and stops requests from stalling because nobody owned them.
- Summaries on long-running jobs. When a work item has weeks of comments, a summary saves whoever inherits it from reading the whole history. The minutes saved scale with the length of the thread.
- Drafting replies and articles. A draft you edit is faster than a draft you write. For routine client responses and first-pass knowledge-base entries, the edit-from-draft pattern is consistently quicker.
Where the case is weaker is anything a person would do faster unaided, or any output that needs heavy correction. Because credits are finite and visible, your team naturally reserves them for the work where the time saved is real. That self-limiting behaviour is a feature, not a flaw: it keeps the spend pointed at value.
A practical way to start is to turn on the features you expect to use most, top up a modest balance, and watch how quickly it moves over a fortnight. The draw-down tells you which actions your practice actually leans on, and from there you can size your top-ups or set an auto-recharge threshold with real numbers rather than guesses. Pairing this with your existing workflow setup means the AI sits inside the work you already do rather than alongside it.
Budgeting with confidence
The point of a metered wallet is not to make AI cheap; it is to make AI predictable. You decide the balance, the balance caps the spend, and the spend follows real usage rather than seat counts. For a principal weighing whether automation belongs in the practice, that predictability is what turns a vague worry about cost into a line item you can actually plan around.
If you want to see how the credit wallet works alongside work items, triage and the rest of the platform, take a look at our pricing or start with Finye and top up a small balance to test it against your own workflow.