Where AI helps a small accounting practice
A small accounting firm has a lot of language work sitting around the numbers. Drafting client emails, turning a messy query into a clear reply, summarising a change in regulations, drafting a first pass of a memo, tidying notes after a call. None of it replaces the judgement, but a good model takes the friction out of the writing.
Drephos puts the leading models in one place, so a partner can pick Claude for a careful draft and GPT or Gemini for something quicker, all from one workspace.
Why prepaid credits suit tax-season work
Accounting work is seasonal in a way that per-seat subscriptions handle badly. The run-up to a filing deadline is a burst of long days, and then it quiets down. A per-seat plan charges the same in January and in the sleepy weeks after, so you pay a full year of seats for a few intense months.
Drephos is prepaid instead. One shared wallet, topped up in USD packs, drawn down only when someone runs a paid model. You buy more credits when the season demands it and spend almost nothing when it does not. Credits are valid a full year, refreshed on every top-up, so a top-up before tax season is not wasted if some of it carries into the quiet months.
Here is a practice of twenty across a year, billed both ways. The seat plan cannot tell January from August.
Un plan por puestos
$6,000
al año, 20 puestos facturados cada mes
| Mes | Coste |
|---|---|
| enero | $500 |
| febrero | $500 |
| marzo | $500 |
| abril | $500 |
| mayo | $500 |
| junio | $500 |
| julio | $500 |
| agosto | $500 |
| septiembre | $500 |
| octubre | $500 |
| noviembre | $500 |
| diciembre | $500 |
Drephos
$2,620
al año, según el uso real
| Mes | Coste |
|---|---|
| enero | $380 |
| febrero | $260 |
| marzo | $300 |
| abril | $320 |
| mayo | $180 |
| junio | $120 |
| julio | $60 |
| agosto | $50 |
| septiembre | $130 |
| octubre | $200 |
| noviembre | $280 |
| diciembre | $340 |
Estos meses de Drephos son un ejemplo, no un presupuesto. El coste real depende del uso. Un mes tranquilo cuesta casi nada y uno intenso solo cuesta lo utilizado.
Keeping client information in bounds
Client confidentiality is the whole job, so the controls matter. Conversations are not used to train models. Memory is opt-in, set per chat, and easy to clear. You add staff with a single join link that you can lock to your firm's email domain, set a monthly cap per member, and see a per-member breakdown of usage, so the shared wallet and who is using it stay understandable.
When a per-seat plan makes more sense
To be fair: if every person in the practice uses AI heavily and evenly all year round, a fixed per-seat plan is easy to budget and there is less to think about. Drephos fits better when your usage is seasonal or uneven, when you want more than one model, or when you would rather pay for what the firm actually uses than a seat for every name. The pricing page shows how the credits work.