Plan your AI spend in one place
AI costs sprawl across subscriptions, pay-as-you-go API tokens, GPU compute, and one-off training runs — and they add up fast. This planner lets you list every AI cost as a line item, group them by category, and instantly see your monthly total, annual projection, and cost per seat, so you can budget confidently and spot where the money is going.
The four buckets of AI spend
Most organisations’ AI costs fall into four distinct categories, and separating them makes forecasting and control much easier:
Subscriptions — fixed monthly fees for SaaS AI tools, including productivity assistants, coding assistants, and team plans for chat interfaces. These costs are predictable and per-seat pricing makes them scale with headcount.
API token usage — pay-as-you-go costs for calling model APIs directly. These are variable and can spike sharply with usage changes. The price depends on the model tier, prompt length, and output length. Input tokens are generally cheaper than output tokens on most provider pricing schedules.
Compute — GPU instances, inference endpoints, and self-hosted model infrastructure. Often the most significant and least predictable category for teams running models themselves. Costs scale with inference load and model size.
Training and fine-tuning — one-off or periodic costs for fine-tuning runs, embedding generation at scale, and dataset preparation. These are often project costs rather than ongoing monthly expenses.
How it works
Each line has a category, a name, a monthly amount, and a quantity (seats, units, or instances). The planner multiplies amount by quantity for each line, sums them into a monthly total, multiplies by twelve for the annual figure, and divides by your team size for the per-seat number. A category breakdown shows the share of spend going to subscriptions, APIs, compute, and training. Everything is saved in your browser so your plan persists between visits.
Estimating API costs you cannot predict exactly
Token usage is the hardest category to forecast. A practical approach is to instrument a representative day of activity, measure the actual token volumes, and extrapolate. Where live usage data is not available, estimate from task type: a typical chat exchange might use a few hundred tokens; an automated document processing pipeline might use tens of thousands per document. Multiply by volume and model price per million tokens, then add a buffer. Revisit the estimate monthly once you have actuals.
Tips for an accurate budget
- Separate fixed from variable. Subscriptions are predictable; API and compute lines are not — add a 20–30% buffer to usage-based lines.
- Track per-seat cost as your headline metric for comparing teams and justifying spend to finance.
- Revisit monthly. AI pricing and your own usage both move quickly; reconcile the plan against actual invoices and adjust.
- Add a contingency line so an unexpected usage spike does not blow the budget.
- Set usage alerts with your provider. Most API providers let you set spend alerts; use them so a runaway automation does not produce a surprise bill before you notice.