How do I estimate each feature's usage share?

Multiply each feature's average tokens per call by its monthly call volume to get its token consumption, then express each as a percentage of the total. The allocator handles the normalisation if your percentages don't add to exactly 100.

Why normalise the shares?

So that 100% of your spend is always allocated, even if your entered percentages don't sum perfectly. If your three features are entered as 45, 30 and 25, they already sum to 100; if they sum to 90 or 110, each is scaled proportionally.

Should I use token share or call share?

Token share is more accurate, because cost is driven by tokens, not request count. A feature with few calls but huge prompts can cost more than a high-volume feature with tiny prompts.

Does this account for input vs output token pricing?

Not directly — it allocates a known total by share. For per-feature spend from first principles, estimate each feature's input and output tokens with the LLM API cost calculator first, then use those figures as your shares here.

Is my data sent anywhere?

No. Everything runs in your browser. No spend figures or feature names are uploaded, stored or logged.

What is the Per-Feature AI Cost Allocator?

Free per-feature AI cost allocator. Enter your total monthly LLM spend and each feature's usage share to split your API bill across summarization, Q&A, classification and more — for accurate per-feature P&L, all in your browser. It runs free in your browser on Gera Tools, with nothing uploaded.

Per-Feature AI Cost Allocator

Name: Per-Feature AI Cost Allocator
Creator: Gera Tools
License: https://creativecommons.org/licenses/by/4.0/

Get one useful tool a week

Like this tool? Enter your email and we'll send you one genuinely useful Gera tool a week — plus a link to come back to this one. No spam, one-click unsubscribe any time.

Turn one API bill into a per-feature P&L

When everything goes through a single API key, your monthly LLM invoice is one opaque number. But your CFO wants to know which feature is driving cost — is it the summarizer, the chatbot, or the classifier? This allocator splits your total spend across features by usage share, giving you a clean per-feature cost line you can drop straight into a P&L or unit-economics model.

How it works

You enter your total monthly spend and a list of features, each with a usage share as a percentage. The tool normalises those shares so they always sum to 100%, then allocates the spend proportionally: a feature with a 45% share of token consumption gets 45% of the bill. Because cost is driven by tokens, the most accurate share for each feature is its token consumption — average tokens per call multiplied by monthly call volume — rather than its raw request count.

Worked example

Suppose your monthly AI bill is $1,200 and you have three features:

Feature	Avg tokens/call	Monthly calls	Token share
Document summariser	4,000	500	55%
Q&A chatbot	800	1,200	26%
Classification	300	2,500	20%

Total tokens: 2,000,000 + 960,000 + 750,000 = 3,710,000. The shares (55%, 26%, 20%) don’t sum to 101% exactly — the tool normalises them automatically. Allocated cost:

Summariser: 55% × $1,200 = $660/mo
Q&A chatbot: 26% × $1,200 = $312/mo
Classification: 20% × $1,200 = $240/mo (approx, before normalisation)

Now the summariser’s true cost-per-call is $660 / 500 = $1.32, which immediately raises the question of whether a cheaper model or prompt compression would cut it.

Why this matters for unit economics

Without feature-level attribution, you cannot calculate the margin of a specific product feature. If the summariser drives only 20% of revenue but 55% of AI cost, it is a candidate for optimization before it becomes a profit drain. Knowing which feature is expensive also tells you where caching is most valuable — a result that repeats even 30% of the time can halve that feature’s effective cost.

Tips for accurate attribution

Use tokens, not requests. A feature with long prompts can dominate cost even with few calls. Weight by tokens for a true picture.
Separate input and output. If features differ sharply in output length, estimate each feature’s cost with the cost calculator first, then feed those dollar figures back in as shares.
Track over time. Re-run monthly to spot a feature whose share is creeping up — that is usually where an optimization (caching, a cheaper model) pays off most.
Tag spend at the source. For ongoing accuracy, attach a feature tag or metadata field to every API call so your provider logs can confirm these shares.