Economic exposure calculator

The cost of a blind spot is the product of an accumulation rate and a detection delay. Without a signal, the loss shows up at billing time or in a complaint. This tool quantifies the exposure of not monitoring over twelve months, and therefore the value observability can avoid. It is designed for a discussion with a finance function or a risk committee.

Prices change fast and vary by contract: this guide gives none; use your provider’s dated price list. The price fields are therefore empty: as long as they stay empty, the token-related items are shown in tokens and calls; enter your input and output prices to get them in euros.

Parameters

Volume and tokens

LLM calls per day/ dayAverage input tokensper calltokensAverage output tokensper calltokens

Your prices (optional)

Input priceyour provider’s dated price list€ / M tokensOutput priceyour provider’s dated price list€ / M tokens

Prompt bloat

Monthly input token growth% / moUndetected monthsone episode per year, 12 at mostmonths

Agent loop

Share of calls wasted in avoidable loopssums up r x (L - K) / K from Part V%

Cache

Achievable response cache hit rateleft unexploited; a hit avoids the whole call%

Hallucination

Expected incidents per monthprobability x exposed incidents/ moCost per incidentremediation, reputation, regulatory exposure€

Exposure

Illustrative parameters, to replace with your own values. No price is provided: without prices, the token-related items are shown in tokens; enter the input and output prices from your provider's dated price list to get them in euros. The base cost is the legitimate spend of the service; the exposure is the avoidable or at-risk amount that observability makes visible. Every item covers twelve months: prompt bloat counts one episode per year, undetected for M months then fixed. The hallucination defaults are arbitrary: replace them with your own estimates. Working document, not financial advice.

Formulas

They simplify the model in Part V. Every item covers the same period, twelve months.

Item	Calculation
Cost per call	t_in x p_in + t_out x p_out
Monthly base cost	V x cost_per_call x 30
Annual base cost	monthly_base_cost x 12
Prompt bloat	monthly_input_base_cost x g x M x (M + 1) / 2, one episode per year, M at most 12
Unexploited response cache	h x annual_base_cost
Agent loop	wasted_share x annual_base_cost
Hallucination	expected_incidents_per_month x cost_per_incident x 12

Without prices, the calculator applies the same formulas to volumes: t_in and t_out multiplied by V x 30 give the month’s base tokens, and every token-related item is expressed in input and output tokens instead of euros.

Differences from Part V, item by item:

• Prompt bloat: same formula as Part V. To stay within twelve months, the calculator assumes a single episode per year, undetected for M months (12 at most), then fixed.
• Cache: this is a response cache, where a hit avoids the whole call. A prompt cache only lowers the price of input tokens read from cache; with one, this item overstates the saving.
• Agent loop: Part V computes V x r x (L - K) x cost_per_step x 30 per month. The calculator sums it up as a share of wasted calls, equal to r x (L - K) / K when each step costs one average call and the V of Part V counts tasks of K steps, that is K calls each.
• Hallucination: Part V writes P_hallucination x N_exposed_incidents x cost_per_incident. The calculator directly asks for the expected number of incidents per month, that is the product P x N, and multiplies it by twelve.

With the default values (50,000 calls per day, 1,500 input and 400 output tokens) and no prices, the base volume is 2.25 billion input tokens and 600 million output tokens per month, or 27 and 7.2 billion per year. The calculator then shows about 14.11 billion avoidable tokens over twelve months, or 41% of the annual token volume: 1.8 billion input tokens of prompt bloat (four undetected months), 10.26 billion tokens of unexploited response cache (5.4 million calls) and 2.05 billion tokens of agent loops (1.08 million calls). Hallucination is shown separately, in euros: 288,000 euros. To turn tokens into euros, multiply input tokens by p_in and output tokens by p_out, or enter those prices in the calculator.

Hallucination can dominate the total

With the default values, hallucination is worth 288,000 euros per year (3 incidents per month, 8,000 euros per incident, over twelve months), and its weight against the token-related items depends on your prices: with prices entered, it can dominate the total. These values are arbitrary: they come from no study. Replace them with your own estimates before presenting the result.

Revised on 2 October 2026: prices removed; token-related items show in tokens and calls, and in euros only with your prices.