$660.27 for a Prompt Style Guide: The Receipt

01 / The $660.27 receipt is mostly people, not tokens

A team prompt style guide should cost less than the confusion it removes. Before writing ten pages about tone, put a price on drafting, review, testing, and maintenance.

Here is an illustrative budget for a six-person team. Labor is valued at $60 per hour. That is an assumed loaded planning rate, not a salary claim or necessarily an additional cash expense. All API rates below are hypothetical arithmetic inputs, not CheapRelay prices or measured benchmarks.

• ■ Drafting: one owner × four hours × $60 = $240.
• ■ Review: six people × 30 minutes × $60 = $180.
• ■ Testing: two people × two hours × $60 = $240.
• ■ API testing: 100 requests at the token assumptions below = $0.27.
• ■ Total initial economic cost: $660.27.

The expensive mistake is not spending another cent on testing. It is scheduling another meeting without deciding what the guide must standardize.

02 / Buy six decisions, not a book of prompt slogans

Keep the first version short enough to use during actual work. Give the drafting owner six concrete decisions to document.

• ■ Task: start with an action and a deliverable, such as “Extract invoice fields into JSON.”
• ■ Context: separate source material from instructions using consistent, named delimiters.
• ■ Output: specify required fields, allowed values, length limits, and whether extra commentary is prohibited.
• ■ Uncertainty: define when to return null, flag missing evidence, or ask a question instead of guessing.
• ■ Examples: include one accepted output and one rejected output, with the reason for rejection.
• ■ Ownership: record the template owner, version, and test cases required before a change ships.

Replace vague rules such as “be precise” with observable requirements. “Return dates as YYYY-MM-DD; preserve an unknown date as null” gives reviewers something they can check.

03 / The $0.27 testing line, with the arithmetic exposed

Assume each of 100 test requests uses 1,500 input tokens and 300 output tokens. At illustrative rates of $1 per million input tokens and $4 per million output tokens, input costs $0.15 and output costs $0.12. Total: $0.27.

This simplified calculation assumes uncached input, no separate cache-write charges, no retries, and no tool charges. Replace those assumptions with the billing categories and actual rates applicable to your workload before approving a budget.

Spend the four testing person-hours on a small fixed set of real tasks. Include missing fields, contradictory instructions, unusually long source material, and content that looks like an instruction but belongs to the document. Compare the old and new templates against the same acceptance checklist. Record failures; do not declare a quality improvement from a handful of pleasing answers.

04 / A 300-token guide adds $0.30 per thousand calls

Suppose the existing request contains 1,200 input tokens and produces 300 output tokens. Across 1,000 calls, the same illustrative rates produce a $2.40 bill: $1.20 input plus $1.20 output.

Appending a 300-token guide to every request adds 300,000 input tokens, or $0.30. The revised bill is $2.70, assuming output length stays unchanged. Send only the rules needed for that task rather than automatically pasting the entire team handbook.

CheapRelay offers pay-as-you-go access with no monthly fee or subscription. Its real-token billing includes cache reads and writes, so keep those categories visible in your receipt rather than treating cached context as free. Sign up for CheapRelay to price a small pilot using your actual request mix.

05 / The break-even target is $299.70 a month

Now make the labor assumption explicit: six people each save three minutes per workday over 20 workdays. That equals six hours, valued at $360 per month. This is a target to validate, not a promised result.

Subtract one hour of monthly maintenance at $60 and the additional $0.30 token expense. Net modeled benefit: $299.70 per month. The $660.27 setup cost breaks even in approximately 2.2 months.

Track correction time before and after rollout. If the assumed savings do not appear, shorten the guide, repair the examples, or retire unused rules. A prompt standard earns its place through less rework, not more documentation.