Savings calculator worksheet

Estimate what your team is burning on AI review noise.

Your AI code review tool produces a lot of comments. A large fraction of those comments are false positives your engineers dismiss, and you are paying for every one of them twice. Once on the vendor invoice, and again in the attention cost of the senior engineer who had to read the comment before ignoring it. This worksheet measures both numbers.

Fill in the inputs with your numbers. The math is transparent so you can verify every step. The output is a range, not a point estimate, because the real number depends on your codebase and your team.

Seven inputs, four outputs, ten minutes. Every default is based on published industry figures or conservative midpoints. Override any default where you have better data. The result updates live as you type.

Your inputs

Count of engineers whose pull requests go through the AI review tool you are currently using.
Default: 4. Industry median for teams on trunk-based or short-lived-branch workflows. Long-lived feature branches tend to be 1 to 2 per engineer per week.
Default: 100. Many teams run their AI reviewer on every pull request automatically. Adjust downward if yours only runs on large changes.
Default: 40. Midpoint of published false-positive rates for Copilot review, Cursor, and Greptile on adversarial benchmark suites. Use your own number if you have internal data.
Default: 10. Covers reading the concern, checking the code, validating, deciding what to do with the comment, and context-switching back.
Default: $95. Senior engineer in North America, base $150k plus benefits and overhead, divided by 2,000 working hours.
Default: 8. Nocturne's measured rate on the same adversarial benchmark suite used for Input 4. Observed range is 5 to 12. The actual rate on your codebase is determined in the 30-day pilot.

Your outputs

Output 1. Total PRs reviewed by AI per year 10,000
Output 2. Annual false positives on current tool 4,000
Output 3. Current annual cost of FP investigations $63,333
Output 4. Projected annual savings with Nocturne
$50,667
Recommended tier: Standard at $2,000/month flat base. At projected savings of $50,667, the performance share is $10,133. Net annual savings to the client, after the Nocturne fee, is $16,533 (monthly base) or $20,133 (annual prepay).

How the math works

Output 1: Total pull requests reviewed by AI per year

Formula: engineers × PRs per engineer per week × 50 weeks × AI review coverage percent

Example with defaults and a 50-engineer team: 50 × 4 × 50 × 100% = 10,000 PRs/year

Output 2: Annual false-positive investigations on your current tool

Formula: total PRs per year × current tool FP rate

Example with defaults: 10,000 × 40% = 4,000 false positives per year

Output 3: Current annual cost of false-positive investigations

Formula: false positives per year × minutes per investigation / 60 × loaded hourly rate

Example with defaults: 4,000 × 10 / 60 × $95 = $63,333 per year. For a 200-engineer team, the number scales to roughly $253,000 per year. For a 500-engineer team, roughly $633,000 per year.

Output 4: Projected annual savings from switching to Nocturne

Formula: (current FP rate − Nocturne FP rate) × total PRs × minutes per investigation / 60 × loaded hourly rate

Example with defaults: (40% − 8%) × 10,000 × 10 / 60 × $95 = $50,667 per year saved. This is the number the 30-day pilot would measure directly on your actual pull requests. If the measured number comes out higher, your savings are higher. If it comes out lower, you pay the pilot fee and walk away with data.

Which Nocturne tier fits your team

Nocturne pricing is flat base fee plus performance share. Every tier charges a predictable monthly base, plus 20 percent of the measured monthly savings the service actually delivers. When Nocturne does not save you money, you only pay the base. When Nocturne saves you a lot, both sides win.

Tier Flat base fee Performance share Team size
Pilot (30 days) $2,500 one-time None Any size
Standard $2,000/month 20% of measured savings Up to 50 engineers
Growth $4,500/month 20% of measured savings 51 to 200 engineers
Enterprise $9,000/month 20% of measured savings 200+ engineers

Annual prepay: 15% off flat base fee only. The performance share is unchanged.

Decision matrix

Projected annual savings (Output 4) Recommended tier Total annual cost (monthly base) Net savings to client
Less than $25,000 Not a fit yet N/A Use the open-source version directly, no managed service
$25,000 to $100,000 (up to 50 engineers) Standard ($2,000/month) $29,000 to $44,000 Client keeps roughly 0% to 71% of savings
$100,000 to $300,000 (51 to 200 engineers) Growth ($4,500/month) $74,000 to $114,000 Client keeps roughly 26% to 62% of savings
$300,000 to $1,000,000 (51 to 200 engineers) Growth ($4,500/month) $114,000 to $254,000 Client keeps roughly 62% to 75% of savings
$1,000,000 to $3,000,000 (200+ engineers) Enterprise ($9,000/month) $308,000 to $708,000 Client keeps roughly 69% to 76% of savings
$3,000,000 and above (200+ engineers) Enterprise ($9,000/month) $708,000 and up Client keeps roughly 76% or more of savings

Month-1 pilot: Every engagement starts with a 30-day pilot at a flat $2,500 fee. At the end of month 1, you receive a written savings report with the measured false-positive reduction on your actual pull requests. If the measured savings come in below an agreed threshold, you exit at the end of month 1 with no further obligation and you keep the report. If the measured savings clear the threshold, the engagement converts to one of the tiers above.

Reality check

This worksheet is a back-of-the-envelope estimator, not a guaranteed savings figure. The real number comes from the 30-day phase-one pilot, which measures actual false-positive rates on your actual pull requests. The worksheet exists so you can sanity-check whether a pilot is worth your time before scheduling an intro call.

Things the worksheet does not capture (and that the real pilot does)

Things the worksheet intentionally excludes

These are real, but quantifying them is unreliable and including them in the pitch is how vendors lose credibility. The worksheet stays on hard dollars only.

How to use this worksheet in a conversation

Fill in your numbers. Save a copy. Bring it to the thirty-minute intro call. The call starts with your worksheet on the screen and we walk through each input together, adjusting where your data differs from the defaults. By the end of the call, we both know whether a pilot is worth running.

If the answer is no, the call ends there and costs you nothing. If the answer is yes, we schedule the pilot kickoff for the following week.

Ready to run the numbers against real PR volume?

Start at nocturnehq.com. We respond to every well-formed inquiry within three business days.

Worksheet version 2.0. Updated 2026-04-13. Nocturne is a GPOut Labs product.