Handbook calculators

Sum latency across retrieval, inference, tool calls, and post-processing.

Calculate monthly AI cost from request volume and unit cost.

Calculate weighted utility for quality, cost, and latency. Adjust weights to match strategy.

Calculate a weighted four-axis evaluation score. Each input is 0-100.

Calculate an experiment decision score. Positive scores suggest rollout; negative scores suggest rollback or iteration.

Annual value of saved time converted to labor cost, divided by build plus annual operating cost. 0% breaks even; positive means it is worth the investment.

Rate four signals from 0-100 each and take the weighted sum. Thresholds are yours to set, e.g. 70+ is Active and below 40 is Risk. Tune the weights to your business.

Remove customers acquired during the period to see how many existing customers stayed. Churn = 100 - retention.

Customer lifetime value divided by acquisition cost. Roughly 3x or higher is healthy; below 1x is a loss-making model.

How much existing-customer revenue is retained and grown after expansion, contraction, and churn. Above 100% means you grow without new logos.

The share that moved from one pipeline stage to the next. The starting point for finding where deals stall.

Combine fit, intent, value, and risk into a simple prioritization score.

Score each candidate tool 1-5 on five criteria and take the weighted average. Treat permissions and operating cost as pass/fail gates, not scores. Tune weights to your priorities.

Estimate monthly build minutes saved from cache hit rate and build volume.

Estimate how many review-wait days OTA updates save in a quarter.

Estimate download time from bundle size and network speed.

Calculate dropped frame percentage from profiling results. Target: 5% or lower.

Calculate net revenue retention from starting ARR, expansion, contraction, and churn.

Calculate account priority by separating fit, intent, readiness, and expansion. Scores above 4.0 are candidates for AE intervention.

Estimate ROI for an AI GTM pilot. Pipeline impact is adjusted by expected close rate and gross margin.

Estimate conservative net annual value after a successful POC. It combines time savings, revenue uplift, and risk reduction, then subtracts annual product cost and implementation cost.

Calculate gross margin after monthly AI execution cost and service cost.

Score whether a signal is fresh, specific, reliable, actionable, and connected to feedback. Scores above 4.0 deserve fast intervention.

Score revenue impact, data readiness, adoption fit, risk control, and learning speed from 1 to 5 to select pilot candidates.

The ratio includes purchases without wishlists. Do not multiply it by demo-to-wishlist conversion. Applying lifetime refunds to first-week sales is approximate.

Defaults use Shopify medians. Checkout-start rate is calculated; enter your own AOV.

Attribution-based, not incremental. Actual profit is lower if existing-customer purchases are included.

Use this for a quick point estimate. The default multiplier of 1 gives unadjusted attributed conversions. No public search multiplier is available; enter your experimental estimate. Use the simulator above for intervals.

Assumes a concave power curve. Estimate b from internal experiments; there is no public standard value.

No public incremental-share estimate was found. Treat it as a scenario until an on/off experiment. At K of 1 or above, the sum diverges and no result is displayed.

Defaults use RevenueCat's health category. Direct payments without a trial are excluded.

The 0.016% meeting default is a lower-bound calculation per message sent from Belkins. The 18% win rate is opportunity-based and only a weak proxy for meeting-to-win. Replace with your own values.

Enter visit-to-trial and trial-to-paid rates for two paths to compare paid customers per 1,000 visits. Defaults compare ChartMogul's card-required trials with all trials. Replace with your own rates.

Time units follow p and q. Defaults are illustrative secondary citations (D); replace them with comparable-product estimates. If q is no greater than p, adoption declines from the start without a later peak.

A rough normal approximation. Lower k further if the sample conditions differ from yours.

Combine the prior mean and concentration with pilot results to calculate the posterior mean.