Skip to content
StartupCFO logoStartupCFO.AI
All industries

Metric Tree

AI Software Metric Tree

ARR like SaaS, with inference cost inside COGS and the GPU and model-routing drivers that move it. 111 metrics, each with its definition and formula, mapped from the north star Annual Recurring Revenue, the profitability root Operating Income, and the survival metric Cash Runway down to the operating drivers you can act on.

Growth & revenueProfit & efficiencyCash & capitalOperating drivers/driver pushes up or downappears elsewhere in the tree

Growth trunk

What you are building: the north-star metric and what it is made of.

Profitability trunk

Whether the business earns money: operating income, gross profit less operating expenses.

Survival trunk

How long you can keep building: runway and the cash, burn and revenue behind it.

Supporting metrics

Also part of the AI Software tree in ClariFi: the ratios and diagnostics you read alongside the trunks.

Growth
Retention
Customers
Subscription
Contracts
Pipeline
Profitability
Unit economics
Efficiency
Workforce
Expenses
Working capital
Cash

All 111 metrics in the AI Software tree

Definitions and formulas as ClariFi uses them. Metric names in formulas refer to other entries on this page.

Growth and revenue

44 metrics

Contracts

Contracted ARR
Annual value of the customer contracts in force at the month end, from the contract register. A renewal is never counted twice. Shown with its coverage of MRR × 12 (or trailing-12-month revenue).
Customer Concentration Index (HHI)
Herfindahl-Hirschman index of customers' shares of contracted annual value.
Remaining Performance Obligations
Contracted revenue not yet recognised, from the contract register. Open-ended contracts are reported separately.
Revenue Concentration
The largest customer's share of contracted annual value, customers grouped by normalised name.

Customers

Churned Customers
Customers lost in the month.
Customer CMGR
Compound monthly growth rate of customers over the last twelve months.
Formula: pow(Total Customers ÷ Total Customers (12 months ago), 1 ÷ 12) − 1
Customer Growth Rate
Month-over-month change in total customers.
Formula: Total Customers ÷ Total Customers (last month) − 1
Gross Customer Churn Rate
Customers lost in the month as a share of the customers at the start of it. Where customers are not under contract, the company defines what counts as a churned customer and the figure is recorded directly.
Formula: Churned Customers ÷ Total Customers (last month)
Net Customers
New customers less churned customers.
Formula: New Customers − Churned Customers
New Customers
Customers acquired in the month. Reactivated customers are not new: counting them would understate CAC.
Retained Customers
Customers at the start of the month less those who churned.
Formula: Total Customers (last month) − Churned Customers
Total Customers
Paying customers at the end of the month. For a marketplace, demand-side active users only.

Growth

MRR CMGR
Compound monthly growth rate of Total MRR over the last twelve months.
Formula: pow(Total MRR ÷ Total MRR (12 months ago), 1 ÷ 12) − 1
MRR Growth Rate
Month-over-month change in Total MRR.
Formula: Total MRR ÷ Total MRR (last month) − 1
Revenue Growth Rate (MoM)
Month-over-month change in net revenue. Over a period it stays a MONTHLY rate, compounded, so the monthly bands still apply.
Formula: Net Revenue ÷ Net Revenue (last month) − 1
Revenue Growth Rate (YoY)
Change in net revenue against the same month a year earlier. Seasonality cancels.
Formula: Net Revenue ÷ Net Revenue (12 months ago) − 1

Pipeline

Average ACV
Average annual contract value of valued, active contracts.
Sales Cycle
Average days from opening a deal to closing it, over deals closed in the window.
Weighted Pipeline
Pipeline deals' annual value × their win probability; a deal with no probability contributes zero.
Win Rate
Closed deals won as a share of deals closed in the window, renewals included.

Pricing

Annual Plan Discount
Discount offered for paying annually instead of monthly.
Annual Plan Mix
Share of customers on annual plans.
Average Revenue per Customer
Average monthly revenue per customer, as the client records it. (Fintech: revenue per account per month; Technology: average revenue per customer per month.)
Inference Units
Inference served in the month, in the unit the company meters and prices: millions of tokens, requests, minutes of audio. The divisor of inference cost per unit.

Retention

Cohort Net Dollar Retention (TTM)
Contract value in force at this month end of customers whose contracts were in force twelve months earlier, over their value then. Biased downward: an expansion recorded by editing a contract changes both ends of the year and is not counted, while churn always is. No benchmark band until register history exists.
Cohort Retention
Retention by acquisition cohort from the client's cohort tables, each cohort divided by its own base.
Expansion MRR Rate
Expansion MRR as a share of last month's MRR.
Formula: Expansion MRR ÷ Total MRR (last month)
Gross Dollar Retention (TTM)
Trailing-12-month gross dollar retention: the twelve-month-old MRR kept after contraction and churn, excluding expansion. Floored at zero and flagged outside 0-100%.
Formula: max(Total MRR (12 months ago) − Contraction MRR summed over the last 12 months − Churned MRR summed over the last 12 months, 0) ÷ Total MRR (12 months ago)
Gross Revenue Churn
MRR lost to contraction and cancellation in the month.
Formula: Contraction MRR + Churned MRR
Gross Revenue Churn Rate
Gross revenue churn as a share of last month's MRR.
Formula: (Contraction MRR + Churned MRR) ÷ Total MRR (last month)
Monthly Gross Dollar Retention
The share of last month's MRR kept this month after contraction and churn, excluding expansion.
Formula: (Total MRR (last month) − Contraction MRR − Churned MRR) ÷ Total MRR (last month)
Monthly Net Dollar Retention
The share of last month's MRR kept this month after expansion, contraction and churn. A monthly figure, never compared with an annual one or with a benchmark.
Formula: (Total MRR (last month) + Expansion MRR − Contraction MRR − Churned MRR) ÷ Total MRR (last month)
Net Dollar Retention (TTM)
Trailing-12-month net dollar retention. When ClariFi calculates it, it is approximated from twelve months of MRR movements, which also counts customers acquired during the year, so it is not compared with a benchmark; a cohort figure the company records itself is compared with the benchmark.
Formula: (Total MRR (12 months ago) + Expansion MRR summed over the last 12 months − Contraction MRR summed over the last 12 months − Churned MRR summed over the last 12 months) ÷ Total MRR (12 months ago)
Net Revenue Churn
MRR lost to contraction and cancellation, less expansion MRR.
Formula: Contraction MRR + Churned MRR − Expansion MRR
Net Revenue Churn Rate
Net revenue churn as a share of last month's MRR.
Formula: (Contraction MRR + Churned MRR − Expansion MRR) ÷ Total MRR (last month)

Revenue

Net Revenue
Revenue recognised in the month from every source, recurring and one-time, net of credits and refunds booked as contra-revenue.

Subscription

Annual Recurring Revenue
Total MRR annualised (× 12). A company that reports contracted ARR instead can have that figure recorded directly; contracted ARR from the customer contract register is shown separately.
Formula: Total MRR × 12
Churned MRR
MRR lost from customers who cancelled entirely.
Contraction MRR
MRR lost from existing customers downgrading or using less.
Expansion MRR
MRR gained from existing customers upgrading, adding seats or using more.
Net New MRR
New plus expansion MRR, less contraction and churned MRR. All four movements are required: an unrecorded churn is not a zero one.
Formula: New MRR + Expansion MRR − Contraction MRR − Churned MRR
New MRR
MRR from customers acquired in the month.
Retained MRR
Last month's MRR less contraction and churned MRR.
Formula: Total MRR (last month) − Contraction MRR − Churned MRR
Total MRR
The recurring portion of monthly revenue from active subscriptions, measured at the END of the month.

Profit and efficiency

45 metrics

Efficiency

Burn Multiple
The quarter's net burn over the ARR added in the quarter. Subscription clients only; needs four months.
Formula: (Total Expenses summed over the last 3 months − Net Revenue summed over the last 3 months) ÷ ((Total MRR − Total MRR (3 months ago)) × 12)
Magic Number
Net new MRR over the trailing three months, annualised, divided by the previous quarter's sales and marketing spend. Subscription clients only.
Formula: Net New MRR summed over the last 3 months × 12 ÷ sum(Sales and Marketing Spend (3 months ago), 3)
New MRR Acquisition Cost
Sales and marketing spend per dollar of new MRR.
Formula: Sales and Marketing Spend ÷ New MRR
Revenue Burn Multiple
A year's net burn over the annual run-rate revenue added over the year (year-on-year, so seasonality cancels); quarter-on-quarter, badged, with fewer than 15 months.
Formula: (Total Expenses summed over the last 12 months − Net Revenue summed over the last 12 months) ÷ ((Net Revenue summed over the last 3 months − sum(Net Revenue (12 months ago), 3)) × 4)
Revenue Quick Ratio
MRR gained (new plus expansion) for every dollar of MRR lost (contraction plus churn).
Formula: (New MRR + Expansion MRR) ÷ (Contraction MRR + Churned MRR)
Rule of 40
Annualised revenue growth plus net margin, in points.
Formula: pow(1 + Revenue Growth Rate (MoM), 12) − 1 + Net Margin

Expenses

COGS
Direct costs of delivering the product or service: hosting, infrastructure, support and payment processing. A ledger's Cost of Revenue line maps here; sales and marketing are operating expenses, not COGS.
G&A Expenses
General and administrative costs: rent, legal, accounting, HR and office.
Inference Cost
Cost of running models to serve customers in the month: model-API usage, GPU and serving infrastructure, and retrieval or vector-database costs. A component of COGS, recorded beside it, the COGS figure already includes it, so it is never added to COGS a second time. The part of COGS that grows with usage.
Inference Share of COGS
Inference cost as a share of COGS. The higher it is, the more gross margin moves with usage rather than with headcount.
Formula: Inference Cost ÷ COGS
Model Training Cost
Cash spent in the month on training or fine-tuning models: compute for training runs, data acquisition and labelling. Lumpy, a model trained once a year lands in one or two months, so the tree shows a trailing twelve-month average beside it; record 0 in a month with no training rather than leaving it blank. A component of operating expenses (usually R&D), recorded beside them and never added a second time. Whether the ledger expenses or capitalises training is the accountant's call; this column records what the ledger says.
Model Training Cost (12-month average)
Model training cost averaged over the trailing twelve months, what a model trained once a year costs per month in steady state. Blank until twelve months of training cost are recorded (a month with no training is recorded as 0, not left blank).
Formula: Model Training Cost summed over the last 12 months ÷ 12
Operating Expenses
Costs other than COGS: salaries, marketing, rent, software and professional services.
Other COGS
COGS other than inference: hosting, support, payment processing and third-party services. The remainder of COGS after inference cost, so the two always add up to COGS.
Formula: COGS − Inference Cost
Other Operating Expenses
Operating expenses not recorded as R&D, sales and marketing or G&A: the remainder after the three departments, so the four always add up to operating expenses.
Formula: Operating Expenses − R&D Expenses − Sales and Marketing Spend − G&A Expenses
R&D excl. Model Training
R&D other than model training: the remainder after training cost, so the two always add up to R&D.
Formula: R&D Expenses − Model Training Cost
R&D Expenses
Research and development costs: engineering salaries, tools and infrastructure.
Sales and Marketing Spend
Cost of the sales team, marketing campaigns, advertising and other customer acquisition.
Total Expenses
Every cost of the month, operating and non-operating. ClariFi computes it from COGS, operating expenses and the source's own below-the-line amount unless a person entered it.

Profitability

EBITDA
Earnings before interest, taxes, depreciation and amortisation: net income plus D&A, interest expense and income tax, less interest income, built from identified ledger lines.
EBITDA Margin
EBITDA as a share of net revenue.
Formula: EBITDA ÷ Net Revenue
Expense Ratio
Total expenses as a share of net revenue.
Formula: Total Expenses ÷ Net Revenue
Fully Loaded Gross Margin
Gross margin after charging the trailing twelve-month average of model training cost against the month: what gross margin looks like once the models that serve the revenue are paid for. Below Gross Margin by the training charge; blank until twelve months of training cost are recorded. Not a benchmark figure, peers report gross margin before training.
Formula: (Net Revenue − COGS − Model Training Cost summed over the last 12 months ÷ 12) ÷ Net Revenue
Gross Margin
Gross profit as a share of net revenue. For a marketplace that reports gross merchandise value (GMV) as its revenue, this is the margin on GMV and is not compared with a benchmark.
Formula: (Net Revenue − COGS) ÷ Net Revenue
Gross Profit
Net revenue less COGS.
Formula: Net Revenue − COGS
Net Income
Net revenue less total expenses.
Formula: Net Revenue − Total Expenses
Net Margin
Net income (net revenue less total expenses) as a share of net revenue.
Formula: (Net Revenue − Total Expenses) ÷ Net Revenue
Operating Income
Net revenue less COGS and operating expenses: what the business earns from operating before depreciation is added back (EBITDA) and before interest, tax and other items below the operating line (net income). The root of the profitability trunk.
Formula: Gross Profit − Operating Expenses
Operating Margin
Operating income as a share of net revenue.
Formula: Operating Income ÷ Net Revenue

Unit economics

Average Revenue Per Account
End-of-month Total MRR divided by end-of-month customers.
Formula: Total MRR ÷ Total Customers
CAC
Sales and marketing spend per new customer over the trailing three months. With spend and no new customers it is not meaningful and is shown as such, never as a figure.
Formula: Sales and Marketing Spend summed over the last 3 months ÷ New Customers summed over the last 3 months
CAC (same month)
Same-period CAC: this month's sales and marketing spend per new customer this month. Shown beside ClariFi's trailing-three-month CAC.
Formula: Sales and Marketing Spend ÷ New Customers
Cost to Serve
COGS per customer, over the average customers of the month.
Formula: COGS ÷ ((Total Customers (last month) + Total Customers) ÷ 2)
Customer Lifetime (months)
1 ÷ the trailing-12-month churn rate, capped at 60 months; blank with fewer than 3 churned customers in the year.
Formula: min(1 ÷ (Churned Customers summed over the last 12 months ÷ sum(Total Customers (last month), 12)), 60)
Gross Margin (trailing 12 months)
Gross profit over the trailing twelve months as a share of revenue over the same months, used by LTV and Payback; the month's own gross margin, badged, when fewer months exist.
Formula: Gross Profit summed over the last 12 months ÷ Net Revenue summed over the last 12 months
Inference Cost per Unit
Inference cost per unit served, in the unit the company meters (a million tokens, a request, a minute). The figure model routing, caching and GPU utilisation move; what the customer pays per unit must stay above it.
Formula: Inference Cost ÷ Inference Units
Inference Units per Customer
Inference units served per customer in the month, over the average customers of the month. Rising usage per customer precedes expansion revenue, and inference cost.
Formula: Inference Units ÷ ((Total Customers (last month) + Total Customers) ÷ 2)
LTV
Gross-margin lifetime value per customer: monthly revenue per customer × trailing-12-month gross margin × lifetime, where lifetime is 1 ÷ the trailing-12-month churn rate, capped at 60 months and blank with fewer than 3 churned customers in the year.
Formula: Monthly Revenue per Customer × Gross Margin (trailing 12 months) × Customer Lifetime (months)
LTV:CAC Ratio
Gross-margin LTV divided by CAC: how many times a customer repays what it cost to acquire.
Formula: LTV ÷ CAC
Monthly Revenue per Customer
What an acquired customer pays each month, used by Payback and LTV: ARPA for subscription clients; net revenue over the month's average customers for contractual clients (marketplaces: per demand-side user, GMV on the principal basis); not defined for charities, biotechs, or clients whose customers are not contractual without a confirmed churn definition.
Formula: Average Revenue Per Account
MRR per New Customer
New MRR divided by new customers.
Formula: New MRR ÷ New Customers
Payback Period
Months of gross profit per customer needed to repay CAC: CAC ÷ (monthly revenue per customer × trailing-12-month gross margin).
Formula: CAC ÷ (Monthly Revenue per Customer × Gross Margin (trailing 12 months))
Revenue Payback
CAC ÷ monthly revenue per customer: payback on revenue rather than gross profit. Shown beside Payback Period, never stored or targeted.
Formula: CAC ÷ Monthly Revenue per Customer

Workforce

Headcount
Full-time-equivalent employees at month end.
Revenue per Employee
Net revenue in the month divided by headcount.
Formula: Net Revenue ÷ Headcount

Cash and capital

18 metrics

Cash

Cash Balance
Cash and cash equivalents at the end of the month. May be negative (overdrawn).
Cash Runway
Months of cash left: the latest balance, projected forward with the burns recorded since, divided by the larger of the average net burn of the months that burned among the last three (the last four when fewer than two of the last three burned) and the net burn of the most recent month that burned; when fewer than two of the last four months burned, divided by the plain three-month average net burn. A month with no recorded figures counts as burning, at no less than the largest net burn recorded in the last twelve months (more when the balances on either side imply more). Blank when not burning, when the latest balance is more than three months old, or when the projected cash ran out before this month; shown as 120 months when longer; "Out of cash" when the latest recorded balance and the cash projected from it are both zero or negative.
Months of Cash
Cash balance ÷ the average monthly total expenses over up to twelve months (at least three recorded). Includes donor-restricted cash; never blank for a break-even organisation.
Net Burn
Net cash consumed by operations in the month: total expenses less net revenue, floored at zero. Over a period it is rebuilt from the components and shown as a monthly figure.
Formula: max(Total Expenses − Net Revenue, 0)

Working capital

AR 31-60 Days
Receivables overdue by 31-60 days.
AR 61-90 Days
Receivables overdue by 61-90 days.
AR Current (0-30 days)
Receivables within payment terms (0-30 days).
AR Over 30 Days %
Receivables more than 30 days overdue as a share of all receivables.
Formula: (AR 31-60 Days + AR 61-90 Days + AR Over 90 Days) ÷ Total Accounts Receivable
AR Over 90 Days
Receivables overdue by more than 90 days.
AR Over 90 Days %
Receivables more than 90 days overdue as a share of all receivables.
Formula: AR Over 90 Days ÷ Total Accounts Receivable
Cash Conversion Cycle
Days sales outstanding less days payable outstanding.
Formula: Days Sales Outstanding − Days Payable Outstanding
Cash Ratio
Cash ÷ accounts payable.
Formula: Cash Balance ÷ Total Accounts Payable
Current Ratio
(Cash + accounts receivable) ÷ accounts payable. ClariFi records no inventory or other current items, so today this is the same formula as the Quick Ratio.
Formula: (Cash Balance + Total Accounts Receivable) ÷ Total Accounts Payable
Days Payable Outstanding
Average days taken to pay bills: accounts payable ÷ COGS × 30.
Formula: Total Accounts Payable ÷ COGS × 30
Days Sales Outstanding
Average days to collect payment: accounts receivable ÷ net revenue × 30.
Formula: Total Accounts Receivable ÷ Net Revenue × 30
Quick Ratio
(Cash + accounts receivable) ÷ accounts payable. Identical to the Current Ratio until inventory is recorded.
Formula: (Cash Balance + Total Accounts Receivable) ÷ Total Accounts Payable
Total Accounts Payable
Bills the company owes vendors and suppliers, at month end.
Total Accounts Receivable
Invoices owed to the company and not yet paid, at month end.

Operating drivers

4 metrics

Demand

Qualified Pipeline
Total annual contract value of open, sales-qualified opportunities at month end.

Operations

GPU Utilisation
Share of reserved or provisioned GPU capacity actually used for inference during the month. Higher utilisation lowers inference cost per unit; a rise precedes a fall in inference cost.
Small-Model Share of Requests
Share of requests served by the smallest model that meets the quality bar. Routing more traffic to smaller models lowers inference cost; a rise precedes a fall in inference cost.

Product

Output Quality Score
Share of sampled model outputs rated acceptable against the company's own evaluation set during the month. Falling quality precedes churn and contraction.

Questions about the AI Software tree

What is the north star metric for a AI Software company?
In ClariFi's default AI Software tree the growth trunk is rooted at Annual Recurring Revenue (subscription revenue; the other revenue model uses a different root). Total MRR annualised (× 12). A company that reports contracted ARR instead can have that figure recorded directly; contracted ARR from the customer contract register is shown separately. The profitability trunk is rooted at Operating Income: gross profit less operating expenses, whether the business earns money. The survival trunk is rooted at Cash Runway, which tells you how long you can keep building.
What metrics should a AI Software startup track?
This tree defines 111 metrics for AI Software: the three trunks (Annual Recurring Revenue, Operating Income and Cash Runway) with the formula components and operating drivers under them, plus supporting ratios such as margins, retention, unit economics and working capital. Start with the roots and the first level beneath each, then add drivers as you can measure them.
What is the difference between a component and a driver?
A component is part of a metric's formula, so the parent is calculated from it exactly (gross profit is revenue minus COGS). A driver is an operating metric that tends to move the parent but is not part of its formula, for example pipeline or activation rate. Drivers show direction, not a guaranteed effect.
How is Annual Recurring Revenue calculated?
Annual Recurring Revenue = Total MRR × 12.

Source: ClariFi metric catalogue, synced 2026-10-13 (catalogue version 75d1d2c5).

Go deeper

Service

Fractional CFO for startups

Board packs, forecasts, and fundraising support without a full-time hire.

Free tool

Burn Rate Calculator

Gross burn, net burn, and burn multiple from your monthly cash flows.

Guide

Finance for Marketplace and GMV Businesses

A finance playbook for two-sided platforms covering take rate, gross versus net revenue, liquidity, and unit economics.

Insight

What Is a Metric Tree? How One Number Breaks Down Into the Levers That Move It

A KPI list can tell you a number missed. A metric tree tells you why: one outcome at the top, broken into the metrics that produce it, down to the levers you can pull.

Glossary

ARR (Annual Recurring Revenue)

Annualized value of your subscription revenue at a point in time.

Free tool

Runway Calculator

See how many months of runway your cash and burn give you.