Method
How AgentX calculates the numbers
If your auditor asks how a number was derived, this is the document to share. It explains the math behind the cash ledger, the capacity log, and the grade.
Overview
Two ledgers, two units
AgentX reports two things: cash saved (in dollars, sourced from your general ledger) and capacity created (in hours, sourced from your agents). The two are reported separately and never combined into one number.
Cash ledger
Cash saved — dollars, sourced from your GL
Each entry is one GL event the customer attests is AI-caused. Five categories: GL delta, cancelled subscription, avoided backfill, cut overtime, reduced contractor. The total is the headline number, with an attribution haircut applied so the methodology does not overclaim.
Headline formula: net = sum of entries × (1 − attribution haircut)
The haircut is reported as a band (low to high), not a point. The band is disclosed on every entry.
Capacity log
Capacity created — hours, never dollars at the headline
Each entry records hours absorbed by an agent, derived from metered outputs times a customer-stated baseline time per output. Capacity does not become cash through multiplication. It becomes cash only when the hours reconcile to a specific GL event.
Pre-deployment baseline
Two measurements before any cash is credited
Before any AI-attributed cash is recorded, the customer captures (a) a paired shadow agent run and (b) an independent 12-month sample. Both must be present and customer-signed. This is the contemporaneous control group the post-deployment attribution rests on.
Rate cards
Rates are derivations, not numbers
A rate card on AgentX is not a database row. It is a five-artifact derivation: the GL extract that produced the rate, the denominator definition, the allocation basis, the utilization assumption, and a countersignature. Rate cards expire on the budget cycle and must be re-signed.
Grade
Two measurements, never blended
The audit-trail grade has two measurements: count coverage (how much of the headline volume came from metered counts) and rate provenance (how much came from finance-signed rates). They are reported as a pair, never averaged into a single letter. The grade is a self-assessment until an independent party reviews it.
Regulatory
Three clocks the methodology tracks
EU AI Act (Aug 2026 provider, Aug 2027 deployer), SEC AI-washing posture, and external audit standards. Any number crossing from internal management metric into regulated representation is gated and disclosed. The audit pack surfaces both the recomputation inputs and the signature.
Size
Four modes by company size
The methodology presents differently depending on company size. At 50 people, the headline is founder-attested. At 5000+, the headline is recomputable from your own data and the attribution chain is peer-reviewed. The system detects size at registration and presents the right mode; an admin can override with a written rationale.
Modes:
Audit trail
What an auditor can verify
Ready to put this in place? Connect your first agent.