AGENT CREDITS · PRICING MODEL
Internal pricing model. Sign in with your AppZen Google account.
Open to anyone with an @appzen.com account.
Opens on the 9/18 decision: Expense margin 90%, template credit as the minimum, current credit as the maximum.
What this is. All 260 agents in TJ’s 9/16 sheet, across three products. The page opens on Anant’s 9/18 decision and matches the 9/18 repricing sheet cell for cell. Show TJ’s 9/16 sheet switches to TJ’s settings, where every figure matches his sheet and Hyram’s doc. Change any setting and everything recomputes.
Formula. Credits per agent run = ROUNDUP( standard cost / (1 - target margin) / price per credit ), at least 1. Standard cost prices a run’s own tokens at the Haiku 4.5 list whatever model served it: input $1.00, output $5.00, cache read $0.10, cache write $1.25 per 1M tokens. Actual cost never exceeds standard cost on any of the 260 agents (for Expense the two are equal), so the MAX(actual, standard) in Hyram’s doc always comes out as the standard cost. Every run is rounded up on its own, and every average on this page is an average of those rounded credits.
The 9/18 decision. Expense margin goes from 70% to 90%; AP stays at 60% and Inbox at 80%. Template credit as a floor: a standard agent is never charged less than its template credit. Current credit as a maximum: an existing automation is never charged more than today, unless that would leave its margin below 60%. The minimum is applied last, so it wins when the two conflict.
The table. Four groups: TJ’s 9/16 sheet (his figures, fixed at his margins), the template, the new credit at this setting, and the SOP match score. Hover a heading for the column’s name in the 9/18 sheet; Export CSV uses those names. Margins are measured against today’s actual cost. 23 rows each cover several versions of one automation that bill different credits today; Current credit per run is TJ’s single figure for the row.
Standard agents. A matching name is not enough. Each agent’s current SOP was compared word for word with the 16 published templates that carry a credit and 9 earlier standard SOPs still in use. An agent counts as standard only if it keeps at least 80% of the template’s SOP and is no more than twice as long. 34 of 260 qualify (AP 25, Inbox 9). Template credits were read from prod on 21 September: Extraction 3, PO Matching 3, GL Coding 3, Duplicate Resolution 1, Statement Reconciliation 7, Payment Status 3. No Expense template carries a credit.
Per agent run, and per transaction. A transaction is one AppZen invoice for AP, one expense report for Expense, and one email thread for Inbox, where agents run again on each new message. Most transactions run more than one agent, so credits per invoice adds up every agent run on it. Credits per agent run is averaged over real runs, so credits per invoice is that figure times runs per invoice. The Averages control changes only the average margin tile: each agent counts once, as in the sheets, or weighted by runs.
The window. Runs and transactions are from 8 to 13 September 2026, the window that reproduces every cost column in TJ’s sheet (260 of 260 agents), Active customers only. Invoices with an agent counts invoices submitted in that window that had at least one agent run. 13,058 of 186,763 agent runs in the window bill nothing, so cost is incurred on more runs than credits are charged on.