Working tool · Agent consolidation

The Agent Cost Model

Model what a centralized GTM agent actually costs to run: tokens, platform and upkeep, in plain language a CFO or controller can gut-check. Pick a use case from Where to point an agent, adjust the sliders to your team, and export the model as a PDF or spreadsheet. Companion to the Agent Consolidation field note.

Showing demo data. Edit any input to make it yours.

Inputs

Every field has a plain-speech explanation. A "run" is one completed task: one account researched, one record routed, one battlecard refreshed, one health score written.

Use case

Picking a use case pre-fills realistic defaults below. Everything stays editable. Choose Custom to model a workflow of your own.

People using the agent

Seats, not the whole company.

Runs per person per workday

Tasks the agent completes for each person daily.

Workdays per month

Model tier

July 2026 list prices load below and stay editable.

Model price per 1M input tokens ($)

Input = everything you send the model (instructions, context, data).

Model price per 1M output tokens ($)

Output = what the model writes back. Priced higher everywhere.

Input tokens per run

Rough guide: 1,000 tokens is about 750 words.

Output tokens per run

Share of input that is cached: 60%

Centralized agents reuse the same instructions and company context on every run. Cached input is billed at roughly a 90% discount, a structural saving fragmented agents never get.

Batch processing (50% off all tokens)

Applies when runs do not need instant answers, like overnight enrichment or scoring.

Platform & tooling, $ per month

Clay, enrichment credits, hosting, or the vendor seat under the agent.

Upkeep hours & rate

hrs/mo at $/hr

The GTM engineer time that keeps prompts, data and evals sharp.

Settings: bring your own key (optional)

Add an Anthropic API key to unlock the two Claude features on the right. The key stays in this browser tab only; it is never stored or sent anywhere except directly to Anthropic.

The model

Live totals from your inputs. Nothing hidden: the spreadsheet export shows every formula in words.

All-in monthly cost

Runs / month

Tokens / month

LLM cost / run

Cost / user / month

LLM share

Minutes of $35/hr time per run

LLM tokens

Platform

Upkeep

The spreadsheet opens in Excel or uploads straight to Google Sheets, with a Read Me tab explaining every line. The PDF button opens your print dialog: choose "Save as PDF".

Ask Claude (bring your own key)

Two helpers for the parts people guess at.

Or describe a workflow and estimate its tokens

How this model works, in one breath

Monthly runs are people times runs per day times workdays. Each run consumes input tokens (billed at your input price, with the cached share billed at a 90% discount) and output tokens (billed at your output price). Batch mode halves both. Add the platform under the agent and the upkeep hours that keep it sharp, and you have the all-in number. Token prices default to July 2026 list rates for frontier ($5 in / $25 out per million), workhorse ($3 / $15) and efficient ($1 / $5) tiers, and every one of them is editable because your negotiated reality beats my defaults.

What this deliberately excludes: the one-time build cost (that is a project, not a run rate) and the revenue side of ROI, which needs a holdout, not a calculator. Both are covered in the field note.