The AI workforce platform for the back office
RoleAxis scores every back-office role task by task, staffs the tasks that should move to an agent, and holds every action for a person's signature.
Three minutes, no card, no account. Or send one month of exports and get back a priced report of what is wrong in it.
Task by task: keep it human, help the human, or hand it to an agent. The same answer every time, from a deterministic engine.
One agent for the tasks handed over. It reports to the person who holds the role, not the other way round.
Every draft waits for a person. Above a threshold, for the owner. The agent never pays anyone or posts a journal.
Every finding carries an amount. The month report counts only what was signed, from a log nobody can edit.
01 · Decide
Bills, tills, stock, payroll, orders. Each leak looks fine on its own line, and nobody has the week to check every line. The assessment maps one role in three minutes and says, for each task, whether a person should keep it, get help with it, or hand it to an agent.
Name a role, list what it actually does, answer seven plain questions per task. You get a verdict for the role, a reason per task, a hiring profile for what stays human, an enablement plan for what gets help, the matching agent for what moves, and a 30/60/90-day plan. Where no agent fits yet, it says so, and that role goes on the build list.
| Matching supplier bills to orders and receipts | Agent |
| Checking tax on bills and invoices | Agent |
| Chasing overdue customer invoices | Help the human |
| Negotiating terms with a key supplier | Keep human |
Tasks marked "agent" become the agent's job description. Tasks marked "keep human" stay with the person, and the platform never drafts them. The split is written down, and the owner can change it at any time.
02 · Staff
One agent per role, reporting to the person who holds it. The results below are what an agent produces every month; the skills behind them are on each agent's page.
Wrongly taxed, duplicated or unmatched bills are held before the payment run, with the evidence attached. Reminders and supplier queries arrive drafted. The accountant opens only the bills that are wrong.
Every day's takings tie out to the settlement and the ledger by the next morning, or you know exactly where they do not. No more end-of-month surprise on the cash line.
The close blockers are on your desk on day one, not day ten. Accruals are listed, cost movements explained, and the management pack arrives drafted for a person to sign.
Stock moves between branches before it is bought again. Purchase orders arrive drafted with days of cover beside them, and supplier prices that creep get named.
Expiring permits, duplicate bank accounts and missing contracts are caught before payroll runs, not after a fine or a failed payment.
Two are agreed and named: an Order & Fulfilment Agent, waiting on connectors that carry order and delivery data, and a Growth & Profitability Advisor, which needs months of governed findings before it has anything to advise from. After those, agents get built in the order businesses ask for them. None reaches the price list until it runs on real data.
The full workforce →Every agent, live or in development, follows the same rule: it does the routine checks, drafts the action, and a person signs. The agents never pay anyone or post a journal. The full workforce
03 · Govern
The controls are the product. Each one is enforced in code, not in a policy document. Every action moves through one path, and a person signs it before anything changes in your books.
The agent drafts a hold, a transfer, a reminder. It waits. Payments, journals and payroll stay with your team.
Above a set amount the team cannot approve alone. The owner signs, and the agent says why the amount crossed the line.
Every draft, decision and result is appended, never overwritten. The month report is counted from it. Connected credentials are stored encrypted.
Why every action still needs a name on it: in a Boston University and BCG experiment, managers told the work came from an "AI employee" caught 16% fewer errors than managers told it came from a tool, and passed it up for review 44% more often. Naming the worker changed how hard people looked at the work. So the queue never asks whether you trust the agent. It asks a person to sign, and records who did. Source.
The same signature, threshold and audit log for any agent a business already uses, not only ours. The platform governs the work; where the agent came from does not matter.
04 · Measure
For every role, every month, in one report.
Value flagged is what the agents put in front of your team. Value approved is what a person signed off on, with the agent's evidence beside it. The month report adds it up from a log nobody can edit, rejections included. It is never called "saved"; it is what your team decided.
In development Hours: the role score already records each task's share of the week; the report will show the hours that moved from a person to an agent.
| Value flagged | Value approved |
|---|---|
| 0 | 0 |
| Approval rate | Actions drafted |
| 0% | 0 |
Industries
Bills, tills, stock and payroll are the same work everywhere. Each industry page lists the back-office roles, which of them has a live agent, and what is honestly not built yet. All five industries
Connectors
Send a month of exports and the agents read them; nothing has to be connected. Connect the system when you want holds and drafts written back into it.
The connectors page lists exactly what each live connector reads and writes today. Nothing in development is sold before it runs.
Pricing
A base fee plus usage you can count.
AP Control from $99, Revenue & Cash from $79, Close & Reporting from $149, Supply & Inventory from $99, HR & Payroll from $49. Three sizes each, set by how much you put through them. Full pricing and the ladder
A verdict for one role, a task map with a reason per task, a workforce-leverage figure, and the documents the verdict calls for: a hiring profile, an enablement plan, the matching live agent, and a 30/60/90-day roadmap. No card, no account.
No. Every action the agent drafts waits in the approvals queue for a person on your team. Above a set amount the owner has to sign. This is enforced in code, not in a policy.
Any system that can export a spreadsheet: bills, orders, receipts, till sessions, bank lines, stock, employee files. Send a month of exports and the agents read them; nothing has to be connected. Connect the system when you want holds and drafts written back. The connectors page lists what each live connector reads and writes today and what is in development.
A monthly base fee per agent in USD, from $49 to $149 at the entry tier, with a usage allowance in a unit you can count in your own system and a per-unit price above it. Pick the agents you need and add another at any time. Excluding tax. Cancel at any month boundary.
Three minutes, no card, no account. If the number is boring, you lost three minutes. If it is not, talk to us before you talk to anyone else. We are early, and we say so.