

Extend the capabilities of your people. Job Control System (JCS): run operational work as tracked jobs, reviewed by your team and connected to your systems.
Get In TouchInternal testing and select beta testers only.
AI Enterprise is a Job Control System (JCS).
A JCS takes a document or queue item, runs a defined workflow, calls the tools that workflow needs, and records the run.
Example path (in production at Dataforge): a vendor invoice arrives by email. The workflow reads the pages, checks the vendor against an approved-vendor list, structures the charges, handles CAD/USD conversion when needed, and posts clean voucher data to QuickBooks Online. If the vendor is missing or the document is unclear, the job raises an interaction for accounting. When the run finishes, you see status, output, and cost for that job.
"This is a Job Control System—not a chatbot that does whatever it wants. Parts of the flow use agents, but they are highly limited in what they can do."

A short introduction to AI Enterprise and the Job Control System idea—not a walkthrough of the portal.
Operational jobs that reach your systems, involve your people, and keep a durable record.
Prepare and post work into tools you already use. Our in-house path uses QuickBooks Online with Office 365 / Exchange email intake; other accounting and email systems follow the same job model.
Queues and tracked jobs keep the work available for the next step and the next person.
Pause for review when policy or ambiguity requires it, then accept, edit, or reject before continuing.
Status, guidance, and approvals stay on the job record for operational review.
See cost per job and cost rollups across runs so leadership can track what ran and what it cost.
"When a vendor isn't on the approved list, the job stops and accounting decides—not the AI."
Operators work in the portal. Jobs, queues, and batches share one job model.
Sign in and submit a file, queue entry, or saved workflow.
Watch the job, or wait until it needs a person. Status stays explicit: succeeded, failed, cancelled, or waiting on interaction.
When it pauses, review, provide input if needed, then accept, edit, or reject before anything posts.
Check the result in the connected system, or download the deliverable. Cost and history stay on the job.
"The AI has been excellent so far at requesting operator input when it needs a person."
Every run keeps a full, inspectable record.
Succeeded, failed, cancelled, or waiting on interaction. Leadership can see what ran and what each job cost.
From the job record you can inspect inputs, outputs, summaries, and cost rollups. Traces answer what the system did when something goes wrong.
A job can request operator input, keep a checkpoint, and resume after accept, edit, or reject. Those decisions stay on the job record.
We run AI Enterprise on our own books—not only in demos.
Vendor invoices and general expenses arrive by Office 365 / Exchange email. Jobs check vendors against our approved-vendor database, raise an interaction when a vendor is missing, and can create the vendor file after accounting confirms. Public vendor details and a short note on what the vendor does are recorded to support later expense guidance. CAD and USD are detected with conversion when needed. Voucher data entry is cleaner and more precise than our manual process, including expense classification.
"The data entry is better than we can do by hand. Voucher data is clean and precise—even expense classification is more consistent."
This is a controlled workflow, not an open-ended agent. Many steps prepare a draft and run it through a validator before tools are used, which keeps behaviour limited and cost low—about 5 to 10 cents CAD per invoice in our current mix. Operator notes act as standing policy: change a note on a vendor and later jobs follow it without relying on someone to remember.
Volume is modest (~100 documents/month) and document quality varies widely. Time saved is about three hours a week today; the larger win is quality and consistency.
From current production and beta use. Results vary by tenant and document mix.
90%+
Share of vendor invoice volume on the JCS path with little or no manual re-keying since the latest workflow version. The rest needs operator review or exception handling.
5–10¢
From our in-house QuickBooks Online path: constrained workflow steps and draft-then-validate processing keep run cost low.
99%
Share of remittance work completed through the automated remittance workflows.
Excellent
Long documents, translation jobs, and audio or video through to a formatted report in current use.
"Many steps don't get tools directly—they process the data, create a draft, and run that draft through a validator. That keeps costs way down. We process invoices for around five to ten cents CAD."
Guide jobs with plain written guidance. Write it once, attach it to a scope, and later runs follow it—like a constantly enforced policy. No coding required.
"For Vendor X, post freight to account 6100. If the invoice currency is USD, pause for review."
Accounting can change a vendor note once; every later job follows it—like a constantly enforced policy.
Jobs, queues, batches, interactions, operator notes, and integrations.
One run with status, history, cost, and deliverables. Portal submit and queues feed the same job model.
Input queues for jobs. Named buffers for intake and handoffs (for example new invoices → ready to book) so work moves through stages without living in the inbox.
Group a set of related data being processed. Keep track of the jobs, re-runs, and interactions that belong to that batch as one unit of work.
When a job needs more input or an operator needs to review what happened, the run pauses for accept, edit, or reject before it continues.
Ordinary written notes that guide how jobs run and when to ask a person. No coding required.
Connect through MCP tool servers and secured credentials (encrypted at rest, not retrievable by users; OAuth where needed) for QuickBooks Online, Office 365 email, and other business apps.
Enroll trusted machines so a job session can reach remote systems under allowlists and credentials.
Available in both deployment models. On-premise is a premium option and subject to site and equipment discovery.
"Some invoices don't have the required data. The system has been versatile so far—far beyond a normal template system."
Real Job Control System runs—starting with how we use AI Enterprise on our own books.
In-house AI Enterprise path: vendor invoices and expenses from Office 365 into QuickBooks Online—over 90% automation, ~100 docs/month, ~3 hours/week saved.

It helps your people get more operational work done. As a Job Control System (JCS), it runs that work as tracked jobs with review, integrations, and a full run history.
Jobs run as controlled workflows with confirmation checks and validators before posting. When a document is unclear or policy requires it, the job pauses for operator input. Accounting can also review after posting. Someone remains responsible for what lands in the connected system.
Yes. Dataforge runs AI Enterprise in-house for vendor invoices and expenses into QuickBooks Online from Office 365 email, and it is used in other production areas as well. Access remains BETA for select beta testers.
Named credentials are encrypted at rest and are not retrievable by users. Workflow steps are highly limited: many agents prepare a draft and pass it through a validator instead of open-ended tool loops. That keeps behaviour constrained and cost per invoice low.
On-premise is a premium service. Potential on-premise deployments need to pass a site and equipment discovery to be considered.