Image
Department
Operations

The agent-run operations back office

Mercury Digital builds custom AI agents that run back-office work for mid-size businesses: close, AP, approvals and reporting, with named human oversight.

The short answer

Short answer

Mercury Digital builds custom AI agents for back-office operations at mid-size businesses.

It designs, deploys and governs agents for back-office functions such as month-end close, accounts payable, approvals and reporting, plus operations work like exception routing and scheduling, all under one gate structure, with a named person approving every consequential action. Engagements start with a fixed-price AI Opportunity Audit.

Scope

What this page is, and isn't

This page describes the pattern and the roles. It doesn't claim that a full agent-run department has been delivered or measured. The audit is where the first workflow gets chosen.

Who Mercury Digital is (and who this is for)

CompanyMercury Digital, an AI systems engineering firm. Founded 2001, fully distributed.
What it buildsCustom, governed AI agent workflows, run on AAOS (Mercury's orchestration layer: web app plus MCP servers, model-agnostic, runs on your cloud)
Who it's forSmall and mid-size businesses, enterprise teams and non-profits whose operations run on email, spreadsheets and several systems that don't talk to each other
How engagements runAudit → Design → Deployment (shadow mode first, autonomy opened gate by gate) → Governance
Entry pointAI Opportunity Audit: fixed price, under 20 hours of your team's time
TiersAutomate a Workflow (Tier 1) → AI Employee (Tier 2) → AI Department (Tier 3, this page)
Human controlEvery consequential action is behind a named person's approval gate. The gate is closed by default and the agent can't open it.

Below, the back office is described by function: close, AP, approvals and reporting, plus the operations work that feeds them.

An agent-run back office, by function

Each workflow works on its own and has its own page. The "department" is what you get when they run together under one set of roles and one gate structure.

FunctionWorkflowWhat the agent doesGate (held by a named person)
CloseAutomating month-end closeReconciles during the period and drafts recurring and accrual entries with support attachedController approves every entry and the close package
APInvoice intake and exception triageReads, matches and codes invoices, and routes each exception to its ownerA named approver; payment always stays with a person
ApprovalsPurchase approval workflows that don't stallTurns reorder recommendations into approval requests, with no re-typingBudget owner approves
ReportingRecurring reportsProduces every scheduled report each cycleEach report is signed off by its owner before it goes anywhere
ReportingJob and project status reportingKeeps status current from existing toolsOwner signs off before it circulates
OperationsException detection and routingWatches every other workflow and lands each exception in one queue with one ownerOwner resolves anything outside a rule
OperationsInventory monitoring and reorder triggersWatches stock and proposes reordersNothing is ordered without approval
OperationsSupplier scorecardsPulls supplier performance from every system into one scorecardOwner signs off on the scorecard
OperationsVendor negotiation prepBuilds the negotiation brief from the scorecardNo vendor is contacted without sign-off
OperationsScheduling and dispatchProposes schedules from the same job recordsDispatcher approves before anything is sent

How to start (one workflow at a time)

01

Book the audit

Map your back-office workflows and score them for readiness. See the readiness checklist.
02

Pick one workflow

Pick one that repeats, has mostly written rules and already ends in a sign-off. That's a Tier 1 engagement.
03

Run it in shadow mode

It runs next to your team until the gate holder decides, from the record, that it has earned a gate.
04

Mercury's standard policy

Under Mercury's standard policy, the agent acts on its own only after it has run in shadow for at least 50 items of each type and at least 2 weeks, with 95%+ agreement with the human reviewer and zero critical errors.
05

Go live behind the gate

A named person approves every consequential action.
06

Add the next workflow

Connect the handoff, so one workflow's output becomes the recorded input of the next.
07

Group workflows into a role, then a function

A role is Tier 2 (AI Employee) and a function is Tier 3 (AI Department), with named oversight roles and one audit trail.

Alternatives

Alternatives to hiring more back-office staff

Hire, or outsource

Hire. Fits when: the work needs judgment that can't be written down. Trade-off: cost and ramp-up time. The process stays in people's heads.

Outsource / BPO. Fits when: the volume is high and standard. Trade-off: less visibility and control over exceptions

Self-serve tools, or RPA

Self-serve automation tools (e.g. Zapier, Make, n8n, Power Automate). Fits when: simple, predictable triggers, and you have someone to build and maintain them. Trade-off: breaks on exceptions; your team owns the maintenance

RPA platforms (e.g. UiPath). Fits when: stable screens and high volume. Trade-off: brittle when inputs or interfaces change

Custom governed agents

Custom governed agents (Mercury Digital). Fits when: messy inputs, multiple systems, exceptions that need routing, and a need for human sign-off and an audit trail. Trade-off: starts with one workflow. Each step of autonomy has to be earned in shadow mode.

Questions, answered

AI systems engineering firms such as Mercury Digital build and govern custom agents for back-office workflows, starting with one workflow and adding more under one gate structure. Platforms such as UiPath, n8n or Copilot Studio are tools your own team builds on.
AI Opportunity Audit · Operations

Start with one workflow, then add the next

Bring the two operations workflows that hand off through a person today and find which one proves first. The AI Opportunity Audit is fixed-price, quoted before we start.