AI agent · Finance

AI bookkeeping that keeps the books current, with a person approving anything unclear

AI bookkeeping is the daily job of the Finance Agent we run in Microsoft Teams. It books what matches the rules our finance team sets, reconciles payouts and sends everything else to a person. Below: how it works, and its live log, misses included.

Get your free AI readiness assessment

See its activity log

Built and run in our own business

Works with your team in Microsoft Teams. Can be powered by OpenAI, Anthropic, Google or open-source models.

Finance Agent

Live in our business

43%of this agent’s work automated

Its specialists

BookkeepingLive
TaxLive
CSP BillingPilot

DepartmentFinance
Tasks this month100–500
Last activeToday
Talk to it inMicrosoft Teams

Live from our own systems, updated every hour. See the activity log

Short answer

The Finance Agent is our AI agent for bookkeeping and finance compliance. Bookkeeping labels new bank and card lines, matches payments to invoices and bills and reconciles card-processor payouts, acting alone only when a line exactly matches a rule our finance team set. Tax tracks filing deadlines and cross-checks sales tax collected against tax filed. Anything outside the rules goes to a person as an approval card in Microsoft Teams, a person files every return, and every action is logged.

What it does

What our AI bookkeeping agent does

BookkeepingBooks bank, card and payout activity by rule and sends every exception to a person.

TaxTracks filing deadlines and cross-checks sales tax collected against tax filed.

In Microsoft Teams

Talk to the AI bookkeeping agent in Microsoft Teams

Anything outside the rules comes to the right person as an approval card in a one-to-one Teams chat. They approve, modify or reject it without leaving Teams. Approvals use Microsoft Teams Adaptive Cards, so they work on desktop and mobile.

Teams is the interfaceYour team asks, approves and follows up in Microsoft Teams: a channel for team work, a one-to-one chat for personal approvals.

Your systems stay yoursOur own back end is Odoo, but the agent connects to the ERP, CRM, helpdesk or accounting system you already use through its API.

One governed agentEvery action runs through our AI Command Center: a switch, a daily budget and an audit log for each agent.

Microsoft TeamsChat · Finance AgentChatFiles

Finance Agent · Bookkeeping7:04 AM

One card charge needs you. It is a new vendor, so no rule covers it. I suggest Software subscriptions.

Book this charge?

Charge: $129.00 · new vendor
Account: Software subscriptions

ApproveModifyReject

Approved. Booked $129.00 to Software subscriptions.

Illustration with sample data.

Specialist · Bookkeeping

Bookkeeping: AI bookkeeping for bank, card and payout activity

Bookkeeping labels new bank and card lines, matches payments to invoices and bills, pairs transfers between our own accounts and reconciles card-processor payouts. It acts on its own only when a line exactly matches a rule our finance team set. Anything else goes to a person as an approval card in Microsoft Teams, and if no one decides in time, nothing is booked.

LiveBookkeeping · part of the Finance Agent

Keeps the books current every day: bank, card and payout activity matched and booked within rules our finance team sets, and anything outside them sent to a person. Month-end without the backlog.

44.9%of the Finance Agent’s work

90%handled by the agent, last 30 days

100–500tasks this month

What it works on

Bank categorization rules
Online payment reconciliation
Company card expenses
International payment fees
Receivables
Billing inbox

Bank and card lines labeled by ruleNew lines are labeled and booked when they match a rule exactly; the rest wait for a person.

Payments matched to invoices and billsCustomer payments and bill payments are matched by reference and amount.

Payouts reconciled every 2 hoursCard-processor payouts are reconciled against the payments and fees inside them.

Collections, with a daily digestIt drafts payment reminders and posts a daily aging digest in Teams.

Specialist · Tax

Tax: filing deadlines tracked and sales tax cross-checked

Tax tracks filing deadlines and cross-checks the sales tax we collected against the tax we filed, so nothing is missed. A person reviews and files every return.

LiveTax · part of the Finance Agent

Tracks filing deadlines and cross-checks sales tax collected against tax filed, so nothing is missed. A person reviews and files every return.

4.5%of the Finance Agent’s work

Under 50tasks this month

What it works on

Filing calendar
State coverage check

A filing calendarEvery filing deadline in one calendar, with reminders before each one is due.

Collected against filedSales tax collected is compared with sales tax filed, and any gap is flagged.

A person files every returnThe agent prepares and checks; people review and file.

Scope

AI bookkeeping, and what stays with your accountant

AI bookkeeping takes over the daily, rule-based part of the books: labeling bank and card lines, matching payments to invoices and bills, reconciling payouts and chasing overdue invoices with reminders a person approves. The judgment stays with people. Your controller writes the rules the AI bookkeeping agent follows, approves every exception, closes the month and signs off the accounts. Your accountant keeps the year-end work, tax advice and anything that needs professional judgment, and Tax only tracks deadlines and cross-checks what was collected against what was filed. The result is AI bookkeeping that keeps the ledger current every day, with the decisions still made by your finance team.

Daily: the AI bookkeeping agentBank and card lines, payment matching, payout reconciliation and reminders, inside your rules.

Weekly: your finance teamApproves the exceptions, reviews the aging digest and updates the rules when something new appears.

Month-end and year-end: your accountantCloses the books, prepares the accounts and gives tax advice, starting from a current ledger.

Specialist · CSP Billing

CSP Billing: monthly Microsoft CSP invoices, checked against last month

CSP Billing prepares the monthly Microsoft CSP billing from our distributor’s invoices and checks every customer against last month. A person approves every invoice and bill before it is posted.

PilotCSP Billing · part of the Finance Agent

Prepares the monthly Microsoft CSP billing from TD SYNNEX's invoices and checks every customer against last month. A person approves every invoice and bill before it is posted.

Under 50tasks this month

Live record

What it actually did

Straight from the audit log the agent writes on every action, for every specialist above. Filter by specialist or by outcome; the misses are shown as plainly as the wins.

Live log · last 7 days · updated Oct 21, 2026, 20:28 UTC · times in UTC

2done by the agent alone

105approved by a person

1waiting or left for a person

107expired or rejected, nothing done

Activity log

Wed, Oct 21

15:17BookkeepingProposed reconciling 1 online-shop payoutExpired, nothing doneLog ref #719

11:43BookkeepingSorted and routed 1 ticketAgent aloneLog ref run 8754

10:54BookkeepingSorted and routed 1 ticketAgent aloneLog ref run 8745

06:30BookkeepingProposed a booking for 1 bank or card lineWaiting for a personLog ref #735

Tue, Oct 20

15:17BookkeepingProposed invoicing 1 paid shop orderExpired, nothing doneLog ref #715

10:45BookkeepingReplied to 19 billing queriesPerson approvedLog ref #14, #25–#27, #39, #46–#48, #102–#103, #109, #117, #145, #166, #214, #705–#706, #708, #721

10:45BookkeepingBooked 27 bank and card lines by rulePerson approvedLog ref #55–#56, #63, #66, #89, #95, #105, #125–#127, #134, #137, #141, #169, #172, #177, #183, #185, #187, #271, #312, #340 …

10:45BookkeepingProposed reconciling 1 online-shop payoutExpired, nothing doneLog ref #686

10:45BookkeepingDrafted a reply to 19 billing queriesExpired, nothing doneLog ref #28–#35, #42, #44, #81–#83, #171, #188, #319, #346, #665–#666

10:45BookkeepingProposed invoicing 1 paid shop orderExpired, nothing doneLog ref #653

10:45BookkeepingProposed a booking for 37 bank and card linesExpired, nothing doneLog ref #76, #80, #86–#88, #90, #104, #112, #124, #128–#130, #138–#139, #193, #197, #199, #338–#339, #342, #347, #350, #352 …

10:45BookkeepingMatched 9 customer payments to invoicesPerson approvedLog ref #50, #75, #174, #179, #196, #249, #252, #269, #364

10:45BookkeepingProposed matching 2 batches of bill paymentsExpired, nothing doneLog ref #337, #354

10:45BookkeepingProposed 1 new booking ruleExpired, nothing doneLog ref #343

10:45BookkeepingDrafted a reply to 17 billing queriesRejected by a personLog ref #18–#24, #36–#38, #40–#41, #43, #100, #111, #173, #317

10:45BookkeepingReconciled 1 online-shop payoutPerson approvedLog ref #316

10:45BookkeepingRecorded 4 offsetsPerson approvedLog ref #6–#7, #135, #310

10:45BookkeepingProposed matching 5 customer paymentsExpired, nothing doneLog ref #195, #216, #221–#222, #308

10:45BookkeepingCorrected 5 receipt linksPerson approvedLog ref #294–#298

10:45BookkeepingCompleted 1 taskPerson approvedLog ref #292

10:45BookkeepingCompleted 4 tasksPerson approvedLog ref #275, #283–#284, #288

10:45BookkeepingSettled 3 online-shop payoutsPerson approvedLog ref #285–#287

10:45BookkeepingCompleted 6 tasksPerson approvedLog ref #272–#274, #278, #281–#282

10:45BookkeepingInvoiced 4 paid shop ordersPerson approvedLog ref #262, #276–#277, #280

10:45BookkeepingAdded order references to 2 paymentsPerson approvedLog ref #261, #279

10:45BookkeepingRemoved 3 duplicate bank linesPerson approvedLog ref #253–#254, #270

10:45BookkeepingUndid 1 incorrect matchPerson approvedLog ref #259

10:45BookkeepingReconciled 2 card-processor deposits to the payments insidePerson approvedLog ref #176, #256

10:45BookkeepingLinked 1 settled transactionPerson approvedLog ref #255

10:45BookkeepingMatched 1 batch of bill payments to their billsPerson approvedLog ref #248

10:45BookkeepingProposed matching 7 customer paymentsRejected by a personLog ref #223–#224, #226, #228, #230, #245, #247

10:45BookkeepingProposed matching 5 batches of bill paymentsRejected by a personLog ref #54, #227, #229, #244, #246

10:45BookkeepingCorrected 3 booking linksPerson approvedLog ref #131, #133, #136

10:45BookkeepingProposed correcting 1 booking linkRejected by a personLog ref #132

10:45BookkeepingSent 3 payment receiptsPerson approvedLog ref #121–#123

10:45BookkeepingPrepared 3 payment receiptsRejected by a personLog ref #118–#120

10:45BookkeepingSent 6 payment remindersPerson approvedLog ref #93–#94, #97, #99, #113–#114

10:45BookkeepingDrafted 1 payment reminderRejected by a personLog ref #98

10:45BookkeepingDrafted 1 payment reminderExpired, nothing doneLog ref #96

10:45BookkeepingDrafted 4 payment remindersRejected by a personLog ref #71–#74

Why you can trust this

It comes straight from the audit log the agent writes on every action, not from a slide.
Each row names the specialist that did the work, so the whole agent is in one log.
It shows the misses too: items still waiting, items left for a person, and items that expired because no one reviewed them in time. When an approval expires, nothing is done.
Each row carries the log’s own reference number, so anyone on our team can open the original entry.

What each outcome means

Agent alone: An exact match to a rule our team set, so the agent did it.
Person approved: The agent proposed it and a person approved it.
Waiting for a person: Proposed and held. Nothing happens until a person decides.
Left for a person: Outside the agent’s rules, so it went straight to our team.
Expired, nothing done: No one decided in time. Nothing was done; it comes back for review.
Rejected by a person: A person turned the proposal down, so nothing was done.

See all our agents

Human control

Where a person stays in charge

The agent acts on its own only within rules we set. Anything that affects a customer, money, a bid or a candidate waits for a person.

A person approves the exceptions

Any line outside the rules, and any change to a customer or vendor balance, waits for a person.

Expired means nothing happens

If no one decides in time, the proposal expires and nothing is booked. It comes back for review.

People file every return

Tax flags and prepares; a person reviews and files each return.

Every action logged

Each step, tool call and approval is written to the audit log, which feeds the activity log on this page.

Switched off in one step

Each agent has a daily cost cap and a kill switch, and is pinned to one model tier, so cost and behavior stay predictable.

Behind the agent

Automations behind the Finance Agent

Scheduled jobs and integrations that work next to the agent. These follow fixed rules rather than AI.

Live

Cloud license billing automationRuns monthly cloud-license billing, refreshes the product catalog daily and turns Microsoft Partner Center invoices into vendor bills every two hours.ERP · Partner Center API · Python

Live

Smart dunning holdPauses overdue reminders while a payment is still settling and releases the hold if the payment fails.ERP accounting

Live

Payment failure alertsAlerts staff to card declines and stalled bank-debit verifications within 20 to 30 minutes.ERP · payment processor APIs

Live

Corporate-card rewards optimizerRecommends the best company card for each bill and audits a year of spend for missed cash-back.ERP · rules engine

Live

Daily exchange-rate feedLoads daily currency rates into the ERP automatically.ERP · money-transfer API

For your business

Put an agent like this to work for your team

Most finance teams start with bank reconciliation or matching bill payments, the repetitive work that eats month-end. We build the agent on your Microsoft Azure, connect it to your accounting system through its API, and set the rules with your controller before it acts on anything. Each agent can be powered by OpenAI, Anthropic, Google or open-source models, chosen per task, and we tell you where each model processes your data. It is one of the AI agents for business we run in our own company every day.

Bank reconciliation

Lines that match your rules are booked; the rest arrive as approval cards.

Accounts payable matching

Bills and payments matched by reference, with mismatches flagged.

Collections follow-up

Reminders drafted from your aging report, sent only after approval.

Engagement	Typical timeline	Investment
AI readiness assessment	About one week	Free
First agent pilot: one process, shadow mode	4–6 weeks	$14,900
Production agent with integrations and approvals	6–10 weeks	$24,900–$49,900
Managed AI operations	Monthly	From $1,900 a month

Want a figure for your organization? Get an estimate in 2 minutes with our pricing calculator.

Prices are our standard US list prices for commercial clients. Model usage is billed at cost and capped per agent per day.

Common questions

AI bookkeeping: questions finance teams ask

Does an AI bookkeeping agent work with QuickBooks?

It can. The agent connects to an accounting system through its API, the way ours connects to our own ERP, and QuickBooks Online has one (Intuit Developer: QuickBooks Online API). We confirm the connection and which actions it may take during the free assessment, before the agent books anything.

Can AI do bank reconciliation?

Yes, within rules. Our agent reconciles card-processor payouts against the payments and fees inside them every 2 hours and books bank and card lines that exactly match a rule. AI bank reconciliation still leaves the judgment calls to a person: anything that does not match arrives as an approval card in Teams.

Can AI bookkeeping replace a bookkeeper?

Not on its own. AI bookkeeping replaces the repetitive entry and matching work, so a bookkeeper or controller spends the day on exceptions, rules and month-end instead of typing. Our AI bookkeeping agent books only what matches a rule your team approved, and every exception still goes to a person.

Is an AI agent for accounting safe with financial data?

It runs on Microsoft Azure, acts only on lines that match rules your finance team approved, and logs every action. Prompts and outputs are not used to train the models, and each agent has a daily cost cap and an off switch.

Which AI models does the agent use?

It can be powered by OpenAI, Anthropic, Google or open-source models, chosen per task. OpenAI, Anthropic and open-source models run in Microsoft Foundry on Azure; Google’s Gemini models run on Google Cloud. We tell you which model is used and where it processes your data before work starts, and prompts and outputs are not used to train the models.

Questions buyers ask

Before you decide

Does the agent post entries without approval?

Only when a line exactly matches a rule your finance team approved in advance. Everything else waits for a person, and you can require approval for every entry while you build trust.

What happens if nobody approves a proposal?

It expires and nothing is booked. The item comes back for review, and the activity log shows it as expired, not as a success.

Which accounting system does it work with?

Ours runs on our ERP. For clients, we connect to the accounting system you already use through its API and confirm the connection during the free assessment.

Find the first process worth automating

Start with a free AI readiness assessment: we look at where AI is in use, what data it touches and which process to automate first.

Get your free AI readiness assessment

See all our agents