AI agent for financial control: what it actually monitors

It’s payday. The balance doesn’t cover payroll, and only now does someone notice that three large clients never paid the invoices due on the 25th. The information had been sitting there for a week, in the bank statement and in the billing system. Nobody looked, because nobody had time to look every single day.

That gap is exactly what an AI agent for financial control is for. Not to do your bookkeeping, and not to replace your finance team, but to watch what nobody watches often enough and raise a flag while there’s still time to act.

Short answer: an AI agent for financial control is software that continuously reads bank statements, receivables, payables and collections, compares actual figures against expected ones and fires alerts when something breaks rules the company defined, such as a low projected balance, a late-paying client or an unusual expense, while suggesting the next step.

What an AI financial control agent is (and what it isn’t)

The difference from the tools you already have lies in the verb. Your ERP records. Your dashboard shows, but only to whoever opens it. The monthly report tells you what already happened. The agent watches and warns: it runs on its own, hourly or daily, and only calls you when there’s a decision to make.

If you already have a well-built financial dashboard, the agent doesn’t replace it. It fixes the fact that dashboards depend on someone remembering to open them at the right moment. In a small business, where the owner doubles as CFO and four other roles, that rarely happens.

What it isn’t: a bot that pays bills on its own. Agents that work well in finance operate at three levels of autonomy, and almost every company should start at the first one.

• Observe: reads the data and alerts. Zero risk.
• Suggest: alerts and proposes the action (chase client X, push supplier Y) with a draft ready.
• Act with approval: sends the reminder or schedules the payment after a human clicks approve.

For a broader look at this autonomy scale outside finance, see what an AI agent does for you (and what it still can’t).

What a finance agent monitors: the 4 fronts

Classic financial control fits into four fronts. The agent doesn’t invent any of them; it just looks at all of them at once, every day.

1. Projected cash, not just current cash

Today’s balance is the most misleading number in finance. What matters is the balance over the next four to eight weeks, crossing scheduled inflows with outflows. The agent recalculates that projection with every new transaction and alerts you when it drops below a floor, say, 30 days of fixed costs.

2. Collections and the receivables curve

Every company has a pattern: how much of monthly revenue usually lands in week one, week two and so on. The agent compares cumulative receipts against that curve and catches the gap early. We walk through an example below.

3. Late payments and behaviour changes

A client who always paid on time and is now five days late tells you more than a client who is always late. The agent learns each payer’s history and flags the break in pattern, which is often the first symptom of trouble on their side.

4. Out-of-pattern expenses

A new supplier with different bank details, a subscription that jumped 20% without notice, a duplicate payment. Nobody catches these by skimming the statement at month-end. A simple rule catches them the same day.

Front	What the agent compares	Example trigger	Frequency
Projected cash	Future balance vs. safety floor	30-day balance below one month of fixed costs	Daily
Receipts	Cumulative receipts vs. historical curve	More than 10% below the curve	Daily
Late payments	Current delay vs. client history	Reliable payer 5 days late	Daily
Expenses	Transaction vs. supplier pattern	15% above average or a new bank account	Per transaction

Why start with late payments

If you can only switch on one front, make it receivables. The market explains why: according to the Serasa Experian Business Default Indicator, released in October 2026, Brazil hit a record 9.3 million companies in default in August, owing a combined R$ 244.4 billion. Micro and small businesses account for 8.8 million of them.

Read that number from the other side. Many of those companies are someone’s clients, and 75.6% of their defaulted debts are owed to non-financial creditors: suppliers, service providers, utilities. If you sell to other businesses, the odds of having a few on that list are real.

Internal controls are often fragile too. The Financial Habits 2025 survey by Sebrae/PR (December 2025) found that 61% of small businesses mix personal and company finances, 30% track cash in spreadsheets, 25% in a paper notebook and 10% don’t track it at all. From that starting point, an automatic alert is worth more than a sophisticated report.

Example: the alert that arrives three weeks early

Take a simulation with round numbers. A B2B services company bills R$ 180,000 a month across 60 clients with an average ticket of R$ 3,000. It starts the month with R$ 40,000 in the bank and R$ 174,000 in planned outflows (payroll, taxes, suppliers, rent and software). If everything comes in as expected, it ends the month with R$ 46,000.

Historically it collects 20% of revenue in week 1, 30% in week 2, 30% in week 3, and 20% in week 4. Here’s what actually happened:

Week	Expected	Actual	Cumulative gap
1	R$ 36,000	R$ 34,000	-R$ 2,000
2	R$ 54,000	R$ 41,000	-R$ 15,000
3	R$ 54,000	R$ 47,000	-R$ 22,000

In week 1, a R$ 2,000 gap is noise. By week 2, cumulative receipts are R$ 15,000 below the curve, about 17% short. That’s when the agent fires its alert, with the list of clients who haven’t paid and each one’s history attached.

If nobody acts and the delay isn’t recovered, the ending balance drops from R$ 46,000 to R$ 24,000: the month’s cushion shrinks by 48%. With the alert in week 2, you have three weeks to chase payments, renegotiate supplier terms or factor receivables calmly, instead of deciding under pressure on payday.

Which alert rules should you set up first?

The most common mistake is switching on too many alerts. Within two weeks the team ignores all of them. Start with five rules and only add another when one stops firing for no reason:

1. Cash floor: 30-day projected balance below one month of fixed costs.
2. Receivables curve: cumulative receipts more than 10% below what’s expected at that point in the month.
3. Pattern break: a client with an on-time track record who is more than five days late.
4. Concentration: a single client accounts for more than 20% of open receivables.
5. Suspicious expense: a payment 15% above the supplier’s average, a duplicate, or one going to a new bank account.

These thresholds are starting points, not universal truths. A highly seasonal business needs a curve per month, not a single one. To pick the indicators that feed these rules, the guide to financial indicators for small businesses helps separate the essential from the nice-to-have.

Where a finance agent gets it wrong

It’s worth being upfront about the limits, because they decide whether the project works.

Dirty data means false alarms. If bank reconciliation is behind or transactions are categorised carelessly, the agent will point to gaps that don’t exist. In practice, the first few weeks are more about cleaning the base than making decisions.

The historical curve breaks in unusual months. Black Friday, annual contract price adjustments or a big new client distort the comparison. The agent needs to know about them, or you’ll get “receipts above expected” alerts that help nobody.

Too much autonomy too soon. Letting the agent send automatic reminders to a strategic client who was late because of a broken invoice link is a fast way to lose the account. Automatic collections suit low tickets and high volume; big clients deserve a human.

If your payments run through Stripe, the 7 weekly loops that catch leaking revenue show the same reasoning applied to a specific gateway.

How to start in 30 days

Week 1: connect the bank statement and receivables. Nothing else. Check that balances match what your finance person sees.

Week 2: build the receivables curve from the last six months and turn on the first two rules.

Week 3: add payables and the expense rules. Review false alarms and adjust thresholds.

Week 4: move to the “suggest” level: each alert comes with a proposed action and a draft reminder. Execution stays human.

The general roadmap for taking any agent from idea to operation is in how to implement AI agents in your company.

Frequently asked questions

Can AI build a cash flow forecast on its own?

It can build and refresh a cash flow projection from bank and ERP data, and do it far more often than a person would. What it can’t do well is guess what isn’t recorded, like a verbal deal with a supplier. The forecast is only as good as the transactions behind it.

What’s the best AI for small business financial control?

The best one connects to the sources you already use, such as your bank, ERP or spreadsheet and payment gateway, and lets you write alert rules in plain language. Before comparing brands, list where your data lives; a tool that forces you to migrate everything rarely pays off for a small company.

Does an AI finance agent replace the accountant or financial analyst?

No. Accountants handle tax and statutory obligations, which require professional accountability. The agent takes the repetitive part of watching numbers every day off the analyst’s plate, freeing time to negotiate with clients and suppliers, which is where cash problems actually get solved.

Is it safe to give an AI agent access to bank data?

To monitor, the agent only needs read access, via Open Finance, an exported statement or an ERP integration. That’s the recommended level to start with. Permission to move money should only exist with human approval for every transaction.

A finance function that warns you first

Financial control in small businesses rarely fails for lack of data. It fails because data arrives late, in a month-end report that merely confirms the problem. A well-configured agent flips that order: the gap shows up in week 2, when it’s still a phone call, not on payday, when it’s already an expensive loan.

With Sherlok, you connect your sources, ask questions like “which clients are paying later than usual this month?” and get the analysis ready, with agents that watch these numbers for you and warn you when something leaves the curve.