Automation for finance teams: what it does for a CFO

The short answer: automation gives a finance team its month back. Invoices go out on the day the work is delivered instead of the week after, collections run on a schedule that doesn't depend on someone remembering, supplier invoices get read and matched before a human sees them, the bank reconciles itself nightly, and the close stops being a 10-day project. The accounting system stays where it is. What changes is how much of the work around it gets typed by hand.

Last updated: September 9, 2026

What does automation do for a finance team?

It takes the re-keying, the chasing and the matching out of the month, and leaves the finance team with the exceptions and the judgment.

A finance team in a 20 to 200 person company usually runs on 3 systems that don't talk: the accounting system (Priority, SAP Business One or Hashavshevet, depending on when the company was founded and who the first accountant was), the bank, and everything else (the CRM, the project tool, the expense app, the email inbox where supplier invoices arrive). The gaps between those systems are filled by people. That's the work automation removes.

Rule-based workflows do most of this without any AI. Where AI (in our stack, Claude) comes in is reading: a supplier invoice that arrives as a PDF photographed on a phone, a customer email that says "we'll pay next week, the PO number changed", a receipt in Hebrew and English on one page. Those steps needed a person 2 years ago and don't now.

Invoicing

Invoices go out late because the person who knows what was delivered and the person who sends invoices are 2 different people, and the message between them is a Slack thread.

The automation connects the delivery record to the invoice. A project marked complete, a shipment confirmed, a monthly retainer date arriving: any of those creates the invoice in the accounting system, with the right customer, items and terms, and sends it. The finance person sees a list of what went out, and steps in only when the workflow flags something (a customer with no PO on file, an amount above the usual, a new customer with no credit terms yet).

Two details that matter in Israel. Invoices above the current threshold need an allocation number from the Tax Authority, and your accounting system handles that, so the workflow creates the invoice inside the system rather than as a PDF from somewhere else. And the "delivery record" has to be real: if your team marks projects complete in a spreadsheet, the automation will invoice from the spreadsheet, and the spreadsheet is where the errors live.

The measure to hold it against is days from delivery to invoice. In companies that run this by hand it's typically 5 to 15 days. Automated, it's 0 to 1. The cash arrives that much earlier, and you didn't change a single payment term.

Collections and dunning

Collections fail for a boring reason: the reminder depends on someone remembering, and the person who has to remember also has 30 other things.

The automated version runs on aging buckets. 3 days before due, a friendly note with the invoice attached. On the due date, a shorter one. At 7 days overdue, a firmer one, still polite, with a payment link if you have one. At 21 days, the finance lead gets a task to call, with the whole history on one screen. At 45, whoever handles disputes gets it. The wording of each step is yours, written once, reviewed once.

Israeli payment terms (net 30, 60 or 90 from end of month) make the aging buckets long, which is exactly why the reminder 3 days before the due date matters: the customer's own payables team has often forgotten the invoice exists.

The AI piece here is small and specific: reading the replies. "Paid yesterday, check your account" pauses the sequence and creates a task to confirm. "We never received the invoice" resends it and logs that. "The PO number is wrong" routes to the account owner. Without that reading step, the sequence either keeps sending (embarrassing) or stops on every reply (useless).

Companies that had no consistent sequence before typically see their collection days fall by a few weeks. If you already run collections tightly by hand, the gain is smaller and it's mostly in hours.

Supplier invoice intake

Supplier invoices arrive through 4 channels (email, WhatsApp, a supplier portal, and the occasional envelope), and someone keys every one of them into the accounting system by hand.

The automation reads them. An inbox rule catches the email attachments, a WhatsApp Business connection catches the photos, and a document-processing step reads each invoice: supplier, invoice number, date, amounts, VAT, line items. It checks the supplier exists, matches the invoice to a purchase order if you use them, flags duplicates (the same invoice sent twice, which happens more than anyone admits), and posts a draft entry in Priority, SAP or Hashavshevet for a person to approve.

That approval step is the point. The system does the reading and the matching; the finance person does the deciding, from a screen that shows the invoice next to what the system extracted. Expect the model to read a clean PDF correctly nearly every time and a phone photo of a crumpled invoice most of the time, and design the review screen for the second case.

What this replaces, in hours, depends on volume. A company processing 200 supplier invoices a month spends roughly 25 to 40 hours keying them. That's most of a working week, every month, spent typing numbers the supplier already typed once.

Bank reconciliation

Reconciliation is matching, and matching is what software is for.

The nightly workflow pulls the bank transactions (through the bank's API where there is one, or an export), pulls the open items from the ledger, and matches: exact amount and reference first, then amount and date within a window, then the fuzzier cases (a customer who paid 3 invoices in one transfer, a supplier who took a discount you didn't agree to). Everything matched is posted. Everything unmatched goes on one list, and that list is the finance team's morning.

Israeli banks vary a lot on API access. Some of the larger ones have usable interfaces for business accounts, others still mean a daily export. Ask your bank before you plan the build, and plan for the export version as the fallback, because the export version always works.

The gain here is knowing where you are. A team that reconciles monthly finds out about a bounced payment 3 weeks late. A team that reconciles nightly finds out tomorrow morning.

Month-end close

The close takes 10 days because it's 40 tasks in 6 people's heads, and each person waits for the one before.

Write the tasks down, in order, with who owns each one and what it depends on. That alone shortens the close, before any automation. Then let a workflow run the checklist: it opens the tasks on day 1, chases the ones that are due, runs the ones that are mechanical (accruals from the fixed list, depreciation, intercompany, the recurring journals), and shows the CFO a board of what's done, what's blocked, and by whom.

The mechanical tasks are the ones to automate first. Accruals for recurring costs, the FX revaluation, the prepayments schedule: these are the same every month and they're where the arithmetic errors live. The judgment tasks (the bad-debt provision, the revenue recognition call on a half-finished project) stay with the controller, and the board shows them earlier because everything before them finished faster.

A close of 3 to 5 working days is a reasonable target for a company this size once the mechanical half runs itself. Faster than that is possible and usually means the company also cleaned up its chart of accounts, which is a separate project.

Expense reports

Expense reports are the process everyone hates, and the one with the fewest excuses for staying manual.

An employee photographs a receipt on their phone. The image goes to a workflow that reads it (amount, vendor, date, VAT, category by rules you set), attaches it to the employee's open report, and asks the manager to approve on their phone. Approved items post to the ledger. Missing receipts are chased by the workflow, and the bookkeeper is off that duty.

Two Israeli specifics. Receipts here are often bilingual and often thermal-printed and faded, so the reading step has to be tested on your actual receipts, from your actual employees' phones, before it goes live. And the VAT recovery rules (which categories, which percentages) belong in the workflow's configuration, reviewed by your accountant, so the categorisation is consistent instead of depending on each employee's memory.

Cash-flow reporting

The cash-flow report is the one the CEO asks for on Sunday night and the one that takes 4 exports and a spreadsheet to build.

Once reconciliation runs nightly and invoicing is automated, the cash-flow report is a by-product. The workflow already has the bank balance, the open receivables by expected date, the open payables by due date, and the recurring commitments. A short Python step rolls them forward 13 weeks; a template drops them into one page; it lands in the CFO's inbox before the CEO wakes up.

The AI addition is a written paragraph on top ("cash tight in weeks 6 to 8 because 2 large receivables land after the quarterly VAT payment; the 3 overdue invoices above would cover it") so the reader gets the point in 20 seconds. That paragraph should be generated from the numbers, and it should say "I don't know" where the model has no basis. That's a matter of how the prompt is written, and something to ask a vendor to show you.

How does this connect to Priority, SAP and Hashavshevet?

The accounting system stays the system of record, and the automation writes into it through its interface rather than around it.

Priority has an OData API that's workable for reading and writing most documents, and it's the one we build against most often. SAP Business One has a service layer that does the same. Hashavshevet is harder: depending on the version, integration means its own API, a file import, or an intermediary, and the answer changes the build. Any vendor who says "we integrate with everything" without asking which version you run hasn't done this before.

The other handoff, and it's the one most companies forget to design, is the external bookkeeper or accountant. Many companies of this size run bookkeeping outside, and the automation should produce what that bookkeeper needs (posted invoices, matched bank lines, filed supplier documents with their extracted data) in the form they need it, so they aren't asked to log into one more system. Ask them what they want to receive. They'll tell you, in detail, because they've been re-keying your work for years.

What should a finance team automate first?

Start with supplier invoice intake or collections, whichever costs you more hours today, and do reconciliation second.

The reasoning: intake and collections both have a clear trigger, a clear rule and a visible result inside a month. Reconciliation is the higher-value one long term but depends on your bank's access, which can add weeks. The close comes after reconciliation, because a close on unreconciled books is a close on guesses. Expense reports are a good project any time; they're just rarely the biggest number.

Put your own hours into the ROI calculator and the order usually sorts itself out. The wider playbook for picking a first process is in the automation basics guide, and if you want to know what the project itself looks like, week by week, we wrote that up as what an automation project looks like.

What should a CFO watch out for?

Watch for anything that posts to the ledger without a review step, anything that only the vendor can change, and anything where the extraction accuracy was measured on someone else's documents.

The review step first. Every automated post should be a draft a person approves, at least for the first quarter, and the approval screen should show the source document next to the extracted numbers. Once you've watched it be right for 3 months you can widen the auto-post rules, one category at a time.

Ownership second. The workflows should run in your accounts, be readable by your controller, and be exportable. If a vendor's answer to "what if we leave" involves a rebuild, keep looking. How we work spells out what we hand over, so you can compare.

Accuracy last. Ask to see the extraction tested on 50 of your own supplier invoices before you sign, and ask what happens to the ones it gets wrong. The good answer is that they land in the review queue with a flag. The bad answer is a percentage with no explanation of the remainder.

Segregation of duties still applies, too. The person who configures the approval thresholds and the person who approves shouldn't be the same person, and the audit log on every automated post is what your auditor will ask for. Build that in from day 1, because retrofitting it is miserable.

More on the industry angle, including the compliance side, is on our finance industry page. If a term in this guide is unfamiliar, the glossary has the short version.

Frequently asked questions

What does automation do for a finance team?

It removes the re-keying, chasing and matching from the month. Invoices are created from delivery records, collection reminders run on aging buckets, supplier invoices are read and matched before a person sees them, the bank reconciles nightly, and the close runs from a checklist that chases itself. The finance team keeps the exceptions and the judgment calls.

Can automation work with Priority, SAP Business One or Hashavshevet?

Yes, and the method differs. Priority has an OData API, SAP Business One has a service layer, and Hashavshevet depends on the version you run, which can mean an API, a file import or an intermediary. A vendor should ask which system and which version before they tell you what's possible.

Is it safe to let software post to the ledger?

It's safe when every automated post starts as a draft that a person approves, the approval screen shows the source document next to the extracted numbers, and every post carries an audit log. After a quarter of watching it be right, the auto-post rules can widen one category at a time.

How accurate is AI at reading supplier invoices?

On clean PDFs, close to perfect. On phone photos of crumpled or faded invoices, most of the time, which is why the review queue exists. Ask any vendor to test on 50 of your own invoices before you commit, and ask what happens to the ones it gets wrong.

What should a finance team automate first?

Supplier invoice intake or collections, whichever eats more hours today. Both have a clear trigger and a visible result within a month. Bank reconciliation comes next, then the month-end close, because a close on unreconciled books is a close on guesses.

Does the external bookkeeper still have a job?

Yes, and a better one. The automation produces what the bookkeeper needs (posted invoices, matched bank lines, filed supplier documents) in the form they need it, so their time goes to review and to the questions that need an accountant, and away from keying in numbers the supplier already typed.

Next step

If your close takes more than a week, or your collections run on someone's memory, that's the conversation. Talk to us with a rough count of monthly invoices in and out, and we'll tell you which process to start with and what the first 8 weeks would look like.