AI in Finance: It’s Becoming an Assistant, Not Just a Reporting Tool

For a while now, AI in finance has meant one thing: nicer dashboards.

Plug it in, get your KPIs, make a graph look better. Job done.

That’s changed. AI is now scanning invoices, matching bank reconciliations, and flagging month-end mistakes before you close the books – but only if your data can actually support it.

We sat down with Kenneth, Technical Lead at Eureka Solutions, who spends his days between software vendors, integration platforms, and the finance teams actually using these systems.

He walked us through what’s new versus what’s just automation with a new name, the questions to ask before switching a feature on, and who’s actually on the hook when AI gets something wrong.

Here’s what he told us:

From reporting tool to assistant

Kenneth tells us that AI has moved from a passive reporting layer to something closer to a colleague.

“It’s shifting from a kind of reporting, passive reporting tool to one where AI is more embedded operationally, is more of an assistant,” Kenneth says.

That now includes:

• Scanning and entering invoices
• Matching bank reconciliations
• Generating items directly from a supplier brochure
• Flagging gaps during month-end close – catching a missed recurring transaction, or a number that looks too far off pattern to be right

The point isn’t just speed. It’s speed and accuracy landing at the same time.

Same conversation, new label?

Digital transformation has been the finance buzzword for over a decade, long before AI joined the conversation.

So what’s genuinely new this time, and what’s just automation wearing a different coat?

Kenneth draws a clear line: “What’s genuinely new with the digital transformation this time is that AI isn’t just automating integrations and workflows and speeding up processes. It’s now actively advising, it’s recommending, troubleshooting and analysing rather than just processing.”

Where it’s not new, he says, is in the integrations and workflows themselves – AI is often just helping build things that already existed.

It’s the difference between a finance function that processes information and one that’s told what to do with it.

The question to ask before any of this works

Every AI conversation in finance eventually runs into the same wall: data.

“Before trusting AI anywhere near finance team data, that needs to be properly structured and controlled and connected so that you’ve got a single source of the truth,” Kenneth says.

“The old rule of garbage in, garbage out still applies, but even more so – like tenfold – to AI. If you’ve got uncontrolled data and it contains errors, then AI is simply going to learn those errors and think they are the process, rather than recommending changes to that.”

That’s the trap. AI doesn’t know your chart of accounts is a mess. It just builds on top of it.

So how do you separate a genuinely useful feature from a sales pitch?

With every NetSuite release bringing new AI capability, it’s easy to feel like you’re constantly behind.

Kenneth’s advice starts before you evaluate a single feature: talk to a partner who already knows what’s live and what’s coming. “Engaging with a partner is the key first step to understand what’s possible.”

From there, he looks for three things a feature actually needs to deliver:

1. Does it save time or reduce errors? Intelligent bank matching and invoice scanning are the obvious examples – less manual input, fewer mistakes.
2. Does it provide actionable business insight? Not just data, but a recommendation attached to it – flagging slow-moving stock in a specific location and suggesting why, and what to do about it.
3. Does it enhance operations elsewhere in the business? Kenneth points to product descriptions written for an ERP record that also need to work as SEO copy on a web store – a job AI can bridge without a human rewriting everything twice.

If a feature doesn’t clear one of those three bars, it’s probably there for the sales conversation, not the finance team.

But will AI replace integrations altogether?

There’s a theory doing the rounds that AI will eventually replace point-to-point integrations entirely – systems just negotiating with each other on the fly.

Kenneth isn’t convinced.

“I think the idea that AI will completely replace traditional integration is more hype. You need to have point-to-point structures in place that ensure security and continuity.”

Where he does see AI adding real value is inside the integration itself:

• Helping build mappings faster
• Handling the small transformation logic that used to need custom coding
• Making live decisions as data flows through

His example: a sales order comes in, AI checks stock, sees the item is unavailable, and automatically raises a purchase order or alerts procurement – without a human needing to spot the gap first.

Assistant, not autopilot

Finance data is some of the most sensitive information a business holds. So what needs to be true before AI gets near it?

Kenneth lists three non-negotiables:

• Confirm the AI provider isn’t training public models on your data, and that it deletes your data after use
• Make sure AI inherits your existing ERP role-based permissions, so a warehouse-access user can’t suddenly ask about payroll
• Keep a human validating anything AI creates, rather than letting it post transactions unsupervised

That last one is where he draws a hard line with clients:

“One of the specific capabilities I advise customers not to switch on yet… is the ability to post transactions. Get used to the AI and how it works first… turn it on as a kind of view access only first and get comfortable with it before turning on the ability to post data.”

If something goes wrong, who’s actually responsible?

This is the question every finance leader eventually has to answer.

Kenneth doesn’t hedge: “Accountability does sit with the business to validate and oversee those outputs.”

When an AI-driven process gets something wrong, he traces it back to one of three causes:

1. The model lacks the necessary knowledge
2. Its reference material is out of date (NetSuite alone upgrades every six months)
3. The underlying source data was wrong to begin with

Understanding which one it was is what stops the same mistake happening twice.

The one thing every finance director should understand first

Before shopping for AI tools at all, Kenneth says the starting point isn’t the tool – it’s the plumbing.

“The one thing every finance director needs to understand is their systems, how they interact and where the data lives, and making sure that they have that one source of the truth.”

Businesses already running an integration platform like Besyncly can put new AI tools straight to work, because the connections already exist.

Businesses still relying on ageing, disconnected systems will find AI simply can’t get the data it needs – no matter how good the tool is.

Modern, connected ERP systems are the ones built for what’s coming. Everything else needs the groundwork done first.

Ready to find out where your systems and data actually stand before switching AI on?

Book a free ERP strategy call →