Published on the 01/07/2025 | Written by Heather Wright

It’s a dance of three parts…

Don’t aim for 100 percent.

That’s the advice from Irene Direnko-Smith, head of process automation at FujiFilm Process Automation when it comes to AI and automation.

Direnko-Smith says she’s seen too many companies get hung up on a small percentage of work that is the most irritating problem they’re seeking to solve. In doing so, though, they often overlook the bigger picture and all the benefits they can receive by just getting started.

“Not many companies know where to get started to be honest.”

“The 90 percent will get you live faster and give you value much quicker,” Direnko-Smith told iStart.

“Oftentimes people focus on the edge cases and the very difficult and complex parts of the process that really irritate everyone and then we forget the bigger picture.”

It’s not an issue unique to AI and automation, with Direnko-Smith noting she’s seen in over many years and across various technologies.

Direnko-Smith’s comments come on the back of a FujiFilm AI in accounts payable automation survey, which found that few Kiwi companies have fully automated their AP processes. The November 2024 survey found just seven percent were fully automated. By June that number was up to around 11-12 percent (from a very small sample size, she notes).

While many businesses are starting to experiment, the majority of experiments remain in the personal productivity space – using Microsoft Copilot to do a task, or having one person complying a spreadsheet faster than before thanks to AI – rather than looking at how an entire process end-to-end can be streamlined with AI and automation.

Direnko-Smith believes we’re missing a trick in focusing on the personal productivity space, with the end-to-end automation offering big productivity gains and enabling businesses to run smoother and better, she says.

“There is significant potential for New Zealand businesses to enhance their AP processes through automation… streamlining processes, reducing manual workloads and introducing AI-driven automation that makes AP more efficient.”

Direnko-Smith says companies are looking to invest and are exploring options around AI and finance automation, but she admits there are sticking points.

“Not many companies know where to get started to be honest.”

Generative AI’s arrival on the scene in late 2022 has garnered plenty of attention, but largely for personal productivity. (And on that note, while generative AI is an important component of AI and automation, Direnko-Smith is keen to stress there are far more layers in the AI and automation equation that add more productivity and gain for companies than just a genAI focus will.)

“When we look at financial process automation end-to-end, people don’t know how they can use the available AI technology for general company productivity and increased efficiency, rather than the personal productivity gains.”

A process such as AP invoice processing involves multiple steps and can touch a wide range of teams from finance and procurement to warehousing.

“When we look at automation of a process end-to-end we need to look at how it impacts individuals at different steps of that process. We need to look at how to make the process more visible and ensure people understand what automation or AI is doing to a particular part of the process, and also make that automation visible to people because AI still has limitations around accuracy so you can’t just let it loose and tell it to automate the entire end-to-end processing.”

The issue of the process is also a key sticking point: Putting AI over a bad process just makes the bad process faster.

Unpacking the process and understanding whether all the steps are necessary and are optimised, is a key part of the process and one which often results in companies realising they have steps in place which are done simply because ‘it’s always been done that way’. Only once the process is optimised should the technology be applied, Direnko-Smith says.

Picking high-volume, high problem, high-cost processes first is another of her recommendations, and feeds back into the notion of going for 90 percent, rather than the 100 percent.

“For companies that have very repeatable, high-volume processes, automating using AI makes a lot of sense and the ROI is very easy to create because you have very repeatable process and there is a volume that stacks up.”

But companies with complex processes that lack visibility can also see big wins, she adds.

“The impact of visibility is a benefit that is really overlooked, in particular when people start putting business cases together. When you have a centralised system that gives you access to that information it means you are making much more informed decisions.”

At a FujiFilm Business Innovation panel discussion earlier this year, Kāinga Ora, which has been using Esker for eight years and is Peppol eInvoicing enabled, revealed 20 percent of its invoices are paid on the same day they receipt the invoice.

It isn’t technology that is holding up the remaining invoices (70 percent of all invoices are processed and paid within two days). Instead, it’s issues around data not aligning – such as the purchase order and invoice information not matching – and exceptions needing to be resolved.

“While eInvoicing helps reduce processing time, it doesn’t guarantee timely payment, especially when discrepancies exist between invoice and purchase order data,” she says.

“It is this dance of three: Process, people and technology. Technology is not going to solve all your problems if you don’t have your people and processes on board. It has to all work together.”

Integrating eInvoice receipting with best-in-class AI can accelerate exception handling and automate non-PO invoice coding.

New Zealand government agencies which process more than 2,000 invoices a year are required to adopt Peppol eInvoicing by January 2026 in a move Direnko-Smith says will likey have a flow on effect for the private sector too.

The November survey showed invoice processing was taking an average of 11-20 days – a stark contrast to top-performing businesses which achieve sub-five-day processing times.

Errors are also high for many with manual processes with 46 percent of respondents reporting frequent discrepancies and errors, which may be flagged by AI error checking and validation tools.

“More businesses need to look at the processes they run and the applications that support them and really ask whether they are getting value out of those technologies and if they’re supporting their people with them, because it is a straightforward process to implement, and the technologies have been around for well over five years and are not that scary.”