A process that once took 26 minutes was reduced to just 2 minutes. That was the result achieved by Amazon Finance after applying generative AI to monitor, analyze and summarize VAT regulatory changes. It is a useful example of how AI can create measurable efficiency when applied to the right process and the right scope of work.
According to AWS, the time required to generate decision-relevant insights from this process fell by approximately 92%, from 26 minutes to 2 minutes. Around 80% of the suggested titles, summaries and impact assessments generated by the system were accepted by Amazon’s tax specialists.
The more important point is not simply the 92% figure. The value came from applying AI to a specific, frequently repeated process that previously required substantial manual effort. For businesses, this is also a practical way to approach AI automation: start with processes where efficiency improvements can be clearly identified and measured.
The opportunity is often hidden in everyday work
Many business activities appear insignificant when viewed individually: extracting information from emails, entering data into spreadsheets, consolidating reports, preparing quotations, updating customer information, checking documents, preparing meeting minutes, searching for information or following internal approval steps. A task may take only 10, 20 or 30 minutes, but when it is repeated dozens or hundreds of times each month by multiple employees, the cumulative time and cost can become substantial.
This is where AI and automation can often generate relatively quick improvements. Research by Jung Ho Choi and Chloe Xie, using data from 79 small and medium-sized businesses and more than 200,000 accounting transactions, found that AI adoption was associated with approximately a 55% increase in weekly client-support productivity. Around 8.5% of accountants’ time shifted away from routine data entry toward higher-value activities such as business communication and quality assurance, while monthly closing time was reduced by approximately 7.5 days.
These findings show that the value of AI goes beyond reducing labor costs. When manual work is reduced, the same team can handle a larger workload, reports can be completed earlier, customers can be served faster, and employees can spend more time on work that requires experience, judgement and communication.
What EPS has seen in client implementations
In several projects implemented by EPS Investing for clients, reviewing existing processes and applying AI and automation has helped businesses achieve approximately 20–30% cost savings within suitable areas of work. Actual results vary depending on the business, workload, process structure and the extent to which activities can be automated.
The strongest opportunities often arise in activities with relatively high frequency, significant manual effort and a reasonable degree of standardization. A process may previously have required employees to receive information, enter data, check it, consolidate it and prepare a report. After redesign, some steps can be automatically received and processed by the system, with an initial output prepared before being passed to the responsible person for review.
The savings therefore do not necessarily come from reducing headcount. In many cases, the greater value is that the same team can handle more work, complete tasks faster and spend less time on repetitive activities. If a task that previously required 30 minutes can be reduced to 20 minutes and is performed 1,000 times per month, the business saves more than 160 working hours. As multiple processes are improved in this way, the impact on cost and productivity can become significant.
AI still requires human oversight
A process using AI does not necessarily need to be 100% automated. In areas such as finance, accounting, contracts, quotations, legal work or customer communication, AI can support reading, classification, consolidation, comparison and initial drafting, while decisions, accountability and exceptional cases still require human oversight.
The research by Choi and Xie also indicates that AI and professional experience can complement each other. More experienced professionals tend to intervene more when the system shows lower levels of confidence. A process that automates 60–70% of repetitive work while keeping people involved at the points requiring judgement may therefore already deliver substantial value. The key is to design the right division of work between technology and people.
Good automation starts with reviewing the process
Applying AI to a poorly designed process may make the work move faster, but it does not necessarily make the business operate better. If a process currently has six steps when four would be sufficient, it should be reviewed before automation. If the same data is being entered repeatedly, the information flow should be reconsidered. If responsibilities and decision rights between departments are unclear, AI is unlikely to solve the underlying problem.
This is why AI automation is closely connected with operational restructuring. Before selecting technology, a business needs to understand how work is currently performed: where inputs come from, who processes them, which steps create value, which steps can be removed, which can be automated and where human review or approval is still required.
In client projects, EPS prioritizes using the platforms and systems that the client is already using. The focus is first on improving processes and operating efficiency, and then on applying technology in a way that fits the actual operations of the business.
Measure the results before scaling
AI does not need to produce dramatic savings across the entire company to justify investment. A process that reduces costs by 20% but consumes hundreds of working hours each month may deserve greater priority than a process that can be 80% automated but occurs only a few times per month.
A practical approach is to identify a small number of processes with clear potential, measure their current time, cost and quality, review how the work is performed, automate the appropriate parts and then measure the results again. If the improvements are demonstrated, the business can gradually expand automation to other processes. This also reduces the risk of investing in AI simply because it is a trend without establishing its economic value.
More broadly, AI is becoming another tool for restructuring how work is performed: reducing manual activities, shortening processing times, improving management information and making better use of people. The ultimate objective remains straightforward: improving business operating efficiency.
Sources: Amazon Web Services, Delivering Regulatory Insights 92% Faster Using AWS with Amazon Finance Technology; Jung Ho Choi & Chloe L. Xie, Human + AI in Accounting: Early Evidence from the Field, Journal of Accounting Research, 2026; Bain & Company, Your AI Budget Is Growing. Your Returns Aren’t. Here’s Why., 2026.
EPS Investing | Operational Restructuring
EPS Investing supports businesses in reviewing and restructuring management and operating processes, identifying opportunities for improvement, and applying AI and automation where appropriate to increase productivity, reduce manual work, control costs and improve operating efficiency.
EPS prioritizes practical issues where results can be measured and, where possible, works with the platforms and systems clients are already using.





