Amazon’s finance teams are relying more on AI—and not just for the simple stuff

Amazon.com’s finance teams are using generative AI in new ways to more efficiently comply with complex tax regulations, help evaluate the financial impact of new products and better analyze changes in revenue.

It’s part of an effort by Amazon to further apply AI to help tackle more complicated finance functions, moving beyond the automation of rote back-office processes taking place at many companies. Amazon has expanded the areas of finance where it uses generative artificial intelligence over the past year, executives said. About a year ago, the company applied GenAI to core back-office functions such as fraud detection, account reconciliation, financing forecasting and contract reviews.

The moves come as the e-commerce giant has made a hard push into AI, investing billions in the startup Anthropic and planning to spend more than $100 billion on data centers. Its investments in tech infrastructure are driving cash capital expenditures, which totaled $31.4 billion in the latest quarter, up from $16.4 billion a year earlier. Chief Executive Andy Jassy in June said that AI will result in a smaller workforce in the next few years.

Amazon is using generative AI for finance work ranging from report writing and document summarization to synthesizing data sources and reshaping how the company generates financial insights, Chief Financial Officer Brian Olsavsky said in an interview. “Generative AI is transforming our finance function at Amazon," Olsavsky said. “It’s making our processes faster and more efficient, helping us identify trends and opportunities sooner, and enabling our teams to focus more on strategic thinking."

These use cases have expanded as the technology, particularly agentic AI, has advanced.

Tax time

For one, generative AI helps Amazon’s finance team to improve its efficiency and consistency in complying with Internal Revenue Service rules on tax issues ranging from transfer pricing to research and development and patents.

The team traditionally pulled together many data sources and tried to solve “incredibly complicated problems," said John Felton, finance chief at Amazon Web Services, the company’s cloud-computing unit.

The company then began turning to AI for help with interpretation of rules and regulations and tax-related work. Much of that work is increasingly automated and done faster, while still aided by human oversight, he said.

Companies analyze their transfer pricing, or internal corporate transactions across borders, to make sure they’re comparable to market prices. Amazon uses generative AI to analyze more than 600,000 companies as part of the benchmarking analysis.

The AI tool could reduce the time to complete such an analysis by at least half, Felton said, citing initial testing.

Revenue deep dive

Amazon’s finance staff is using a generative AI chatbot to review and thoroughly analyze revenue and customer data. For example, finance leaders use natural language to create and query dashboards showing revenue on a weekly and monthly basis over time, Felton said. They can ask the agent why revenue rose last week, or why a customer recently changed purchasing habits or stopped or started buying.

AWS employees use the tool daily for generating internal finance reports, Felton said. “We were starting to play with it a year ago, but today it’s really working," he said.

Felton said he would have to triple-check such summaries a year ago because hallucinations, which are misleading or mistaken answers to a prompt, were frequent. He still double-checks the results, but the effort required in doing so has “gone way down."

Amazon said its finance team is still experimenting with using AI agents to help approve employee purchases of things like digital advertising, event production and software licenses from outside vendors.

Many companies are applying more generative AI to their finance work after nearly three years of experimentation. Large corporations like Amazon have more freedom to allocate resources to their AI efforts and move faster than smaller businesses, some of which are stuck doing finance tasks in a pre-AI landscape, said Mark D. McDonald, an independent consultant and former senior director at research and consulting firm Gartner.

The companies at the forefront of AI in finance “get to focus on nonstandard back-office things that are more tailored to their business and that they almost have to build themselves because there’s no off-the-shelf solution," McDonald said.

Product finances

Amazon agents are also assisting in estimating the return on investment and potential pricing of a new Amazon product that’s about to be built. The finance and product management teams, among others, for decades have wrestled with the drafting of a product document, known as PR/FAQ, sometimes over the course of weeks. The document lays out the business case and customer experience for a new product. Software teams used to then build the product based in part on that document.

Felton said it was a rigorous process to write the PR/FAQ. It still is rigorous, but AI agents are heavily involved. The teams input a few prompts and receive a close-to-finished product. AWS provides more than 240 cloud products.

“I can definitely see a world where a press release is iterated on with humans, but very quickly, and then it is handed off to software agents to go build that product and you can go from kind of idea to product very, very quickly," Felton said. “That is not science fiction in my mind. I think that we are close to seeing that world."

Felton said he’s excited to someday witness 15 or more AI agents interacting with each other to provide him insights and complete key finance tasks. “There’s a bunch of stuff I do today that I did not think I would be doing," Felton said, referring in part to using agents.

AI is already making the Amazon workforce more productive, Felton said, adding that it’s too early to say if it will cause a net reduction in jobs and declining to comment on potential cost savings. Jassy last year said the company had generated about $260 million in annual efficiency gains due to generative AI.

Felton said his job’s biggest constraint is time. “I just don’t have enough hours in the day to do all the stuff I want to do," he said. “A lot of what GenAI does is it gets rid of the muck, the maybe the not-so-fun parts of your job."

Amazon executives said AI is changing what finance talent looks like. “In addition to strong traditional technical skills, our finance leaders need to think creatively, adapt quickly and turn AI-generated information into meaningful business insights," Olsavsky said.

Eighteen percent of CFOs have eliminated finance jobs due to AI implementation, with the majority of them saying accounting and controller roles were cut, according to a survey released this month by executive search firm Egon Zehnder.

Almost every finance professional at a company will need to become more analytical and strategic, more akin to business advisers, as AI warrants less manual intervention, said Ash Mehta, a senior director analyst at Gartner.

“The skills that made finance professionals successful in the past may not make them successful in the future," Mehta said.

Write to Mark Maurer at mark.maurer@wsj.com