Amazon leans into generative AI to manage its finances

The e-commerce giant initially adopted rules-based systems, one of the earliest forms of AI that use a set of rules to solve problems and make decisions, in its finance organization, which it then augmented with machine learning. Generative AI now assists finance employees with more complex analyses.

Its finance teams are turning to generative AI in areas such as fraud detection, contract review, financial forecasting, personal productivity, interpretation of rules and regulations, and tax-related work, moves in part aimed at reducing costs, boosting efficiencies and increasing accuracy, company executives said. These use cases are in a mix of experimentation and implementation stages.

“While experimentation and getting to know the technology are things that we really want to speed up, actually deploying this into production and making sure that we are in a well-controlled situation is very, very important for us," said Dave George, vice president of finance technology at Amazon.

More broadly, Amazon is shifting its focus to AI innovations, with the capabilities of its cloud-computing unit Amazon Web Services driving a surge in companywide sales for the most recent quarter.

Amazon has said it expects cash capital expenditures to meaningfully increase this year as it makes investments in technology infrastructure, specifically generative AI efforts. The company booked $13.9 billion in cash capital expenditures during the first quarter, up 6% from the prior-year period. Amazon Web Services comprised 17.5% of Amazon’s $143.31 billion for the quarter, up from 16.8% of $127.36 billion the prior-year period.

Value creation from AI

The tone has changed in recent months across industries for finance executives navigating generative AI, from fascination with technological advances to a serious focus on its application and generating value from it. Some of the largest tech companies continue to be more aggressive and more willing to adopt generative AI.

Amazon’s in-house tax personnel make use of the technology as well. The tax compliance team created a tool that uses generative AI to help validate inbound invoices for value-added taxes, which businesses typically pay in the different stages of production of a good or service.

When the team receives an invoice, the tool automatically checks the invoice’s validity before payment occurs, George said. Instead of manually validating these invoices, generative AI allows the team to automate the process, he said.

The many use cases

Generative AI also helps Amazon with the review of the many contracts it has with suppliers and customers. A tool takes into account Amazon’s standard reference data, policies and history of previous contracts to determine the changes needed to accommodate the contracts. AI then extracts key information for a more streamlined human review.

“There’s a benefit there because humans are not wading through very, very dense contracts," said George, who leads the finance organization’s two generative-AI working groups, one centered on innovation and the other on security.

The company trains models on vast data sets of financial transactions to detect and prevent financial fraud and anomalies. The models can identify patterns and irregularities that would be virtually impossible for humans to spot, said John Felton, chief financial officer of Amazon Web Services.

“This enhanced fraud detection capability not only protects our bottom line but also helps us ensure compliance," said Felton, who, along with George, reports to the company’s CFO, Brian Olsavsky, who has held the role since 2015. The company declined to comment on estimated cost savings.

These controls around transactions increasingly incorporate generative AI, and models that can interpret and explain the findings. Employees who were previously working with a random sample now focus on a list of the highest risk transactions enhanced by generative AI, George said.

Prior to generative AI, employees worked with a random sample, he said. “The humans now have a much more rewarding job," George said.

A potentially up-and-coming area of AI application for corporate finance is making complex queries to retrieve certain data; for example, the ability to quickly obtain information on sales to new clients in Mexico that subsequently placed orders within a month, said Mark D. McDonald, a senior director at advisory firm Gartner who is focused on AI in finance.

Slow march of AI

But for many companies, finance has been slow to adopt the technology compared with other parts of the business such as sales, marketing and customer support. “The tasks of finance are rooted in the analysis of numbers and generative AI is a technology rooted in the analysis of words," McDonald said, adding that machine learning is better suited for numerical analysis.

Amazon is one of the few companies that has built its own large language models from scratch. Amazon Web Services has said tens of thousands of businesses are using its AI app development platform, known as Bedrock. Most companies pay to use proprietary models from providers such as OpenAI and fine-tune those models as needed, or build their own generative AI tools using open-source models.

AWS employees use an AI-app builder in Bedrock to automate daily work tasks such as helping to write documents and summarize meetings, Felton said.

One of the big advantages of generative AI is the opportunity to grow the personal productivity of employees by allowing them to expand their skills and tackle more critical thinking, George said. Accounting and finance departments at companies across industries have long been areas replete with repetitive tasks.

But George also acknowledged a “valid concern" from employees on the technology’s potential effect on job security. “I think there’s a sweet spot where we can get the benefit of generative AI from a corporate perspective as well as a human experience perspective," he said.