Hands off the task, eyes on the outcome
By implementing AI agents for autonomous invoice processing, an accounts payable team saves time, streamlines operations and enhances compliance.
THE SITUATION
As a global consumer products company evolved and expanded, its procurement policies and processes lagged. A lack of standardization across the growing business challenged the company’s accounts payable (AP) clerks and accountants, who frequently had to conduct time-consuming research before they could process an invoice for payment. The inconsistencies also left the end-to-end process prone to error and exposed the company to potential risk.
With a focus on enabling future growth and business, the company redesigned its AP department and worked with Deloitte to build a strong foundation that could support its migration to more modern tools. The AP department transformation began with reestablishing more standard processes and heightened controls, prior to an enterprise resource planning (ERP) upgrade and robotic process automation, augmented by Deloitte resources. But even though the infrastructure and tools were better, process issues persisted.
Artificial intelligence (AI) had long been part of the conversation between the company and Deloitte. The question was never “if” the AP department would implement the emerging technology, but “when” and “how.” Company leaders were following the advancements and believed the technology was ready for exploring how an AI-driven solution could finally address the research-heavy, error-prone processes plaguing the AP function. And, given the company’s growth trajectory, leaders were also eager to discover how AI could expand AP capacity. Implementing AI agents, built for purpose (without the need to sleep!), could enable the AP function to do exponentially more to help grow the business—without growing the back office.
The ERP upgrade wasn’t a requirement for an AI implementation, but with it operating in a steady state, company leaders believed the time was right to move AI from conversation into action. And because they felt the door was open to further innovation, rather than implementing autonomous processing in phases leaders chose to shift from “low-touch” to “touchless” all at once.
ONE INVOICE COULD TAKE 30 MINUTES AND ENDLESS EMAILS. THEN THEY WENT TOUCHLESS.
THE SOLVE
Deloitte’s Finance Transformation and Operate services teams worked together closely to address the client’s frustration with limited benefits from automated AP processes and the resulting delays they caused. They proposed Deloitte's Zora AI™ Perform AP Invoice Management solution, which can go beyond robotic process automation and inherent ERP functionality by using advanced AI/machine learning technology to understand a company’s specific processing challenges and the appropriate corrective actions.
To customize the solution for the client, Deloitte trained the AI models on the client's vendor invoice intake process, controls, policies and overall AP environment. This helped autonomous AI agents learn to “read” the information, even handwritten or nonstandard invoices, and identify vendor, business unit, currency, date and general ledger coding (if available). Since the company worked with both domestic and foreign suppliers, which require different tax treatments and coding, Deloitte also built a foreign-based data integration tool to process international tax lines appropriately.
The solution was implemented so that the company’s businesses could continue to send invoices to existing AP mailboxes to be routed to Zora AI. When invoices contained codes or edits added manually by internal subject matter specialists, the agent could make intelligent recommendations to automatically complete invoice fields.
Additionally, Zora AI provides a “confidence score” for the information it predicts or parses from an invoice, which helped build trust right from the start. As part of Deloitte’s Operate services, humans review and verify the agent’s recommendation, but business stakeholders can also use the score to determine whether additional reviews or manual updates are needed.
Throughout the process, accountants and AP clerks have enhanced visibility into which invoices remain open (and why), with Zora AI issuing automated reminders to help speed up the invoice cycle time. Most issues that previously required back-and-forth emails between Deloitte and company accountants can now be resolved directly within the system.
AI WAS BUILT TO CHASE DOWN THE ISSUES. HUMANS ARE ESSENTIAL FOR THE RESULT.
THE IMPACT
The combined Deloitte and client team anticipated an increase in the number of invoices processed from input through rejection or approval without human interaction. But the actual touchless processing rate—92%—exceeded expectation.
Ultimately, the team is targeting a touchless throughput of nearly 99%.
Why not 100%? Given the size of the company and the breadth of its facilities, the occasional outlier or aberration is inevitable due to human error (and human complexity). And the human element will always be key to the business.
The time savings are dramatic. Before implementing Zora AI, it might have taken as much as 30 minutes to process a single invoice. Now, the company has reduced its invoice processing time by an average of 50%–75%, to a matter of minutes or even seconds.
The compliance enhancements are meaningful. Zora AI reduces the risk of duplicate payments and helps support timely payments to suppliers and vendors. The AI agent enables end-to-end audit trails and visibility into invoice processing. Because the system was designed for continuous improvement, learnings from the first month were incorporated into a refresh. Zora AI has already intelligently adapted its approach to resolve a recurring invoice formatting error.
Zora AI supports higher-quality invoice processing. The company’s AP clerks and accountants are now spending less time chasing down general ledger (GL) codes, goods receipts, and purchase order (PO) mismatches—and the reduction in manual processes also lowers their risk of human error while recording numbers, dates or even vendors.
The company has been able to reduce invoice processing FTEs by 50%. As resources are redirected to more meaningful tasks, such as root cause analysis and heightened analytics, a lower-touch approach can unlock a higher return from work across the AP function.
LOWERING TOUCH AND RAISING CONFIDENCE
