How Lexitas Automated 46% of Daily Payment Processing with Boomi AI

A team of up to 20 accountants once spent their days manually matching thousands of lockbox payments to invoices. Today, an AI agent built on Boomi handles nearly half that work, with zero human involvement for most transactions.

A Growth Story That Outran Its Back Office

Lexitas is a leading provider of technology-enabled litigation support services, spanning court reporting, record retrieval, process service, and legal staffing. Since its founding in 1987, it has grown aggressively through acquisition, completing more than 53 deals, over 40 of them in the last six years alone.

That pace built the business, but it also built complexity. Dozens of disparate systems, siloed data, and financial workflows that were never rationalized across the enterprise. Nowhere was the strain more visible than in accounts receivable.

Each day, more than 2,500 payment lines flowed into the company’s main lockbox. Customers routinely paid multiple invoices in a single lump sum with no explanatory note. That left no clear trail for matching a payment to the right open invoices. Data was scattered across the ERP, subsystems, and shared mailboxes, so reconciliation fell to a team of 15 to 20 specialist accountants who dug through emails, ordering systems, and historical records to piece each payment together.

The volume was unsustainable, and the process was error-prone, while deep institutional knowledge of client payment behavior was being consumed by low-value, repetitive work.

The Idea That Came From Finance

The push for automation started on the finance side. Chief Accounting Officer Sherry Bourque identified the lockbox process as a high-value target and brought it to John Baker, CIO and CISO at Lexitas, who had already been building a more integrated technology environment. The two partnered to turn the concept into a production system.

Lexitas had worked with Boomi for around five years, initially to manage integrations across acquired companies and to deploy Boomi Data Hub for deduplicating records across 53 entities. That relationship made Boomi the natural partner for this project.

Boomi ran a proof-of-concept workshop on a pro bono basis; the pilot launched in mid-November 2025, and by February 2026, the solution was in full production. Baker was candid that AI’s reputation initially made the team cautious, but the phased approach won them over. “You worry about all the horror stories from AI,” he said. “Yet, the team got comfortable, the pilot went well, and the production launch went even better.”

How It Works

The solution runs on a multi-agent architecture orchestrated through Boomi AgentStudio, where specialized agents handle different sources. One checks the shared email inbox, another reads the incoming lockbox file, and others query the ERP and subsystems. The agents collaborate, comparing findings and reasoning across data points simultaneously to determine how each payment maps to its invoices.

At the intelligence layer sits Anthropic’s Claude, accessed via AWS Bedrock, with Amazon Textract handling optical character recognition of remittance documents. Boomi Data Hub provides the unified data foundation. Matched payments post directly into the company’s ERP. When the agent cannot confidently resolve a payment, it flags the exception and routes it to the AR team with a summary of its reasoning.

Governance was non-negotiable, especially given Baker’s dual CIO and CISO role. No payment data leaves the Lexitas environment, and none is used for external model training. Boomi Agent Control Tower delivers a full audit trail, real-time dashboards, alerts if agent behavior deviates from norms, and a kill switch to halt activity instantly. Accounting owns the process and retains final authority over every posting.

The Results

In production since February, the numbers are clear:

• 46% of daily payments handled by AI
• 30% fully automated, with zero human involvement
• 16% human-assisted, where the agent does the heavy lifting

Beyond the headline figures, Lexitas reports higher auto-posted rates, better matching accuracy, faster exception resolution, and a meaningful drop in manual burden, freeing specialists for higher-value analysis.

Phase 2 is already underway, extending the framework to credit card processing, a second bank’s lockbox, accounts payable and vendor payments, and workflows beyond AR.

Baker’s advice reflects that momentum. “AI is too strategic to wait on. So the move is to pick something bite-sized, do it quickly, then build from there,” he noted. His three lessons follow the same logic:

• Start small and move fast, since the pilot ran concept to production in about three months
• Model ongoing run costs, not just setup, because LLM inference accumulates at scale
• Treat data access and guardrails as the hardest part, since building the agent is the easy step.

What This Means for SAPinsiders

Let the line of business nominate the first agentic use case. At Lexitas, the CAO surfaced lockbox cash application as the target, and IT delivered it. SAP customers running SAP S/4HANA Finance should apply the same test: an agent built on SAP Business AI and Joule can post to the ERP, while finance retains ownership. Start where the pain is measurable, not where the technology is fashionable.

Make governance the architecture. Lexitas’ results held because no data left the environment, every agent action was auditable, and a kill switch existed from day one. SAP users pursuing agentic automation should demand the equivalent inside their landscape: audit trails, human-in-the-loop approval on financial postings, and controls that satisfy both the CIO and CISO before scale.

Fix the data foundation before scaling the agents. Lexitas’ hardest problem was clean, governed access to the right sources, not building the agent. That maps directly to SAP transformation where master data quality, deduplication across acquired entities, and unified data access determine whether AI agents can reason accurately across SAP and non-SAP systems. SAPinsiders should treat master data cleanup and integration as the prerequisite that enables each new agentic use case to reuse the same foundation rather than rebuild from scratch.