MRS BPO and the money everyone gave up on

In 1991, Jeff Freedman was about to apply to law school when his brother Saul suggested a different occupation: collecting money that people had not paid. There was a small obstacle. The brothers had no clients, and the creditors they wanted to serve already had collection agencies. A new agency needed a reason to exist.

The brothers’ answer, as Jeff later recalled, was to ask for the rejects. Keep your existing agency. Send us the accounts it has stopped working. Pay us on contingency. A creditor could test MRS without replacing a trusted supplier. The pitch reduced the decision to something much smaller: would you let another firm try?

The second chance was the business

Many prospects declined, Jeff says. Some agreed. Those returned accounts produced recoveries, and the brothers began building a reputation. When regional banks consolidated and larger banks absorbed their clients, the relationships survived. In his telling, that continuity gave MRS its opening to grow.

There is a useful business lesson here. Asking a buyer to abandon an incumbent creates work and risk for the buyer. Offering to tackle neglected work gives the newcomer a place to demonstrate value. That is an interpretation of MRS’s origin, rather than a formula for every market. It depends on there being something recoverable in the rejected pile.

“Don’t fire your current agency.”JEFF FREEDMAN / RECALLING THE ORIGINAL PITCH

MRS expanded beyond that original niche. Its work includes first- and third-party collections, accounts receivable management and back-office services. Its customers span banking, telecom, retail and other credit businesses. In January 2018 it bought Alabama contact-center operator Vantage Sourcing, adding people and capacity to a business built around difficult conversations.

The person paying is not the person calling

A creditor hires MRS to recover overdue money. The consumer encounters MRS because an account needs attention. Those two relationships create a demanding design problem. A collection process must recover revenue while handling someone who may be anxious, short of money, disputing the balance or simply unwilling to talk.

The consumer website makes its offer concrete: account access, payment options, potential discounts and contact through text, an online portal or a representative. Available arrangements depend on the account. For a creditor, the service is outsourced collection work. For the consumer, its usefulness lies in being able to review options and manage payments without fitting everything into a phone conversation.

This places MRS between a creditor’s own collections department and the wider market of outsourced agencies and collection software. Its proposition combines operations with technology. A software vendor can supply a tool; MRS also undertakes the work surrounding the tool. Whether that combination wins depends on execution, account economics and the creditor’s requirements.

A collector named ADAM

MRS was experimenting with conversational automation before the recent generative AI rush. A SpeechTEK 2020 session description traced its virtual agent, ADAM, to an initial idea using a Raspberry Pi. That is an unexpectedly modest ancestor for a system designed to discuss payments, disputes and bankruptcy.

The company publicly introduced ADAM in August 2019, using IBM Watson technology. It described a conversational voice system that could handle payment arrangements and account notifications, including fraud and deceased-customer cases. After-hours availability was part of the appeal: the computer did not need the call center to be fully staffed.

In a June 2020 account, MRS executive Chris Repholz described consumers shifting their schedules during the pandemic. ADAM handled more calls across the day, including outside normal agent hours. Automation had found a practical use: making a service available when people were ready to use it.

The next upgrade cost control

Experience with one generation of automation did not mean MRS could build everything that followed. Jeff has acknowledged that the business lacked sufficient AI expertise to move as quickly as the founders wanted. Their response was consequential. In December 2024, they sold a majority stake to a technology-focused ownership group and remained involved as shareholders and advisers.

The disclosed strategic price was majority ownership. The rationale was access to capabilities the agency could not develop fast enough alone. Sandeep Behari became CEO. The combination also introduced friction: Jeff has described the difficulty of bringing two cultures together, with technology specialists pushing possibilities and collections veterans bringing operational judgment.

In 2026, MRS described a generative AI successor to ADAM and plans to retire the earlier system. The announced capabilities included identity checks, payment plans, settlements and handling several kinds of account issue. This is a company-described direction, with an important practical hurdle: a consumer must trust the conversation enough to participate.

A better conversation still needs an ending

Consider identity verification. MRS must establish who it is speaking with before discussing account details. The consumer may wonder why an unfamiliar company is asking personal questions. A fluent voice alone cannot resolve that tension. The workflow needs to explain itself, protect information and recognize when the ordinary payment path is inappropriate.

Compliance therefore belongs inside product design. Jeff describes checks and exception dashboards built into MRS’s processes. The transferable idea is straightforward: involve the people who know the difficult cases before automating the ordinary ones. A mistaken process becomes more consequential when software repeats it across many accounts.

The original contingency pitch also has a limit. There must be enough recoverable money to justify the effort. Today’s automation faces a related test: a convenient interaction cannot create disposable income. A payment plan that someone cannot sustain merely postpones the problem. The useful measure is what happens after the conversation ends.