A six-month pilot, a minimal integration, and a customer who had already been burned once.
By month six, Krew’s agents had engaged nearly 100,000 consumers, resolved more than $2M in past-due balances, and freed the equivalent of 192 hours of human-agent capacity every week.
The customer is a leading BPO specializing in servicing and collections for non-credit accounts: consumer rentals, municipalities, utilities, and the like. In a segment defined by thin margins and high volumes of low-dollar accounts, cost efficiency is not a nice-to-have, it is the whole game. Every dollar spent servicing a small balance eats directly into unit economics, so the BPO has long treated operational efficiency as a core competitive advantage.
That mindset made them an early adopter of new technology. A technology-forward BPO, they owned and managed their own technology stack, giving them the ability to evaluate, integrate, and pressure-test new systems on their own terms.
Their appetite for innovation led them to adopt AI early. Their first attempt, with a long-established incumbent vendor, was a failure, and not a marginal one. Over the life of that engagement, consumer NPS fell sharply in servicing conversations. For contingency accounts, the vendor barely recovered enough in gross dollars to cover the cost of the system. The agent was neither intelligent nor empathetic, could not negotiate effectively, and resolved very few outstanding accounts. Despite promising the BPO the ability to cancel if the product was not working, the vendor refused to allow it and insisted on a cancellation fee totalling the rest of the contract value.
That experience left a mark. The BPO became deeply skeptical of AI agents for servicing and collections, a domain where tone, compliance, and consumer trust carry real weight, and where a bad interaction can do more harm than no interaction at all. The skepticism went all the way to the top: the Chief Information Officer was a net detractor on AI.
“I was very much against using another AI vendor at the beginning. I fought really hard against using Krew at first.”
Chief Information Officer, BPO servicing 2M+ accounts
Start small, measure directly, and let the numbers make the case
When the BPO was introduced to Krew, the initial reaction was hesitation, and understandably so. They had been burned once, and their most technical decision-maker was the hardest to convince. Rather than push for a large commitment, the engagement started small: a limited contract that let the BPO evaluate Krew on real accounts, with real consumers, against outcomes they could measure directly, without betting the business on an unproven vendor.
The customer went live with a minimal-lift integration, a file transfer over SFTP, with the aim of proving out a single metric in the first month: can the agent bill enough in contingency fees alone to justify its own cost?
The results were immediate. During the limited six-month pilot, the BPO saw efficiency gains equivalent to 192 hours of human-agent capacity every week, for a fraction of the cost. Krew’s agent engaged nearly 100,000 consumers. In first-party servicing, it directly resolved more than $2M in past-due balances. In contingency collections, it cost less than 3¢ per dollar liquidated.
Hours of human-agent capacity freed each week
Attributed to Krew, by month of the pilot.
Consumers engaged
0K
Past-due resolved
$0M+
Cost per $ collected
<3¢
$37 collected for every dollar spent
On contingency collections, the program returned $37 in collections for every dollar spent on Krew’s agents, a 9x return once liquidation uplift and labor savings are counted together, on a net liquidation uplift of 4%.
How it was measured. Results reflect collections on past-due balances of more than $70M across 79,428 accounts. Return is calculated as incremental net cash plus realized labor cost savings from reduced headcount, divided by AI program spend. Incremental net cash is derived from the 4% net liquidation uplift measured during a 12-week randomized comparison against a control group. Labor savings are based on fully loaded cost and were realized after deployment as manual workload fell on outreach the platform handled.
What a dollar of AI spend returned
Uplift measured in a 12-week randomized comparison.
Collected per dollar spent on Krew’s AI agents
$0
Return on the program
0x
Liquidation uplift plus labor savings
Net liquidation uplift
0%
Against the control group
“With our previous vendor, nothing worked. There were no results. With Krew, we saw a clear uplift within days.”
Chief Information Officer, BPO servicing 2M+ accounts
Half of servicing requests resolved with no person involved
In first-party servicing, 51% of consumer requests were resolved without any human intervention, and 82,000 end-to-end workflows were executed by the agent alone.
How it was measured. Results reflect first-party servicing across roughly 20,000 consumer accounts. A request counts as resolved when the conversation was completed without live-agent involvement and the consumer either confirmed their inquiry was resolved or rated the interaction 8 or higher on a 0 to 10 post-call survey. End-to-end workflows are full-scope tasks, as defined by the BPO, completed without live-agent intervention.
Requests resolved with no person involved
Resolved means the consumer confirmed it, or rated the conversation 8 or higher out of 10.
End-to-end workflows executed
0K
Full-scope tasks, as defined by the BPO, completed without a live agent
“Krew is the best applied AI partner we’ve ever worked with.”
Chief Information Officer, BPO servicing 2M+ accounts
Prepared by
Michael Goh
Cofounder, Member of Technical Staff
Formerly led speech benchmarking at Artificial Analysis. Previously a FinTech VC and Bain consultant. MS Computer Science, University of Chicago; BA Economics, University of Oxford.
Rafael Khaykin
Cofounder, Member of Technical Staff
Formerly a research engineer at Figure Robotics, working on dexterity algorithms and latency research. Computer Science, University of British Columbia.
This case study describes one customer’s deployment over the period stated, measured as described above. The results are specific to that customer’s portfolios, workflows, and configuration; they are not a prediction of what any other deployment will show, and actual results will vary by portfolio, configuration, and context. Nothing here is a guarantee of performance. Figures are aggregated, identify no individual consumer, and are published with the customer’s agreement. The customer and its previous vendor are not named.
This document is provided for informational purposes only and does not constitute legal or regulatory advice. Motivated by Krew’s Collaborative Assurance Framework’s principles of shared accountability and streamlined assurance, we partner with our customers, combining our AI-driven credit servicing platform’s built-in security safeguards with each customer’s own controls, system configurations, and user training, to help manage regulatory risk. It remains the responsibility of each customer to ensure compliance with all applicable federal, state, and local statutes. All rights, responsibilities, and liabilities of Krew in relation to its customers are governed exclusively by the terms of Krew’s customer agreements. This document neither forms part of, nor alters, any contractual agreement between Krew and its customers.