93% of Collections Agencies Are Using AI. The Other 7% Are Not the Problem.

ACA International's 2026 benchmarking data shows that 93% of collections agencies are now using or evaluating AI and machine learning technologies - up from 49% in 2023. That near-doubling in two years is one of the fastest technology adoption curves the ARM industry has seen.

It is also one of the most misleading numbers in the current collections narrative.

The framing that almost all agencies now "have AI" obscures the more consequential question: what does having AI actually mean? ACA's own analysis answers it directly. The firms pulling ahead in 2026 are not the ones that have AI. They are the ones that have embedded AI across the entire collections journey - account scoring, prioritization, consumer outreach, quality and compliance monitoring, negotiation support - rather than deploying it at one layer and calling it done.

The agencies at risk in 2026 are not the ones that skipped AI. They are the ones that put AI on the outreach layer, measured the contact-rate lift, and stopped.

The Konur Consulting take: 93% adoption is not the same as 93% transformation. The performance divergence that ACA's data documents is not a technology gap. It is an operating model gap - and it is widening.

What ACA's data actually shows

ACA's 2026 Convention and Expo materials, drawing on the organization's February 2026 benchmarking report, describe an industry at a fork. A few data points that matter:

• 93% of agencies are using or evaluating AI - up from 49% in 2023 and 73% in 2024
• 76% of companies plan to increase technology spending over the next two years
• 48% cite agent productivity and margin improvement as the primary motivation for that spending - down from 28% citing regulatory compliance in 2024
• 47% identified quality and compliance monitoring as their leading AI use case, followed closely by written communication at 47%, scoring and treatment strategies at 45%, and voice communication at 44%

The last data point is where the divergence becomes visible. The agencies deploying AI across quality and compliance monitoring, scoring, treatment strategy, and communication simultaneously are the ones building compounding advantages. The agencies that deployed a voice AI, measured outreach volume, and returned to their normal operating model are not in that cohort.

ACA's own language on this is precise: "AI/ML engagement increases with scale, but the real divide isn't just who uses AI/ML - it's how deeply they commit to it and whether they're building capabilities of their own."

The two-generation gap

The collections industry in 2026 is effectively running two generations of AI deployment simultaneously.

The first generation is AI-as-outreach-tool: automated voice agents, AI-generated text messages, AI-drafted collection letters. This is the layer most agencies reached in 2024 and early 2025. It produces measurable contact-rate improvements and is the technology that 93% of agencies now have in some form.

The second generation is AI-as-operating-model: AI embedded in the account-level decisions that precede and follow outreach. Which accounts get worked, in what sequence, with what treatment strategy, based on what scoring inputs. How QA is conducted on every consumer interaction rather than a sampled subset. How compliance monitoring catches problems before they become complaints. How negotiation support gives agents real-time data at the moment of consumer contact.

The gap between these two generations is not a technology gap - the tools for second-generation deployment are available to agencies of every size. It is an operating model gap. Second-generation AI deployment requires process redesign, not just tool addition. It requires decisions about data architecture, workflow integration, human-in-the-loop checkpoints, and performance measurement that cannot be made by a technology vendor and cannot be purchased off a shelf.

Why this gap is widening

Three dynamics are accelerating the divergence between first-generation and second-generation AI agencies:

Margin compression is not uniform. Agencies running AI at the outreach layer have improved contact rates but have not necessarily changed their unit economics on contested or thin files. Agencies running AI at the scoring and prioritization layer are working fewer files more effectively - routing resources to accounts with the highest recovery probability and reducing cost on accounts that would have underperformed regardless of contact intensity.

Compliance costs are rising for agencies without governance. The pattern ACA documented in June 2026 - AI agents contacting consumers about already-paid debts because of data sync failures - generates complaints, and complaints generate regulatory scrutiny. Agencies with AI governance infrastructure catch these errors before they become consumer interactions. Agencies without it learn about them from complaint logs and, increasingly, from state AG offices.

Creditor selection criteria are shifting. As creditors become more sophisticated about agency AI capabilities, the question is shifting from "do you use AI?" to "how do you use it, and how do you govern it?" An agency that can demonstrate AI-embedded QA, documented compliance monitoring, and audit-ready interaction logs is a different kind of partner than one that can demonstrate contact-rate lift from an AI dialer.

What the gap looks like in practice

The difference between a first-generation and second-generation AI deployment is visible at the account level:

In a first-generation deployment, the AI determines how to contact a consumer. In a second-generation deployment, the AI also determines whether to contact this consumer at all, now, and with what message - based on a real-time scoring model that incorporates payment probability, account age, dispute history, and portfolio recovery economics.

In a first-generation deployment, QA reviews a sampled subset of AI interactions manually. In a second-generation deployment, AI monitors every interaction for compliance flags, escalation triggers, and performance signals - and surfaces the cases that require human review rather than sampling randomly.

In a first-generation deployment, agents receive accounts with balance and contact information. In a second-generation deployment, agents receive accounts with AI-generated scoring context, suggested approach, real-time negotiation support, and settlement authority that adjusts based on consumer response signals.

The output difference is not marginal. It is the kind of compound operating advantage that closes the margin gap between the agencies growing share and the agencies holding on.

What to do Monday

• Score your current AI deployment against the full collections journey. Account scoring - outreach sequencing - consumer communication - QA and compliance monitoring - negotiation support. Where does your AI touch each stage? Where does it touch none?
• Identify the highest-value unworked layer. For most first-generation agencies, the scoring and prioritization layer offers the highest marginal return on AI investment - routing resources to recoverable accounts rather than spreading effort evenly across the portfolio.
• Measure the right things. First-generation metrics are contact rate, right-party contact rate, and promise-to-pay rate. Second-generation metrics are net recovery per worked account, QA defect rate, and complaint-to-interaction ratio. If you are measuring the first set and not the second, your operating model is still first-generation even if your technology is not.
• Define your governance layer before the next deployment. Every second-generation AI capability requires a corresponding governance decision: which outputs require human review, how that review is documented, and how errors are caught. Building the governance layer after the deployment is the pattern that generates complaints.

FAQ

If 93% of agencies are already using AI, what advantage is there in doing more?

The advantage is in depth, not presence. An agency using AI for outreach has a contact-rate edge. An agency using AI across scoring, QA, compliance monitoring, and negotiation support has a unit-economics edge. The first is measurable in months. The second compounds across every account worked.

Is second-generation AI deployment only available to large agencies?

ACA's data shows AI/ML engagement increases with scale, but it also notes that "smaller firms tend to sit on the sidelines or test vendor offerings." That is a choice, not a constraint. The tools for second-generation deployment are available to agencies of most sizes. The barrier is operating model redesign, not technology access.

How does this connect to the M&A consolidation signal in collections?

The agencies being acquired in current collections consolidation are disproportionately those with undifferentiated operations and thin margins. Second-generation AI deployment changes both - it improves margin by reducing cost per recovery and differentiates the operation by making AI-embedded QA and compliance monitoring visible to creditors. The agencies that will be acquirers rather than targets are the ones building this layer now.

93% adoption means the easy part is done. The compound advantage belongs to whoever builds the operating model underneath the technology.

Konur Consulting works with collections agencies to move from first-generation AI deployment - outreach automation - to second-generation AI operationalization: scoring integration, QA automation, compliance monitoring, and the operating model redesign that makes the compounding advantage sustainable. Reach out at info@konurconsulting.com to map where your current deployment sits and what the next layer looks like.

---

Source - ACA International benchmarking: ACA International, "Collection Agencies Report Record Account Volumes as AI Adoption Surges," February 2026. acainternational.org

Source - ACA 2026 Convention materials: ACA International, 2026 Convention and Expo Session Schedule, Orlando, July 22-24, 2026. acainternational.org