What is automated resolution? The metric that proves your AI is actually working

Every AI customer service vendor leads with a number. Seventy percent automated. Eighty percent contained. Ninety percent deflected. There is one question those numbers almost never answer: did the customer actually get what they came for?

A conversation that ends isn't the same as a conversation that's resolved. It's why automated resolution rate has become the north star for enterprise CX teams burned by AI that looked good in demos and underdelivered in production.

This guide covers everything you need to know about automated resolution: how it's defined, how it's measured, what automated resolution rates are realistic by industry, and where the metric is heading next.

What is automated resolution?

An automated resolution (or AR) is a conversation between a customer and an AI agent that reaches a successful outcome without involving a human agent, and meets three criteria:

1. Relevant: The AI agent understood the customer's actual inquiry and responded directly to it. Not a redirect to an FAQ page. Not a generic response that happened to close the ticket.
2. Accurate: The response was correct and consistent with current policies, knowledge, and connected systems.
3. Safe: The agent engaged appropriately and on brand.

All three must be met. A conversation that avoids a handoff but gives a wrong answer doesn't count. A response that was technically accurate but off-topic doesn't count.

Automated resolution vs. containment vs. deflection

These terms appear constantly in AI customer service platform evaluations. They are not interchangeable.

The numbers tell the story

Across +350 global AI deployments:

• Containment rate averages around 72%
• Automated resolution rate averages around 52%, with best-in-class rates reaching 84%+

These are measured on the same conversations. That 20-point gap is conversations that look resolved by one measure and aren't by the other. For most enterprise contact centers, that's where the real problem lives.

Containment can be gamed by blocking escalation paths. It's a ceiling, not a result. Automated resolution requires a genuine outcome.

Can AI actually resolve complex multi-step issues, not just FAQs?

The concern is valid for the previous generation. Scripted and flow-based chatbots answer questions, they don't execute workflows. They have no path into backend systems, and no ability to adapt when a conversation takes an unexpected turn.

Autonomously resolving a complex issue—a flight rebooking, a billing dispute, a ride that won't end because GPS failed—means understanding what the customer needs across multiple steps, connecting to the systems that hold the relevant data, and taking action based on what those systems return. That requires reasoning, not scripting. And it requires deep integration, not surface-level API access.

Ada is built for both. The Reasoning Engine™ interprets each conversation dynamically, not following a fixed path, but responding to what's actually happening. Playbooks give teams a structured way to define complex workflows that connect to live systems, execute reliably, and improve over time.

Examples of complex, multi-step issue resolution by industry

Dott, a micromobility operator handling 2 million customer contacts a year across 400+ cities, built 25 API-powered automations, including a workflow that resolves one of their most operationally complex issues in real time.

When GPS fails and a rider can't end their trip, Ada interprets the request, determines what's needed, and triggers the API call to close the ride directly. As Dott's COO Nicolas Gorse put it: "No human can instantly end your ride. It's only possible with Ada." Dott's automated resolution rate went from 32% to 77%, with AR more than doubling within months of launch.

AR rate reflects this at the workflow level. The overall rate is an average across every use case in scope. Complex transactional workflows pull it down early, then climb just as quickly as integrations deepen. The rate for a specific workflow tells you exactly what's working and exactly what to fix.

Autonomous resolution across voice, email, and chat

Automated resolution is a channel-agnostic standard—relevant, accurate, safe, without human involvement—whether the conversation happens over chat, email, or voice.

Ada’s Agentic Customer Experience (ACX) Platform acts as a single point of orchestration across all channels. There's no separate voice system running different logic, no email bot operating on outdated information. When you update a policy, close a knowledge gap, or refine a workflow, it propagates everywhere at once. That means improving AR on one channel lifts the others, and your team maintains one system instead of managing each channel separately.

The business case: what automated resolution is actually worth

The standard ROI model for AI customer service goes like this: count the conversations that didn't reach a human agent, multiply by cost per ticket, declare victory. It's a useful number, but an incomplete one.It treats every customer conversation as a cost to be avoided.

When AI truly resolves conversations, something changes. Contact volume stops being a threat and starts being a signal. Every conversation becomes a data point: what customers need, where products fall short, where policies are creating friction, where revenue is being left on the table.

The organizations that see highest returns from AI aren't the ones who reduced contact the most. They're the ones who turn every conversation into a revenue event:

• A rebooking done right is a loyalty retention.
• An account recovery done right is a churn prevention.
• A billing question answered on the first try is a satisfaction signal that compounds over time.

The question isn't just how many conversations didn't reach your human agents. It's how much value the conversations that did get resolved actually generated.

What that looks like in practice: The IPSY story

IPSY's results from their first four months with Ada: 64% increase in automated resolution rate, 41% improvement in CSAT, 943% return on their AI investment.

Those numbers are real. But TJ Stein, Head of Customer Care at IPSY, is deliberate about how he frames them. "This isn't a cost-cutting story," he said. "It's a strategic reinvestment story. Every dollar we save through efficiency gets channeled back into elevating the member experience. That's how you turn Customer Care from a cost center into a growth driver."

What that reinvestment looks like in practice: IPSY created a dedicated AI strategy role and is growing the ACX team. Anna Prince now leads the program full time, running cross-functional partnerships with Engineering to deepen integrations and expand what the AI agent can do. The team is exploring AI-powered commerce flows to drive cross-sell and upsell directly inside support conversations. And IPSY's Customer Care team has become the internal benchmark for AI deployment across the company—other departments are now following their lead.

"The executive team is paying attention to more than just bottom-line savings," TJ said. "They're seeing a fundamental shift in how efficiently we operate, and that's opened up strategic conversations about AI deployment across the entire company."

That's what the shift from containment to resolution actually unlocks. Not a smaller team doing the same work, but a more capable team doing different work, and a business that finally has the data, the capacity, and the confidence to grow into it.

Beyond automated resolution: what comes next

Automated resolution is the right metric for AI customer service today, but we can do better. AR labels an entire conversation with one pass/fail verdict, but conversations often have more than one intent.

An intent is a single thing a customer is trying to accomplish—check a balance, rebook a flight, update an address. A conversation that resolved three out of four customer intents is counted the same as one that failed on the first question.

Ada's answer is Intent Resolution Rate (IRR). Rather than scoring a conversation as a single pass/fail, IRR decomposes it into individual intents and scores each one independently. A four-intent conversation gets four scores. A failure on intent three surfaces a specific, actionable signal tied to a specific surface: a Playbook to fix, a knowledge article to update, an integration to deepen.

Why intent resolution is the next evolution for automated resolution

IRR changes performance tracking and analysis in three main ways.

As Ada's CPTO Mike Gozze put it: "AR told you when a conversation ended cleanly. IRR tells you when the customer's life got better." IRR is Ada's near-term direction, being developed and deployed with customers now.

Frequently asked questions

What is automated resolution in AI customer service?

A conversation resolved by an AI agent without human involvement that was relevant to what the customer asked, accurate with respect to company policies and systems, and safe. All three conditions must be met. Avoiding a handoff alone doesn't count.

What is automated resolution rate (AR rate)?

The percentage of customer conversations an AI agent resolved without human involvement, in a way that was relevant, accurate, and safe. Formula: AR rate (%) = resolved ÷ (resolved + not resolved). Industry average: 52%. Ada customers at mature deployments have reached up to 84%.

What's the difference between automated resolution rate and containment rate?

Containment tells you a human wasn't involved, not whether the issue was solved. It measures an exit, not a resolution, and it can be gamed by blocking escalation paths. Automated resolution requires a genuine outcome.

How does Ada measure automated resolution?

Ada uses large language models to automatically evaluate a statistically significant sample of closed conversations after each one ends. Each conversation is assessed against three criteria—relevant, accurate, and safe—and labeled either resolved or not resolved. A conversation must meet all three criteria to count as an automated resolution.

Ada's AI scientists validated this methodology to an error margin of 0.06%, meaning the AR rate reflects what is actually happening in a deployment, not an approximation.

The process is fully auditable. AI Managers can see the reasoning behind every individual label, drill into specific transcripts, flag cases where they disagree with the model's assessment, and submit feedback that trains the evaluation model over time. Every correction flows back into the model, so measurement accuracy improves as the deployment matures.

Ada's AR measurement is designed to be transparent and resistant to gaming—the same standard applied consistently across every conversation, with no black-box scoring.

What's a realistic AR rate for an enterprise company?

Across +350 global AI deployments: containment rate averages around 72%, and automated resolution rate averages around 52%. Ada’s best-in-class retail and ecommerce customers have reached 80–85% or higher for high-volume transactional use cases. Ada’s best-in-class fintech customers have reached 84% AR rate. Ada’s best-in-class travel customers have reached 77% AR rate. Most enterprise deployments launch in the 50–65% range and improve as knowledge coverage, Playbook complexity, and backend integrations develop.

Can AI resolve complex, multi-step issues or just FAQs?

Scripted or flow-based chatbots can't. They can only answer questions, not execute workflows. Ada's architecture handles multi-step, multi-intent conversations that require real action in backend systems: rebooking flights, processing refunds, recovering accounts, resolving billing disputes.

Will CSAT decrease if automated resolution increases?

The data shows the opposite. eSky's AI CSAT rose 19 points in the same period their AR rate climbed 17 points. Customers care how fast and accurately their problem was solved. When automated resolution is genuine, CSAT follows.