Deflection vs Resolution: How to Stop Counting Ignored Chats as Wins and How Chatbase Closes the Gap

TL;DR: Deflection vs. Resolution

• Deflection measures support requests handled without human agents; resolution measures issues solved by AI.
• A high deflection rate does not guarantee success. Track repeat contacts, CSAT, and escalations to spot unresolved issues.
• Calculate AI resolution rate by dividing independently resolved eligible conversations by total eligible AI-handled conversations and multiplying by 100.
• Improve resolution by refining knowledge sources, automating common tasks, and enabling smooth handoffs to humans.
• Chatbase automates support tasks, connects business systems, and monitors AI performance.

Deflection vs. resolution highlights a key challenge in AI-powered customer support: fewer tickets do not always mean fewer customer problems. Chatbots may keep requests out of the human support queue while leaving customers with unanswered questions, abandoned chats, or reasons to contact support again.

Deloitte Digital’s 2026 Global Contact Center Survey found that more than half of surveyed consumers said customer service quality stayed the same or worsened in 2025, despite leaders’ growing confidence in their customer experience performance.

So, what is the difference between deflection and resolution? Deflection measures how much support demand avoids human agents, while resolution measures whether AI actually solves customer issues.

This guide explains how to calculate AI resolution rate, spot misleading deflection metrics, and choose the right support KPIs. It also explores how Chatbase Helpdesk helps teams automate support, complete customer requests, and escalate complex issues with context.

What Is the Deflection Rate?

Deflection rate measures how many customer support requests are handled without reaching a human agent. This can happen when customers find answers in a help center, follow an FAQ, use a self-service portal, or interact with a chatbot instead of submitting a ticket.

For support teams, a higher deflection rate can mean fewer tickets for agents to handle and lower support costs.

For example, imagine 100 customers ask a chatbot about their order status. If 30 find the information they need without contacting a human agent, the team may record a 30% deflection rate, depending on how it defines and tracks deflection.

But there is a catch: avoiding a ticket doesn't necessarily mean solving the customer's problem.

Some customers may leave without an answer or return later with the same question. As Chatbase’s guide to AI in customer service explains, teams should look beyond deflection and track metrics such as resolution rate, re-contact rate, and customer satisfaction to see whether automation is genuinely helping.

What Is Resolution Rate?

Resolution rate measures the percentage of customer conversations an AI agent resolves independently, without creating a ticket or escalating to a human agent. A resolution can mean answering a product question accurately, checking an order status, processing a refund, updating a delivery address, or helping a customer change an order.

For example, if an AI agent handles 100 eligible conversations and independently resolves 18, its AI resolution rate is 18%. The other conversations may need further review to determine whether the customer received an answer, abandoned the chat, or required human assistance.

How to calculate AI resolution rate

AI resolution rate = (AI-resolved conversations ÷ Total eligible AI-handled conversations) × 100

To calculate this metric consistently, define what qualifies as a resolved conversation and which interactions count as eligible. An unanswered question, abandoned chat, or conversation that ends without a human handoff should not automatically count as resolved. Where possible, verify outcomes through completed actions, repeat-contact data, customer feedback, or conversation reviews.

Alongside resolution rate, track two additional metrics to understand AI support performance:

• Re-contact rate: The percentage of customers who contact support again about the same issue. A high rate can indicate that the initial response did not fully resolve the problem.
• Containment rate: The percentage of conversations handled entirely by the AI without transferring to a human agent. High containment can reduce agent workload, but it does not guarantee successful resolution.

Chatbase’s AI support analytics can help teams evaluate agent performance and identify opportunities to improve outcomes. Tracking these metrics together gives a clearer picture of whether an AI agent is reducing workload and solving customer problems.

Deflection vs Resolution: Side-By-Side Comparison

The difference between deflection rate and resolution rate comes down to what each metric measures. Deflection tracks how many support requests avoid human assistance, while resolution measures whether an AI agent successfully solves the customer’s issue. Both can reduce agent workload, but they indicate different levels of success.

For example, if an AI agent deflects 23 conversations but independently resolves only 18, the five-conversation gap warrants investigation. Some customers may have abandoned the chat, while others may not have received a complete answer. These figures are illustrative, and the gap depends on how each metric is defined.

To improve both outcomes, teams can use AI agents for customer support to automate routine requests while retaining clear escalation paths for complex issues. The goal is to reduce unnecessary tickets without mistaking fewer conversations for better support.

Why High Deflection Can Hide Bad Support

A high deflection rate only tells you that customers did not reach a human. It does not tell you why. A customer who found the right answer and a customer who gave up look exactly the same on the dashboard, and that is how high deflection can hide bad support.

Here is where deflected tickets often end up:

• Abandoned chats: the customer gets a vague answer, closes the window, and still counts as deflected.
• FAQ loops: the bot keeps sending the same help article, the customer rephrases the question, and the conversation ends with nothing solved.
• Hidden escalation paths: the human option is buried or missing, so the customer leaves and tries email, phone, or social media instead.
• Angrier repeat contacts: the same customer returns days later with the same problem, plus lost time and less patience.
• Silent churn: some never contact support again. They just cancel.

Why does this happen? Basic bots are rewarded for keeping people away from agents. Deflection is easy to measure and always looks good, and the less a bot hands off, the higher the number climbs, whether or not anyone was helped. Resolution is harder to measure, so teams often leave it out of the report.

A simple way to spot the problem is to watch what happens around the deflection number. If deflection rises while CSAT stays flat, repeat contacts increase, or phone and email volume goes up, the bot is moving tickets, not solving them. A useful check is whether the same customer contacts you again about the same issue within 48 to 72 hours.

That is why you should judge an AI customer support agent on outcomes. To spot these gaps, review abandoned conversations, repeat contacts, escalation patterns, and customer satisfaction alongside deflection rate. Chatbase’s guide to why AI customer support fails explores these risks.

AI Support Pricing Models Compared: What Counts as a Billable Resolution?

A billable resolution is a customer issue that meets an AI vendor’s criteria for a charge. However, not every AI customer support platform charges per resolution. Some use message credits, conversation volume, ticket volume, or user seats instead.

AI customer support pricing models compared

Here’s how that maps to the platforms buyers compare most often:

This is where the metric and the money meet. If a vendor bills per resolution and defines resolution loosely, you pay for the abandoned chats and unanswered questions this guide warns about. A credit or message model ties cost to work done rather than to a resolution count the vendor defines.

So before choosing a plan, clarify how a vendor treats repeat contacts, reopened issues, partial resolutions, and conversations that need human help. Without a consistent definition, pricing comparisons mislead.

Chatbase uses message-credit pricing on its self-serve plans, so cost tracks usage rather than a resolution count. Enterprise customers can also discuss per-resolution or per-conversation pricing where it fits their setup. See the Chatbase pricing page for options.

Resolution Pricing Benchmarks and Drawbacks

Per-resolution pricing connects AI customer support costs to the number of issues resolved. However, it is not automatically cheaper than ticket-based or credit-based pricing. The actual cost depends on the vendor’s rates, billing rules, and your support volume.

Three drawbacks are worth considering:

• Per-resolution pricing can be more expensive than per-ticket pricing. The cost per resolved issue may exceed the fee for processing a ticket, depending on the vendor and contract.
• Flat fees may ignore task complexity. A simple FAQ answer and a multi-step refund or account update may carry the same resolution fee, even though they require different levels of work.
• Higher resolution rates can increase total costs. When every successful resolution is billable, automating more requests can raise your monthly invoice even as human-agent workload falls.

How do you benchmark AI resolution pricing?

Compare total cost at your expected conversation volume and resolution rate, not unit prices alone. Include platform fees, usage charges, repeat contacts, and human escalations to estimate the effective cost per genuinely resolved issue.

For a practical framework, see Chatbase’s guide to cutting customer support costs without hurting customer experience. It explains why teams should evaluate cost savings alongside repeat contacts, service quality, and escalation patterns.

The best AI pricing model balances predictable spending with successful customer outcomes. A lower fee per interaction is not necessarily better if customers need multiple attempts or additional human support to solve the same issue.

6 Tips to Tackle the Deflection vs Resolution Gap

Closing the deflection vs. resolution gap doesn't start with a new tool. It takes a clearer definition of success and a steady habit of checking it. Start with these six.

Define resolution as a verified outcome, not a closed chat

A chat that ends is not a problem that is solved. Count a conversation as resolved only when the customer got their answer or action and did not come back about the same issue within 48 to 72 hours. Action: write that rule down and apply it to every report.

Pick one primary KPI and one or two guardrails

Choose resolution rate as the number you manage, then add guardrails such as repeat contact rate, CSAT, or escalation rate. This keeps a rising number from hiding a falling experience. Action: set a target for the primary KPI and a limit for each guardrail.

Track resolution by use case, not as one blended number

Billing questions, returns, and account changes behave very differently, and an average hides the weak spots. Break your AI resolution rate down by topic so you know exactly where customers still get stuck. Action: pull your top five ticket types and report each one separately.

Separate successful deflection from abandonment

A customer who got an answer and a customer who gave up both look deflected. Tag chats that end with no answer clicked, no thumbs up, and no follow-up, and count them apart from real wins. Action: add an abandoned label to your reporting.

Keep an easy path to a human for complex or sensitive issues

Billing disputes, cancellations, and upset customers need a person. Make the handoff visible and pass along the full conversation so customers never have to repeat themselves. Action: test the handoff yourself and fix anything that takes more than one click.

Review a sample of conversations every week and fix the gaps

Most failed answers trace back to missing or outdated knowledge. Read a small sample each week, note where the agent guessed or stalled, and update the source content. Tools like Chatbase Backstage group conversation themes and flag knowledge gaps, so you start with the biggest ones instead of whatever ticket someone happened to open.

Action: block thirty minutes weekly and fix the top three gaps.

Small wins add up quickly. Fix the definition first, report by topic second, and the gap between deflection and resolution starts to shrink.

How to Choose the Right KPI for Your Goal

The right AI customer support metrics depend on what you want to improve. Use one primary KPI to measure progress and one guardrail to catch unintended outcomes.

Why these pairs work:

• Cost: cost per resolved issue shows real savings, and repeat contact shows whether the savings hold.
• Satisfaction: CSAT shows how customers felt, and escalation shows where the AI fell short.
• Automation: deflection shows how much work the AI absorbs, and CSAT shows whether customers were still happy.
• First-contact resolution: FCR shows how often one conversation is enough, and repeat contact confirms it.
• AI effectiveness: verified resolution counts only solved issues, and escalation shows what the AI could not handle.
• Retention: retention shows the long-term result, and CSAT is an early warning if it starts to slip.

What metrics should you use to measure AI customer support success? Start with the business outcome, then check that automation improves it without creating new problems. For example, a higher deflection rate only helps if customers still get the help they need.

How Chatbase Delivers True Resolutions

True AI resolution means solving a customer’s issue, not simply ending a conversation without human involvement. Chatbase combines contextual answers, task execution, human handoffs, and performance insights to help support teams improve outcomes.

The Chatbase Agentic Harness: Built into the foundation

Chatbase’s Agentic Harness integrates the LLM, business context, APIs and integrations, CX capabilities, and security into one architecture, rather than bolting these capabilities on as afterthoughts. This foundation helps agents deliver context-aware answers, execute supported actions, enable human handoffs, and support enterprise requirements through security controls.

Resolve issues through actions

Chatbase agents go beyond explaining solutions by executing supported tasks, such as checking orders, initiating refunds, and making API calls through integrations like Shopify and Stripe. Procedures define plain-language workflows for common ticket types, helping agents handle customer requests more consistently.

Escalate without losing context

When an issue needs human assistance, Chatbase Helpdesk supports ticket creation and AI-to-human handoffs. Preserving conversation context helps human agents continue the interaction without making customers repeat the same details.

CSAT surveys, analytics and Backstage

After a ticket closes, Chatbase can email a Thumbs, Stars, or Smileys survey, and analytics rolls answers into a CSAT score, response rate, and coverage. Backstage audits your agent on a schedule and proposes tested fixes you approve before they ship.

Align pricing with resolution goals

Chatbase self-serve plans use monthly message credits, while Enterprise customers can discuss pricing based on credits, conversations, or resolutions, depending on the agreement. Compare the effective cost per genuinely resolved issue, including repeat contacts and human escalations, before choosing a model.

Explore Chatbase pricing and Enterprise options.

By bringing intelligence, context, integrations, customer experience, and security together, Chatbase helps teams focus on what matters: resolving customer issues accurately, escalating when necessary, and managing support costs.

A Simple Workflow to Raise Resolution With Chatbase

How do you set up an AI agent to resolve support tickets?

Start with your existing support content, automate a few repetitive requests, and improve the agent using real conversation data. You do not need to automate every ticket at once.

Train the agent on your best sources. Add your help center, website content, policy documents, and common support Q&As to Chatbase. Test the answers before deployment to identify missing or outdated information.

Customize instructions for accurate answers. Define how the agent should respond, which sources and policies to follow, what it must avoid, and when it should ask for clarification or escalate. Clear instructions help keep answers consistent with your support standards.

Add Actions and Procedures for common tickets. Start with your three most frequent requests, such as order lookups, billing questions, or refunds. Connect relevant tools through integrations or APIs, then use Procedures to guide the agent through multi-step workflows.

Set handoff rules before going live. Define when the agent must involve a human, such as for unresolved questions, billing disputes, or customer complaints. Enable Chatbase Helpdesk or connect your existing helpdesk so the conversation context carries over, and use CSAT surveys to collect feedback after ticket closure.

Review conversations and fix the gaps. Each week, check unresolved questions, failed actions, repeat contacts, escalations, and customer ratings. Update knowledge sources, refine instructions, and adjust Procedures based on what the conversations reveal. You should also revise any weak answers straight from your chat logs, and Chatbase saves the fix so the agent keeps improving over time.

Firecrawl achieved an 86% AI resolution rate while supporting more than 1.5 million users with one full-time support staff member. Its Chatbase agent connects to Stripe and internal systems to verify payment details and issue eligible refunds, showing how AI Actions and integrations can turn support conversations into completed tasks. Read the Firecrawl customer story to see how it works in practice.

Measure What AI Actually Resolves

Deflection and resolution measure two different aspects of AI customer support. Deflection shows how effectively you reduce human-agent workload, while resolution reveals whether customers get their issues solved. Tracking both alongside re-contact rate, CSAT, and escalation rate helps you identify gaps and avoid counting abandoned conversations as successful outcomes.

The goal is not simply to automate more conversations or reduce ticket volume. It is to resolve routine requests accurately, complete customer tasks wherever possible, and make human assistance easy to access when needed. With Chatbase, you can automate common support workflows, connect your AI agent to business systems, and hand off complex issues with conversation context intact.

Ready to improve your AI support outcomes? Book a demo to explore how Chatbase can help your team automate support while keeping customer outcomes at the center.

Frequently asked questions

1. How do you calculate AI resolution rate?

Calculate AI resolution rate by dividing the number of eligible conversations independently resolved by AI by the total number of eligible conversations, then multiplying by 100. For example, if an AI agent resolves 80 out of 200 eligible conversations, its resolution rate is 40%. Define resolution consistently and verify outcomes where possible, rather than counting every conversation that ends without a human handoff.

2. Is a high chatbot deflection rate always good?

No. A high deflection rate can indicate that self-service is reducing human-agent workload, but it does not prove that customers received the help they needed. Abandoned chats, incomplete answers, and repeat contacts can make deflection look successful when issues remain unresolved. Track deflection alongside AI resolution rate, re-contact rate, and customer satisfaction (CSAT).

3. What is the difference between AI resolution rate and containment rate?

AI resolution rate measures the share of eligible conversations that AI resolves independently. Containment rate measures the share of conversations handled without transferring to a human agent. A conversation can be contained without being resolved, so containment alone is not a reliable measure of customer outcomes.

4. How can you improve AI chatbot resolution rate?

Start by identifying common support requests that AI can reliably handle, such as order tracking, product questions, and routine billing queries. Improve the agent’s knowledge sources and instructions, connect the systems needed to complete supported tasks, and create clear escalation rules. Review unresolved conversations, repeat contacts, and CSAT regularly to identify where the agent needs improvement.

5. Does Chatbase charge per resolution?

Chatbase uses monthly message credits for its self-serve plans. Enterprise customers can discuss pricing based on credits, conversations, or resolutions, depending on the agreement. Review the Chatbase pricing page for self-serve options or book a demo to discuss Enterprise requirements.

6. Which metrics should you track alongside AI resolution rate?

Track re-contact rate, CSAT, escalation rate, and containment rate alongside AI resolution rate. Re-contact rate helps reveal whether customers return with the same issue, CSAT captures customer feedback, and escalation rate shows how often conversations require human assistance. Together, these metrics provide a more balanced view of AI support performance.

7. Can an AI agent resolve support tickets without human intervention?

Yes, when the request falls within its capabilities and it has access to the required information and tools. An AI agent can answer questions, retrieve order details, or complete supported workflows such as eligible refunds through connected systems. Complex, sensitive, or unresolved issues should have a clear path to human support.

8. How do you know whether an AI-resolved conversation was genuinely successful?

Define resolution around the customer’s intended outcome, not simply the end of a chat or the absence of a handoff. Where possible, verify that the requested action was completed, then review repeat contacts, customer feedback, and a sample of conversations. This helps distinguish genuine resolutions from incomplete answers or abandoned interactions.

9. What counts as a billable resolution?

It depends on the vendor, which is the problem. Some count any ticket the AI touched, some exclude handoffs, some re-bill on repeat contacts. Fin counts an implicit resolution if the customer goes quiet for 24 hours; Gorgias waits 72 hours with no human. Before signing, get the definition in writing: does a handoff count, and is a repeat contact billed twice?