What is a good deflection rate?

Published benchmarks put a good deflection rate between 40% and 60%, with 60–80% considered top tier and 80%+ aspirational. The honest answer is that a good deflection rate depends on the contact mix, the industry, and what's happening to CSAT and repeat-contact rate at the same time. A 70% deflection rate paired with rising repeat contacts and falling CSAT is a worse result than a 40% deflection rate paired with high resolution quality and stable CLV.

How do you calculate deflection rate?

The standard formula is: deflection rate (%) = (inquiries resolved by self-service / total inquiries) × 100. A contact center that handles 10,000 inquiries and resolves 3,500 through self-service has a 35% deflection rate. Stricter versions of the formula only count an inquiry as deflected when the customer did not return on the same issue within a defined window.

What's the difference between deflection rate and resolution rate?

Deflection rate measures how many inquiries were handled without a human agent. Resolution rate measures how many inquiries were actually resolved for the customer, regardless of who or what handled them. The two numbers diverge when self-service or AI ends a contact without ending the customer's problem.

What's the difference between deflection rate and containment rate?

Containment rate is typically narrower than deflection rate. Containment usually refers specifically to chatbot or IVR sessions that completed without transferring to a human, measured inside one channel. Deflection rate is the broader, organization-wide measure that includes self-service article views, IVR self-service, chatbot resolutions, AI agent resolutions, and other forms of agent-free handling.

Is deflection rate the right KPI for AI customer service?

Deflection rate is a useful operational input but a misleading primary KPI for AI customer service. It can rise while CSAT, repeat-contact rate, and customer lifetime value all fall, especially when the AI vendor's pricing rewards keeping customers away from human agents. Most modern measurement frameworks pair deflection rate with resolution rate, CSAT on the self-service interaction, and a longer-loop outcome like revenue per conversation or CLV.

Why do AI vendors emphasize deflection rate?

Many AI customer service vendors price per AI-resolved conversation, which makes deflection rate the metric most aligned with their revenue. That alignment is worth noticing when evaluating vendor dashboards, benchmarks, and case studies. Vendors with outcome-aligned pricing models — where revenue grows as customer outcomes improve, not as deflection holds steady — tend to publish a broader metric set.

How does deflection rate relate to customer lifetime value?

The two are linked through trust. A deflection that resolves the customer's problem completely is neutral to (or positive for) customer lifetime value. A deflection that ends the contact without resolving the issue can erode CLV through silent churn and reduced repeat purchase rate. Because deflection rate doesn't distinguish between the two cases, it should be read alongside CLV trends in the contacted cohort to know what it actually means.

What is deflection rate definition formula and benchmarks

Deflection rate is the percentage of customer support inquiries resolved through self-service channels — knowledge bases, FAQs, chatbots, AI agents, IVR — without a human agent. It is calculated as self-service resolutions divided by total inquiries, multiplied by 100, and it became the default scoreboard for self-service investment in the 2010s and for AI customer service tools in the 2020s. A team that handles 10,000 inquiries and resolves 3,500 of them without an agent has a deflection rate of 35%.

The metric is everywhere. Salesforce, Microsoft Copilot Studio, Qualtrics, and most AI vendors price, report, and benchmark on it. Industry guides put a "good" deflection rate between 40% and 60% and call 80%+ the top tier.

This page covers what deflection rate is, how to calculate it, what counts as a deflection, what the benchmarks look like, why the metric took hold, where it falls down as a primary KPI, and what to measure alongside it (or instead of it) when customer trust is part of the business case.

Deflection rate in one sentence

Deflection rate is the share of customer support inquiries handled without a human agent — typically expressed as a percentage of total inquiries.

The deflection rate formula

Deflection rate is almost always calculated the same way:

Deflection rate (%) = (Inquiries resolved by self-service / Total inquiries) × 100

Some teams use a stricter version that only counts a deflection when the customer did not return on the same issue within a fixed window (24 hours, 7 days). A few use a multi-factor version that adjusts for entitlement and success rate, popularized by ServiceXRG:

Deflection = Self-help events × Success rate × Intent rate × No-further-action rate

The multi-factor version is more honest because it strips out the customers who self-served, gave up, and then called anyway. The simpler version is what most dashboards report.

What counts as a deflection?

Self-service channels that typically count toward deflection:

Public help center articles and FAQs read in lieu of contacting support
Chatbots and AI agents that resolve the inquiry without escalating to a human
IVR self-service paths that complete the task (order status, balance check, appointment reschedule) without routing to an agent
In-product help, in-app assistants, and embedded answer widgets
Community forums and peer support where the customer's question is answered without a company employee

What doesn't count, in most definitions: an automated acknowledgment email, a chatbot that hands off to a human after two turns, or an IVR menu that ends with "press 0 for an agent."

A worked example

Imagine a retailer that took 100,000 customer contacts last month. The contacts broke down as:

38,000 customers self-served on the help center and never contacted support
22,000 conversations were resolved by an AI agent without a handoff
5,000 IVR sessions completed a task without an agent
35,000 inquiries went to a human

Total deflected: 38,000 + 22,000 + 5,000 = 65,000

Total inquiries: 100,000

Deflection rate = 65%.

That's on the upper end of the published "good" benchmark and the lower end of "top tier." Whether 65% is actually a good number depends on what happened next — to the customer, to revenue per conversation, and to retention — which is the part the formula doesn't tell you.

Industry benchmarks

Published benchmarks vary by source, but the rough consensus across Salesforce, Balto, eesel, Alhena, and others:

Range	Label	What it usually means
Below 20%	Underperforming	Self-service channels are not findable, not trusted, or not capable.
20–40%	Average	A working knowledge base and a competent chatbot.
40–60%	Strong	Mature self-service plus a deployed AI agent.
60–80%	Top tier	A well-tuned AI agent on most inquiry types.
80%+	Aspirational	Vendor marketing territory. Cross-check it against escalation rates, CSAT, and repeat-contact data before celebrating.

Benchmarks vary widely by industry, contact reason, and how strictly the team defines a deflection. A SaaS company with a strong product help center and a low-emotion contact mix will deflect a higher share than a luxury retailer whose customers contact support to talk to someone.

Why deflection rate became the default metric

Deflection rate took hold because it solved a real measurement problem. Self-service software is expensive to build and maintain. Leadership needed one number to justify the investment. Cost-per-contact made self-service look infinitely cheaper than agent-handled contact, so "the share of contacts we didn't have to staff for" became the proxy for ROI.

Three forces locked it in:

The cost-per-contact math. A typical agent-handled contact runs $5–$15 fully loaded; a self-service resolution is closer to a few cents. Multiplying a small percentage shift by a large contact volume produces an executive-friendly number.
AI vendor pricing models. Many AI customer service vendors price per "AI-resolved" conversation. Their commercial incentive is to count deflections high, which means every vendor demo, dashboard, and benchmark report puts deflection rate at the top.
Operational simplicity. Deflection rate is one number, calculated from data the contact center already has. CSAT, revenue per conversation, repeat-contact rate, and customer lifetime value require more plumbing.

The result is that for a decade, deflection rate has been the metric most often presented to CFOs and boards when AI and self-service investments come up. It's an easy story to tell. The harder question is whether it's the right story.

Where deflection rate falls down as a primary KPI

A high deflection rate can mean two very different things, and the formula can't tell them apart:

Version A — the good deflection. A customer searched for "track my order," found the tracking page in two seconds, got the answer they needed, closed the tab, and never contacted support. The company saved a contact and the customer was served.

Version B — the bad deflection. A customer tried to get a refund, was routed through a chatbot that misunderstood the question twice, hit a help-center article that didn't apply, gave up, and silently moved their business to a competitor. The company also "saved a contact." The dashboard shows the same green number.

The specific failure modes worth watching:

Silent churn. Customers who give up on self-service often don't escalate. They leave. Deflection rate goes up; retention goes down. The two trends are invisible to each other on most dashboards.
Repeat contacts. A chatbot interaction that "deflected" the first contact but generated three follow-ups doesn't show up as a failure in the deflection number. It looks like one deflection plus three new inquiries.
Vendor incentive misalignment. Vendor incentives can shape how success is measured. When pricing models are tied primarily to deflection, organizations should ensure they are also evaluating resolution quality, customer satisfaction, and repeat-contact rates.
Brand erosion in high-relationship verticals. Retail, hospitality, healthcare members, financial advisory, and luxury are not categories where "we kept the customer away from a person" is a thing to celebrate. The metric was built for help-desk volume reduction; it travels poorly into experiences that compete on relationship.

Deflection rate is a useful operational metric, but it becomes misleading when treated as the sole measure of customer-service success.

What to measure instead (or alongside)

If the point of self-service and AI is to serve customers well at scale, the metric set has to capture both halves of that sentence. Five additions cover most of the gap:

1. Resolution rate, not deflection rate. Resolution rate measures the percentage of inquiries that actually got the customer to a complete answer, regardless of who or what handled it. A resolution is a customer outcome; a deflection is an organizational outcome. The two diverge whenever self-service is finishing the contact without finishing the job.

2. CSAT on self-service interactions. Most teams measure CSAT on agent-handled tickets. Fewer measure it on the chatbot and help-center side. If a self-service interaction can't earn a 4-or-5 rating, calling it a deflection is generous.

3. Repeat-contact rate (or first-contact resolution). A self-service "win" that produces a second contact within 24 or 72 hours is not a win. Tracking repeat contact tied back to the original session is the cleanest way to catch the bad deflections the headline number hides.

4. Revenue per conversation. Especially in retail and ecommerce. A conversation that recommended a product, recovered a cart, or saved a cancellation is worth multiples of one that just answered a question. Optimizing on deflection alone leaves that value on the table.

5. Customer lifetime value (CLV) impact. The longest-loop measurement, and the one that surfaces silent churn. If deflection rate is going up and CLV in the contacted cohort is going down, the metric is lying to leadership.

The point is not to retire deflection rate. It's to stop treating it as the whole picture.

The Gladly view: devotion, not deflection

Our position on this metric is on the record. The category-leading framing across our product, pricing, and content is that customer service AI should be evaluated on customer outcomes — resolutions and assists — not on how many customers it kept away from a human.

The argument has two parts. First, the math of trust. Customers who trust a brand spend more, refer more, and churn less; a quick deflection that erodes trust is expensive even when it looks cheap. Second, the math of AI improvement. AI that assists a human agent generates the training data that makes future AI resolutions better. A vendor priced purely on deflection is incentivized to maintain a static metric rather than improve the underlying system.

Brands deploying AI on this model are not trading efficiency for relationships:

KÜHL runs a 59% AI resolution rate with a 120% lift in revenue per conversation.
Breeze Airways has AI enhancing 71% of conversations while maintaining high CSAT.
Smith Optics hits a 67% AI resolution rate on product-help and recommendation conversations — the kind of conversations most "deflection-first" stacks avoid because they're harder to automate.

The throughline: deflection rate optimizes for cost. Resolution-plus-assist optimizes for the customer relationship that drives cost down and revenue up over time. Gladly publishes the longer argument in this post.

How to use deflection rate without being trapped by it

For teams that have deflection rate already wired into reporting and don't want to rip it out:

Pair it with at least one quality metric. CSAT on the self-service interaction, first-contact resolution, or repeat-contact rate. If quality is moving the wrong way while deflection moves the right way, the headline number is hiding harm.
Segment it by contact reason. Password resets and tracking lookups should run a much higher deflection rate than refund disputes or service recovery. A single company-wide number averages those segments into incoherence.
Audit the bad deflections. Pull a sample of "deflected" sessions every quarter and read the transcripts. Count how many were genuinely resolved vs. abandoned. Most teams find the gap surprising.
Pressure-test vendor pricing. If an AI vendor is priced exclusively on "AI-resolved" conversations, ask how they handle a session that resolved a customer surface-level but generated a repeat contact two days later. The answer is informative.
Watch the customer outcome, not the dashboard. Retention, CLV, and net revenue retention in the contacted cohort will tell the truth about whether self-service is working. Deflection rate alone won't.

What is deflection rate?