Research report · June 2026

The 2026 AI Resolution Benchmark

How often do AI agents actually resolve customer conversations, not just deflect them? We measured 140 million conversations across 2,100 companies, by industry and by channel, and tracked what happened to cost per contact.

Download the full report Read the key findings

48 pages · Free · Anonymized, aggregated data from customers who opted in to benchmarking

Pazau Research2026

The 2026

AI Resolution
Benchmark

Resolution rate by industryn = 140M

140M conversations · 2,100 companies48 pp

140M

conversations analyzed, July 2025 to June 2026

2,100

companies across 9 industries and 31 countries

67%

average AI resolution rate under a strict 7-day definition

−41%

median change in cost per contact after 12 months

Key findings

Six things the data says about AI in support.

Every rate on this page uses the same definition of resolution: no human involved, no repeat contact on the same issue within 7 days, and no abandonment.

Finding 01

76% → 48%

Industry sets the ceiling, but it isn't destiny.

Logistics and retail resolve the most, because most of their contacts are “where is my order” and returns. Healthcare and financial services resolve the least, with more identity checks, more systems and more regulated steps. The spread inside each industry is wider than the spread between them: the top-quartile insurer resolves more than the median retailer.

AI resolution rate by industry

AI resolution rate by industry
Segment	Value
Logistics	76%
Retail & e-com	72%
Mobility	70%
Hospitality	66%
Telecom	64%
Airlines & travel	61%
Insurance	55%
Financial services	52%
Healthcare	48%

Finding 02

+13 pts

Voice is catching up fastest.

Messaging channels still lead, and asynchronous channels give the agent time to call systems and check policy. But voice resolution rose from 41% to 54% in a year, the largest gain of any channel, as voice-to-voice latency fell below 800ms and voice agents started running the same procedures as chat.

AI resolution rate by channel

AI resolution rate by channel
Segment	Value
WhatsApp	73%
Web chat	71%
Email	68%
SMS	66%
Voice	54%

Finding 03

−41%

Cost per contact falls by about 40% in year one.

Fully loaded cost per contact (people, tooling, telephony, outsourcing and AI fees, divided by all contacts) fell 41% at the median customer after 12 months. The absolute savings are largest in regulated industries, which start with the most expensive contacts.

Median fully loaded cost per contact (USD)

Before PazauAfter 12 months

Median fully loaded cost per contact (USD)
Industry	Before Pazau	After 12 months
Logistics	$4.90	$2.60
Retail & e-com	$5.80	$3.20
Hospitality	$8.10	$4.90
Airlines & travel	$9.40	$5.90
Financial services	$11.20	$7.10
Insurance	$12.60	$7.80

Finding 04

1.9×

Agents that can act resolve almost twice as much.

Conversations where the agent could take an Action, such as issuing a refund, changing a booking or updating an address, resolved at 74%. Where it could only answer questions, the rate was 39%. The biggest predictor of resolution rate in the dataset isn't the model. It's how many systems the agent is allowed to touch.

AI resolution rate by agent capability

AI resolution rate by agent capability
Segment	Value
Answer-only conversations	39%
Conversations with Actions	74%

Finding 05

14 pts

Deflection overstates resolution by 14 points.

Counting every conversation that didn't reach a person, the way most deflection metrics do, gives 81%. Removing customers who came back within 7 days, or who abandoned mid-flow, brings it to 67%. If a vendor quotes a number without a definition, ask which one it is.

The same conversations, three definitions

The same conversations, three definitions
Definition	Rate
Deflected: never reached a person	81%
…and not abandoned mid-procedure	72%
…and no repeat contact within 7 days	67%

Finding 06

4.58

Customers rate AI resolutions as highly as human ones.

AI-resolved conversations averaged 4.58 out of 5 CSAT, against 4.49 for conversations resolved by people. That comparison flatters AI slightly, because harder conversations go to people, but the gap holds within the same intent. Speed explains most of it: the median AI resolution took 3 minutes 40 seconds.

4.58CSAT, AI-resolved

4.49CSAT, human-resolved

3m 40smedian time to AI resolution

3.1%of AI resolutions reversed by a 7-day reopen

Methodology

How we built the benchmark.

The report was produced by Pazau's CX Research team, led by Jess Alvarez, and reviewed by our Data Protection Officer. Questions about the method, or requests for cuts we didn't publish, go to press@pazau.com.

Read more about how we define resolution in Resolution, not deflection and how we calculate cost per contact in The real cost per contact.

The dataset

140 million conversations handled on Pazau between July 1, 2025 and June 30, 2026, across 2,100 customers, 9 industries, 5 channels and 31 countries.

Opt-in only

Data comes only from customers who opted in to benchmarking. Opting out is a single setting and removes a workspace from all future reports.

Anonymized and aggregated

We computed metrics on counts and outcomes, never message content. No segment includes fewer than 25 companies, and no single company exceeds 10% of any segment.

Definitions

Resolution uses Pazau's billing definition. Cost per contact is fully loaded and compares each customer's 3 months before go-live with months 10 to 12 after.

Full report

Get all 48 pages.

The PDF breaks every finding out by company size and region, and adds resolution rates for the 40 most common intents.

Resolution rate for 40 intents, from order status to card disputes
Cost per contact by industry, company size and region
How top-quartile teams structure procedures and approvals
A worksheet to benchmark your own team in an afternoon

Where would your team land?

Bring 1,000 of your real conversations. In 30 minutes we'll show you how many Pazau Agent would resolve, measured by the same definition as this report.

Book a demo See pricing