Voice AI Index

Oct 1, 2026

Consio Voice AI Index

What thousands of ecommerce conversations tell us about why customers call, and what they actually want.

AI minutes analyzed0▲63.0%vs Aug

Conversations analyzed0▲85.4%vs Aug

Shopify brands0▲22.5%vs Aug

Trusted by leading Shopify brands

Key findings

This month in three numbers

The figures that stand out across every AI conversation we analyzed.

Key findings

This month in three numbers

The figures that stand out across every AI conversation we analyzed.

1 in 5

calls is about an order delivery issue, the #1 reason customers call.

66%

of conversations are resolved fully or partially, as rated by our LLM judge.

3.75 / 5

average quality rating, scored by our LLM judge on every judged call.

Customer issues

Why customers call

The problem behind each call, as classified by our LLM judge on every conversation.

Customer issues

Why customers call

The problem behind each call, as classified by our LLM judge on every conversation.

Share of conversations by issue

% of all calls · hover a row for the most common details

This monthAug

Order delivery issue

18.7%▲4.3 pts

Other

14.1%▼0.2 pts

Unexpected subscription charge

12.0%▲7.9 pts

Product knowledge gap

9.6%▼0.9 pts

Product technical issue

7.3%▼5.6 pts

Return or exchange issue

5.4%▼2.4 pts

Order content issue

4.8%▼1.4 pts

Payment issue

4.5%▲0.1 pts

Account issue

4.5%▲2.4 pts

Checkout issue

4.3%▼0.2 pts

Source: Consio Voice AI Index · Sep 2026

When they call

Calls by day and hour, customer local time

00061218MonTueWedThuFriSatSun

FewerMore

Outside 9-to-616%

On weekends9%

Source: Consio Voice AI Index · Sep 2026

When they call

Calls by day and hour, customer local time

00061218MonTueWedThuFriSatSun

FewerMore

Outside 9-to-616%

On weekends9%

Source: Consio Voice AI Index · Sep 2026

Customer goals

What customers actually want

What the caller wants to get done, and how often the AI gets them there.

Customer goals

What customers actually want

What the caller wants to get done, and how often the AI gets them there.

Ask for a human11.0%

Only 1 in 9 callers asks to speak to a person.

Easiest for the AIInitiate return or exchange · 80%

Hardest for the AITrack order · 61%

Source: Consio Voice AI Index · Sep 2026

Customer goal

% of conversations · share resolved end-to-end by the AI

GoalShare of calls%Resolved by AI

Manage subscription

12.5%

74%

Get explanation

9.3%

66%

Place order

8.6%

61%

Track order

8.5%

61%

Speak to support specialist

7.4%

63%

Initiate return or exchange

7.0%

80%

Get refund

6.9%

66%

Get product information

5.5%

71%

Speak to a human

11.0%

Routed to an agent

Source: Consio Voice AI Index · Sep 2026

Ask for a human11.0%

Only 1 in 9 callers asks to speak to a person.

Easiest for the AIInitiate return or exchange · 80%

Hardest for the AITrack order · 61%

Source: Consio Voice AI Index · Sep 2026

Customer goal

% of conversations · share resolved end-to-end by the AI

GoalShare of calls%Resolved by AI

Manage subscription

12.5%

74%

Get explanation

9.3%

66%

Place order

8.6%

61%

Track order

8.5%

61%

Speak to support specialist

7.4%

63%

Initiate return or exchange

7.0%

80%

Get refund

6.9%

66%

Get product information

5.5%

71%

Speak to a human

11.0%

Routed to an agent

Source: Consio Voice AI Index · Sep 2026

Ask for a human11.0%

Only 1 in 9 callers asks to speak to a person.

Easiest for the AIInitiate return or exchange · 80%

Hardest for the AITrack order · 61%

Source: Consio Voice AI Index · Sep 2026

Customer goal

% of conversations · share resolved end-to-end by the AI

GoalShare of calls%Resolved by AI

Manage subscription

12.5%

74%

Get explanation

9.3%

66%

Place order

8.6%

61%

Track order

8.5%

61%

Speak to support specialist

7.4%

63%

Initiate return or exchange

7.0%

80%

Get refund

6.9%

66%

Get product information

5.5%

71%

Speak to a human

11.0%

Routed to an agent

Source: Consio Voice AI Index · Sep 2026

AI performance

How the AI performs

Every conversation is scored with the same rubric, every month.

AI performance

How the AI performs

Every conversation is scored with the same rubric, every month.

Resolution rate

Judged resolved, fully or partially · last 3 months

▼3.7 ptsvs Aug

65.8%Sep 2026

Source: Consio Voice AI Index · Sep 2026

Avg. quality rating3.75/5▼0.10vs Aug

1★

2★

3★

4★

5★

Source: Consio Voice AI Index · Sep 2026

Escalation rate26.7%▲2.5 ptsvs Aug

Source: Consio Voice AI Index · Sep 2026

Avg. call duration3m 05s▼20 svs Aug

Source: Consio Voice AI Index · Sep 2026

How conversations end

% of all AI conversations

37.9%

26.7%

21.5%

13.9%

Resolved by AI37.9%

Escalated to a human26.7%

Unresolved21.5%

Callback scheduled13.9%

Source: Consio Voice AI Index · Sep 2026

How it works

How we generate the data

No surveys, no sampling. Every AI call goes through the same pipeline.

How it works

How we generate the data

No surveys, no sampling. Every AI call goes through the same pipeline.

Call

Every AI call, inbound & outbound

56,234 calls

Transcript

Speech-to-text, personal data removed

56,234 transcripts

LLM Judge

Gemini 2.5 Flash, same rubric on every call

38,336 calls judged

CategorizationIssue · goal

Quality1–5 score · summary

OutcomeResolved or escalated

Index

Aggregated across brands, monthly

index / month

Source: Consio Voice AI Index · Sep 2026

Methodology

Published every month. Figures are aggregated across all brands, so no single brand can be identified.

Period: Jul to Sep 2026, by calendar month. Month-over-month trends compare Aug and Sep 2026.
Sample: 56,234 AI calls from 185 Shopify brands in September 2026.
Coverage: 38,336 of these calls (68%) are scored by the judge.
Judge: Every scored call is evaluated by an LLM judge with the same rubric: summary, resolution, category, customer issue and customer goal.
Two intents: Customer issue is the problem behind the call. Customer goal is what the caller wants to get done.
Anonymization: Names, emails, addresses, order and payment numbers are removed. No brand is named without consent.
Next edition: Nov 1, 2026