Alisa Kargin, product manager / Cases

Should support be human again?

Companies are moving support to AI because a human contact costs about seven times more. Customers are pushing back, and a lot of them say they would pay to get a person. So I modeled the question a product team actually has to answer: is a paid "human tier" a business, or a nice idea?

43%

of consumers would pay more for a product that guarantees access to human support.

Alchemer, 2026 AI & Customer Expectations Report, 2,009 consumers

What the evidence says

Six numbers frame the decision. Each links to its source at the bottom of the page.

44%

say AI has made it harder to reach a person when they need one.Alchemer, September 2026

64%

would prefer companies did not use AI for customer service. 53% would consider switching.Gartner survey of 5,728 customers, 2024

14%

of customer service issues are fully resolved in self-service. Even "very simple" ones: 36%.Gartner, 2024

7x

cost gap: $1.84 per self-service contact against $13.50 per assisted contact.Gartner customer service cost benchmark, median

700

support roles Klarna replaced with AI, before it started hiring humans back in 2025 as satisfaction fell.Forbes, May 2025

2.5x

Daily AI users are the most likely to say they would definitely pay for a human: 27% against 10% of infrequent users.Alchemer, September 2026

The model

A subscription company with one million customers. Three ways to run support, compared over a year against all-human support. Move any lever. Every default is an assumption, labeled.

The company

AI-first support

Gartner measured 14% for self-service in 2024. Newer AI agents claim far more; 50% is a middle guess.
Percentage points per year. The least known number here, and the one that decides AI-first.

Paid human tier

43% say they would pay. Saying is not buying, so the default is far lower.
Times as often as an average customer. People who need help most are the first to buy it.
Percentage points per year.
Costs per contact: $1.84 for AI, $13.50 for a human (Gartner medians). AI that fails hands the customer to a human, so the company pays both.

All human

a year, against all human
Support cost
Extra churn
Tier revenue
-

AI first

a year, against all human
Support cost
Extra churn
Tier revenue
-

AI first + paid human tier

a year, against all human
Support cost
Extra churn
Tier revenue
extra support cost per member per year, against AI first
what a member pays per year
break-even price per month, before churn effects

Churn is valued at one year of lost revenue per extra customer lost. Baseline churn is the same for every option, so only the difference counts.

What I would test first

The model is most sensitive to two numbers nobody has measured: how many customers really buy the human tier, and how much more those members use it. Both are cheap to learn.

  1. Offer the tier to a small slice of customers

    5% of new and renewing customers see a "guaranteed human" option at checkout, split across two prices.

  2. Measure buying, not liking

    Uptake at each price, then contacts per member and renewal against a matched group without the offer.

  3. Read the AI-first churn at the same time

    Track cancellations that follow a failed AI conversation. That number decides whether AI-first is saving money or quietly losing it.

The rule, agreed before launch: if fewer than 3% buy at the lower price, or members cost more to serve than they pay, we stop and keep AI-first with an easy path to a person.

What the model leaves out

Sources

  1. Alchemer, 2026 AI & Customer Expectations Report (press release, September 29, 2026)
  2. Gartner: 64% of customers would prefer companies didn't use AI for customer service (July 2024)
  3. Gartner: only 14% of customer service issues are fully resolved in self-service (August 2024)
  4. Gartner: Benchmarks to Assess Your Customer Service Costs ($1.84 self-service, $13.50 assisted, medians)
  5. Forbes: Klarna reverses on AI, says customers like talking to people (May 2025)