The 2026 customer support time benchmark
We surveyed 412 support teams. The median first-reply is still 6h14m — and the gap to top performers has never been wider.
Sho Ito
Head of Research · Apr 14, 2026 · 9 min read
Every spring we run the same survey: how long does it actually take to answer a customer? This year we got 412 responses from support leaders across e-commerce, SaaS, and DTC. The headline is that the median is barely moving — but the spread is exploding.
The numbers
- Median first-reply time: 6h 14m (vs. 6h 22m in 2025 — statistical noise).
- Top decile: 8m 31s. Up from 11m last year.
- Bottom decile: 38h 12m. Worse than 2025's 35h.
- Teams using AI-assisted drafting: 41%. Up from 19%.
The median didn't move because the middle of the market hasn't adopted AI drafting yet, and human bandwidth has a hard floor. What's changing is the tails. Top performers are pulling away because they've automated the first 80% of replies and use humans for the hard 20%. Bottom performers are getting worse because ticket volume keeps growing and headcount isn't keeping up.
What top teams do differently
- They draft, not auto-send. 96% of top-decile teams have a human reviewing every AI-generated reply before it ships. Auto-send is a trap — it saves seconds and costs goodwill.
- They route by sentiment, not channel. Negative-sentiment tickets get pulled to a senior agent regardless of whether they came from Gmail, Zendesk, or Intercom.
- They measure draft acceptance rate. The metric isn't "replies sent." It's "drafts accepted without major edits." A 70% acceptance rate is the threshold where AI drafting starts paying for itself.
Where the data surprised us
We expected SaaS to lead e-commerce on response time. It didn't. SaaS support is dragged down by long technical investigations — debugging beats batching. E-commerce teams batch ruthlessly, especially on common questions (shipping, returns), and AI drafting pays off there immediately.
"The teams winning right now treat support like a product surface, not a cost center. Every minute saved on the easy stuff gets reinvested into the hard stuff."
Methodology
Self-reported via a 14-question survey, distributed through three support-leader Slack communities and our customer list. Responses verified by domain (e.g., a freelancer claiming to run support for Acme Co. couldn't submit if Acme's actual head of support already had). We weighted by team size to avoid one-person teams dominating the median.
Full anonymized dataset is available on request — email research@quickply.com.