Skip to content
Unit Economics

AI Cost Per Support Ticket 2026:
What Does One AI-Resolved Ticket Actually Cost?

The true cost of an AI-resolved support ticket — including LLM inference, vector DB retrieval, and infrastructure overhead. Per-ticket costs across 6 models, scaling tables from 1K to 1M tickets/month, and deflection ROI vs human agents. Last verified: 2026-04-01.

9 min read·Updated April 2026
AI Cost Per Ticket (5-turn conversation)
$0.00088
Flash-Lite (cheapest)
$0.00222
GPT-5.4 nano
$0.01025
Claude Haiku 4.5
$0.02825
Claude Sonnet 4.6

How We Calculate AI Cost Per Ticket

A standard customer support conversation involves multiple turns. The token math for a 5-turn conversation with a 500-token system prompt:

# 5-turn conversation token breakdown
system_prompt: 500 tokens (cached after turn 1)
per_turn_user: 150 tokens
per_turn_assistant: 200 tokens
# Input per turn (accumulates)
turn_1: 500 + 150 = 650 input, 200 output
turn_2: 500 + 150 + 350 = 1,000 input, 200 output
turn_3: 500 + 150 + 700 = 1,350 input, 200 output
turn_4: 1,700 input, 200 output
turn_5: 2,050 input, 200 output
total_input: ~6,750 tokens | total_output: ~1,000 tokens

Cost Per Ticket: All Major Models

ModelInput/1MOutput/1MCost (6,750 in + 1,000 out)Cost (cached input)
Gemini 2.5 Flash-Lite$0.10$0.40$0.001075$0.000880
GPT-5.4 nano$0.20$1.25$0.002600$0.002225
Gemini 2.5 Flash$0.30$2.50$0.004525$0.003525
Claude Haiku 4.5$1.00$5.00$0.011750$0.007250
GPT-5.4 mini$0.75$4.50$0.009563
Claude Sonnet 4.6$3.00$15.00$0.035250$0.021750

Cached input assumes 500-token system prompt cached at 90% off (Haiku cache read: $0.10/M). Caching the system prompt saves $0.45–4.50/1K tickets depending on model.

Monthly Cost at Scale

Monthly ticketsFlash-LiteGPT-5.4 nanoClaude Haiku 4.5Claude Sonnet 4.6Human agent cost
1,000 tickets$0.88$2.22$7.25$21.75$3,000–5,000
10,000 tickets$8.80$22.25$72.50$217.50$30,000–50,000
100,000 tickets$88$223$725$2,175$300K–500K
1,000,000 tickets$880$2,225$7,250$21,750$3M–5M

AI handles 40–70% of tickets as full deflections (no human needed). The cost comparison should be against the deflected ticket count, not total tickets.

All-In Cost Per Ticket (Including Infra)

ComponentCost per ticket (Haiku, 10K/mo)
LLM API (Haiku, cached system prompt)$0.00725
Embedding (query, ~50 tokens)$0.0000010
Vector DB query (Pinecone at $70/mo ÷ 10K tickets)$0.007000
App hosting + queue ($100/mo ÷ 10K tickets)$0.010000
Total all-in cost per ticket~$0.025

At 10K tickets/month, infra ($170/mo) adds ~$0.017/ticket on top of LLM cost. At 100K tickets/month, infra becomes a smaller fraction and per-ticket total approaches LLM cost alone.

Deflection ROI Analysis

ScenarioTickets/moDeflection rateAI cost (Haiku)Human cost savedROI
Conservative10,00040%$250/mo$16,000/mo64×
Moderate10,00060%$250/mo$24,000/mo96×
Optimistic10,00075%$250/mo$30,000/mo120×

Human agent cost assumed at $4/ticket (outsourced). All-in AI stack cost (LLM + infra) = $250/mo for 10K tickets (all tickets screened, 6K–7.5K fully deflected).

Model Selection for Support Tickets

  • Simple FAQ deflection (product info, status, returns): Flash-Lite or GPT-5.4 nano — structured retrieval + template response
  • Technical troubleshooting (multi-step, context-heavy): Claude Haiku 4.5 — better reasoning with retrieved docs; cache knowledge base context
  • Complex billing disputes, escalations: Claude Haiku 4.5 + human escalation flag — cost-effective with graceful handoff
  • Enterprise B2B support (custom deployments, multi-product): Claude Sonnet 4.6 — worth the premium when customer LTV is high and resolution quality matters

Calculate Your Support AI Cost

Enter your monthly ticket volume and model to get exact per-ticket and monthly AI costs.

AI API Cost Calculator