# Calculation notes — State of AI in Customer Support, mid-2026 field report

All Cadenflow calculations in the published report, with formulas and inputs. Verified July 31, 2026.

## 1. Inference-cost scenarios (per AI-handled text resolution)

Model list prices as displayed July 31, 2026:
- OpenAI GPT-5.4-mini: $0.375 / 1M input tokens, $2.25 / 1M output tokens (developers.openai.com/api/docs/pricing)
- Anthropic Claude Haiku 4.5: $1.00 / 1M input, $5.00 / 1M output (platform.claude.com/docs/en/about-claude/pricing)

Scenario A — low-complexity text resolution (2 turns, no tools):
~2,500 input tokens (system prompt + KB context + message), ~500 output.
- GPT-5.4-mini: 2500×0.375/1M + 500×2.25/1M = $0.0009 + $0.0011 ≈ **$0.002**
- Haiku 4.5: 2500×1/1M + 500×5/1M = $0.0025 + $0.0025 ≈ **$0.005**

Scenario B — multi-turn tool-using resolution (4 turns, 2 tool calls, context re-sent):
~6,000 input, ~1,200 output.
- GPT-5.4-mini: $0.00225 + $0.0027 ≈ **$0.005**
- Haiku 4.5: $0.006 + $0.006 ≈ **$0.012**

Scenario C — heavy/voice: several times Scenario B once ASR/TTS or frontier-model routing is added; voice infrastructure commonly cited at ~$0.05–0.10/minute (third-party, directional).

Published claim: "on the order of half a cent to a few cents per text resolution at mid-2026 list prices, before retries, evaluation overhead, and orchestration." Excludes: retries, eval/judge calls, embeddings, logging infra — real deployments run higher.

## 2. Freshworks revenue multiple (Cadenflow calculation)

Market cap $3.21B (close Jul 29, 2026) ÷ TTM revenue $871.17M ≈ **3.7x price/sales** (stockanalysis.com/stocks/frsh/, fetched Jul 31, 2026). EV/revenue would be lower (Freshworks holds ~$1B cash). Used as: "public helpdesk comparable trades near ~4x sales."

## 3. Funding aggregates (from funding-tracker.csv, frozen Jul 31, 2026)

Core set = disclosed equity rounds, AI customer-support/service vendors (incl. voice; incl. Redo as an adjacent suite selling AI support resolutions; excl. debt, excl. AI-SDR, excl. eval tooling — exclusions listed in tracker).

- 2026 (Jan–Jul): 6 rounds, $1,786M total. Top 3 (Sierra $950M, Parloa $350M, Decagon $250M) = $1,550M = **86.8% ≈ ~87%** of tracked 2026 dollars. All three are Series D/E at $3B–$15.8B valuations. Largest first-institutional round in the tracked 2026 set: **$5M** (Cue seed).
- 2025: 7 rounds, $856M. 2024: 10 rounds, $586M.
- LIMITATION (published): the tracker is curated around profiled companies and press-covered rounds; small seed rounds are under-captured. Dollar-concentration among top rounds is robust to that bias (adding $200M of missed small rounds would still leave top-3 ≈ 78%); year-over-year deal-count or stage-share trends are NOT inferable from this set and are not published.

## 4. Human-cost baseline (labels used in report)

- "Roughly $5–12 per human-handled ticket" = third-party industry estimates (DigitalGenius $4–12/interaction; channel-dependent to $5–25) — labeled third-party estimate, not Cadenflow measurement.
- Offshore support VA: $900–1,300/mo all-in (salary + 13th month + tools + processing), before owner management time — from agency/platform rate cards (OnlineJobs.ph, Hurupay 2026, agency guides), labeled as such. No per-ticket derivation is published (utilization/tickets-per-hour unknown).

## 5. Multiples cited (Section 1) — basis labels

- Sierra ~79x = $15.8B reported post-money ÷ ~$200M forward-revenue estimate (Getlatka estimate; ARR $150M company-reported Feb 2026). Label: press/estimate-based, forward.
- Decagon ~130x = $4.5B ÷ $30–35M company-reported annualized (Jan 2026); ~375x on Forbes's independent $12M estimate. Label: company-reported vs journalist estimate.
- Wonderful ~110x = $2B ÷ ~$18M company-reported (Q1 2026).
- Freshworks ~3.7x P/S = calculation above.
- Influx (services comparable): Dealroom enterprise-value estimate $12–18M on estimated $15–40M revenue ⇒ <1x. Label: third-party estimates.