Sierra

Bret Taylor's outcome-priced AI customer-service agent platform — $15.8B valuation, $150M+ ARR, 40% of the Fortune 50 as customers, and the canonical case for "pay per resolution" software economics.

1. Core Product / Service

Sierra sells branded AI customer-service agents to enterprises. The product is positioned as a "branded agent persona" trained on the company's knowledge base, integrated with backend systems (refunds, claims, account changes), and configured to handle billions of customer interactions [1][2][7].

Key architectural elements:

• Branded agent identity — the agent represents the buyer's brand, not Sierra's.
• Action-capable agent — beyond Q&A, the agent executes transactions (mortgage refinancing, insurance claims, refunds, plan changes) through deep integration with each enterprise's systems.
• Multimodal channels — chat, voice, email; voice in particular is a 2025-26 expansion vector.
• Outcome-priced contracts — pre-negotiated per-resolution rate; if the case escalates to a human, no charge to the customer [3][6].

Product surface in 2026 has expanded "beyond customer support" toward what CMSWire and Sequoia podcasts describe as a "platform for AI agents that complete jobs" — sales, ops, and beyond [2][7].

2. Target Users & Pain Points

• Buyer: Chief Customer Officers / VP Support at F50/F100. Sierra cites >40% of the Fortune 50 as customers [1].
• Pain points addressed:
  • Legacy BPO and chatbots fail on complex, transactional cases (mortgage refi, insurance claims) that require backend system access.
  • Per-seat or per-hour pricing (BPO) is misaligned with AI's marginal-cost-near-zero economics — Sierra's outcome pricing aligns vendor and buyer incentives.
• Wedge: Bret Taylor brand + outcome pricing + frontier-model quality + deep enterprise integration capacity.

3. Competitive Landscape

Sierra's structural moat is brand + integration depth: Bret Taylor / Clay Bavor's pedigree + the willingness of F50 procurement teams to give an LLM agent deep system access (mortgages, insurance claims) is the high-bar earned trust other vendors lack.

4. Unique Observations

• Pricing model: Pure outcome-based per-resolution. The contract structure: pre-negotiated rate per autonomous AI resolution; if the conversation escalates to a human, no charge. Some accounts also use volume-priced rates for routine interactions, blended with outcome rates for complex ones [6][7]. Bret Taylor publicly evangelizes "pay for a job well done" as the secular software business model for AI.
• Implied $/1M tokens consumed: Sierra resolutions can be very long (multi-turn voice + tool calls + system integrations); estimate 20-100K tokens per resolved case. Per-resolution rates aren't public but FourWeekMBA and pricing-research blogs imply rates in the $1-$5 range per resolution at enterprise scale, with very high-value resolutions (mortgage, insurance claims) priced higher. At $1-5/resolution × 50K tokens ≈ retail rate of $20-$100 per 1M tokens at the model layer. Frontier API cost (2026): $3-$10/1M (output-heavy w/ voice). Markup: ~5-30× over raw API. Materially lower per-token markup than harvey or crescendo — but Sierra captures $/resolution, not $/seat or $/conversation, so the vendor economics work even at thinner per-token markups because each resolution is priced against the cost of a human handling that case ($5-$30 BPO labor).
• Moat type: Brand (Taylor / Bavor founder pedigree, OpenAI board) + distribution (F50/F100 procurement access) + workflow / integration (deep system access to mortgages, claims, refunds). Not a model moat. The brand moat is the most defensible: Sierra got into F50 procurement processes that newer entrants can't fast-follow.
• Customer profile: Blue-chip enterprise. >40% of F50; airlines, banks, telcos, retail, mortgage. Explicitly not SMB or growth-stage tech (Decagon's territory).
• Markup multiple over raw API: ~5-30× per token, but $/resolution vs. $/human-equivalent-labor — the more honest framing — is closer to 0.05-0.2× of human labor cost per case. Sierra captures ~10-30% of the gap between raw API cost and human-labor cost; the buyer keeps 70-90% in savings.
• Industry impact of pricing model: HubSpot publicly switched to per-resolution AI pricing (2025) explicitly modeled on Sierra; the pricing model is now becoming the L4 CX category default [8]. This is rare — the highest-priced upstart resetting category pricing norms within 24 months of launch.
• Strategic context: Bret Taylor is also the chair of OpenAI's board. The OpenAI tie + the "the atomic unit of AI productivity is a process, not a person" thesis converge: Sierra is the operational testbed for OpenAI's enterprise-agent ambitions.

5. Financials / Funding

• Total raised: ~$1.4B cumulative; latest $950M at $15.8B valuation (May 2026), led by Tiger Global and GV [1][2][3].
• Prior round: $175M at $4.5B (Oct 2024).
• ARR: $150M+ (Feb 2026); $100M (Nov 2025); $50M+ first quarter posted = first $50M Q.
• Customer base: >40% of Fortune 50; agents handling billions of interactions across mortgage refi, insurance claims, support.
• Founded: 2023.
• Headcount: not publicly disclosed; ARR/employee ratios suggest lean but growing fast.

6. People & Relationships

• Co-founders:
  • Bret Taylor (CEO) — chair of OpenAI's board; ex-co-CEO of Salesforce; ex-CTO of Facebook; ex-CEO of Quip; co-creator of Google Maps. Generational technology executive.
  • Clay Bavor — ex-Google VP (Labs / VR/AR); the engineering anchor.
• Investors: Tiger Global (lead, May 2026), GV (Google Ventures, co-lead), Sequoia, Benchmark, Thrive Capital.
• OpenAI relationship: Bret Taylor as OpenAI board chair creates a tight feedback loop on enterprise agent capabilities; Sierra is widely read as a "GPT-Enterprise reference customer."
• Reference customers: F50 banks, airlines, telcos, retailers (publicly cited but specific names limited).
• Related wiki: crescendo (HITL competitor at $500M val), decagon ($4.5B val competitor), magic (HITL pattern in different vertical).