- Sierra makes AI agents that talk to a company's customers. The agents work by chat, phone, SMS, email and in ChatGPT. They also do tasks, for example a refund or a plan change.3, 12
- Bret Taylor and Clay Bavor started Sierra in 2023. Sierra launched in February 2024. Taylor was co-CEO of Salesforce. He is also chair of the OpenAI board.16, 12
- Sierra grows very fast. It reached about $200M ARR in May 2026, nine quarters after launch. Investors value it at $15.8B.11, 13
- Sierra sells to large enterprises only. Each contract is custom. Sierra shows no public price list.17, 45
- The main price metric is the outcome. The customer pays when the agent completes the job. When the agent sends the case to a human, the customer usually pays nothing.1, 9
- The next pricing problems are hard. Sierra must define outcomes clearly, price long revenue workflows (the Horizon product), protect margin, and sell to more companies without a price list.
What kind of company is Sierra, and how did it grow?
What does it do? How did it start? How did it expand from product A to B to C, and from one domain to the next?
- Sierra is an enterprise software company. It calls itself "the leading platform for customer-facing AI agents." Sierra21
- A Sierra agent answers customers and does work in the company's systems. It can process a return, change a subscription, or take a payment.1, 8
- Sierra started with chat support for consumer brands. Then it added voice, regulated industries, other countries, and revenue work such as sales and loans.2, 5
How it started
Bret Taylor built Google Maps, was CTO of Facebook, founded Quip, and was co-CEO of Salesforce. He announced his exit from Salesforce in November 2022, the same day ChatGPT launched. Reported16
Clay Bavor worked at Google for 18 years. He led product and design work in Workspace and led Google's AR/VR and Project Starline work. Reported16
The two founders agreed to start a company in early 2023 without a fixed product. They chose customer service because it is a large, measurable job that language models can do. Sierra came out of stealth on Feb 13, 2024 with $110M from Sequoia and Benchmark. The first public customers were WeightWatchers, SiriusXM, Sonos and OluKai. Reported15, 16
Each expansion step also changes the pricing unit. A chat answer is cheap and short. A voice call costs more to serve. A Horizon goal, for example a closed loan, can take weeks and involve many people. The pricing team must decide what "one outcome" means at each step.
What are the key numbers?
Revenue, valuation, employee count, TAM, and more.
TAM: two budgets, very different sizes
There are two ways to size Sierra's market. The difference is the main reason for its pricing model.
Labor lens
About 17 million contact center agents work worldwide (Gartner). The US has about 2.7M customer service reps. At about $42K salary each, that is about $120B of wages per year. Labor is up to 95% of contact center cost.16, 37
AI impact lens
Gartner predicted that conversational AI will cut contact center labor cost by $80B in 2026. The global contact center market may grow to about $500B by 2030.36, 37
Revenue lens (new)
Horizon moves Sierra into sales, retention and loan work. This budget is not in the numbers above. No public source sizes it yet. Analysis5
Seat pricing sells into the software budget. Outcome pricing sells into the labor budget, which is about 20 times larger. This is the strategic logic for the outcome model. It also changes the buyer. The buyer is often the COO or CFO, not only the IT team.
What is the key product, and how is it delivered?
Is it a standard SaaS product with AI? Or does Sierra customize everything, like Palantir with forward deployed engineers (FDEs)? How does it work?
- It is a hybrid. All customers run on one shared platform, Agent OS. Sierra does not write a new code base for each customer.3, 12
- But Sierra sells "agents as a service." A Sierra team helps design, build and improve each customer's agent. This part is similar to the Palantir FDE model.4, 41
- Sierra uses tools to make this service work cheaper and faster. The main tool is Ghostwriter, an AI that builds agents.4
- DiscoverCollect SOPs, call transcripts and KPIs. Agree the outcome definitions. These definitions become the billing terms.
- BuildBuild the agent with Ghostwriter, Studio or SDK. Connect it to the customer's APIs.
- TestRun simulated conversations. Check guardrails, policy and tone.
- LaunchGo live on the agreed channels. Set the rules for handoff to humans.
- ImproveInsights and the agent team find failures. They fix them. Resolution rate goes up.
Who does the work: the "Agent Development" team
Sierra job posts show the service model. Agent Strategists own the customer's outcomes, from pilot to production. Agent Engineers build the agent logic and integrations. Agent Product Managers plan the roadmap for each agent. Forward Deployed Infrastructure Engineers work close to the customer and keep agents stable. Reported42, 41
One report found about 20 open "Agent Engineer" roles in ten cities. This is a people-heavy model. It costs money. It is also a strong reason why customers get results. Reported41
The service team is a cost that the outcome price must cover. Ghostwriter can lower that cost. A pricing question follows: if AI builds the agent, should implementation stay bundled, become a separate fee, or go away for smaller customers? Also note that partner firms (system integrators) may take more of the build work after the Oct 2026 partner program.6
Is Sierra's delivery model like Palantir's?
How does Sierra's forward-deployed model compare with Palantir? What does the org chart probably look like? How is it different from a typical SaaS company?
- Yes, in structure. Sierra uses the same pair of field roles as Palantir: a strategist who owns the customer problem, and an engineer who builds the solution at the customer.47, 42
- Sierra also uses Palantir's feedback loop. Field teams build for one customer. The platform team turns repeated work into product.47, 4
- There are three big differences. Sierra does one job (customer conversations). It goes live in weeks. It charges per outcome, not with a large license.1, 13, 50
| Job to do | Palantir | Sierra | Typical SaaS |
|---|---|---|---|
| Own the customer problem, stakeholders and adoption | Deployment Strategist ("Echo") | Strategist, Agent Development | Customer success manager (CSM) |
| Build the solution inside the customer's systems | Forward Deployed Software Engineer ("Delta") | Software Engineer, Agent ("Agent Engineer") | Customer's own admin, or a partner firm |
| Plan the roadmap for one customer's solution | Shared by the Echo and Delta team | Product Manager, Agent Development | No equivalent role |
| Keep the deployment stable | Part of the Delta team's work | Deployed Infrastructure Engineer | Central support and SRE |
| Build the shared product | Product engineers ("Dev") | Platform Engineering | Product engineering |
| Prove value before the sale | Bootcamp: a working pilot in days | Sales Engineering, Business Value team, pilot agent | Demo and free trial |
- Field buildAn agent pod builds a custom flow for one customer, for example a check of mortgage documents.
- PatternThe same need appears at other customers in the same industry.
- ProductizePlatform Engineering turns it into a standard feature: a skill, a connector, a template, or a Ghostwriter capability.
- Faster next dealThe next deployment needs less custom work. Build cost goes down. Go-live gets faster.
Engineering
- Platform EngineeringAgent Runtime, Agent Builder, Context Engine, Horizon, Inference, Insights, Channels, Payments, Security, SRE
- Applied Researchmodels and benchmarks
- Agent Engineeringpods by industry: financial services, healthcare, insurance, retail, telecom and media, travel, public sector; plus language pods
- Deployed Infrastructurekeeps customer agents stable
Product
- Agent Strategistsby region and industry; own customer outcomes
- Agent Product Managersby industry and language
- Product Design
Sales and Partnerships
- Enterprise Salesarea and regional VPs, sales directors, enterprise account executives
- Sales Engineering
- Partnershipsecosystem, system integrators, public sector, developer relations
Operations
- GTM OperationsAgent Development ops, Business Value, Pricing, order management, sales compensation, GTM engineering
- Business Operations
- People and Recruiting
Other functions
- Marketing and Comms
- Legalcommercial and product counsel
- Security and GRC
- IT
| Dimension | Typical SaaS | Palantir | Sierra |
|---|---|---|---|
| What the customer buys | Software access, by seat or tier | Platform (Foundry, AIP) plus embedded engineers | Platform (Agent OS) plus an agent team: "agents as a service"4 |
| Scope of work | One product's feature set | Broad data and operations problems: defense, health, industry | One job: customer conversations, now also revenue workflows (Horizon)5 |
| Who builds the solution | The customer | Palantir field engineers | Sierra agent pods, more and more with Ghostwriter4 |
| Time to first value | Hours to weeks | Days for a bootcamp pilot; months for large programs50 | About 5–7 weeks to production2, 13 |
| Pricing | Public price list. Seats or tiers. | Large custom contracts. Not public. $11.2B total remaining deal value at end of 2025.50 | Custom. Outcome-based. Not public.1, 17 |
| Scale | Thousands of customers or more | 954 customers; $4.475B revenue in 202549 | "Hundreds" of customers; ~$200M ARR2, 13 |
| Gross margin | About 80% in classic SaaS43 | About 82% in 202550 | Not public. Services and inference push it down.16, 43 |
| What field work produces | Services are a cost, often at near-zero margin | Field work becomes product over time47, 48 | Field work raises the resolution rate, which raises outcome revenue Analysis |
| Main way to scale | Marketing and self-serve | Bootcamps and a stronger product (AIP)50 | Ghostwriter, partners, marketplaces4, 6, 7 |
What this means for pricing
The pod cost sets a floor
A pod of three or more specialists per large customer is expensive. The platform fee and the minimum contract (about $150K per year, estimated) must cover it.18
The pod is a revenue engine, not only a cost
In seat pricing, services do not grow revenue. In outcome pricing, each point of resolution-rate gain creates more billable outcomes. A useful metric: outcome revenue gained per pod hour. Analysis
Avoid the consulting trap
If pods build one-off work that never becomes product, margins stay low. Then pricing must carry more service fees, and deals get harder to scale.48
A short way to say it: Sierra uses Palantir's field model, but it connects the field team to an outcome meter. At Palantir, field work grows the license over years. At Sierra, field work raises the resolution rate, so revenue can go up within the same contract. The pricing team's job is to make sure the pod cost, the platform fee and the outcome rate fit together.
What is Sierra's pricing strategy today?
Which metric? Seats? Usage or tokens? Tiers? Different by product? Base price plus paid add-ons? Public price? Commitment discounts?
- Sierra has no public price, no self-serve plan, and no published tiers. sierra.ai/pricing does not exist.17
- The main metric is the outcome: a resolved conversation, a saved cancellation, a sale or upsell.1
- Low-value traffic, for example routing, has a per-conversation fee. Sierra calls this a "blended" approach.1
- Reported contracts add an annual platform fee and implementation services. Each deal is negotiated.45, 18
| Question | Answer | Source quality |
|---|---|---|
| Pay per seat? | No. Sierra says seat pricing makes vendors want more humans at work, which conflicts with automation.1 | Sierra |
| Pay per token? | No. Horizon launch post: "you don't pay for tokens, you pay for business outcomes delivered. We bear the burden of managing token spend."5 | Sierra |
| Main metric | Per outcome. Outcome types include resolved conversations, e-commerce purchases, saved memberships, upsells and cross-sells. Sierra and the customer agree on clear criteria for each outcome before launch.1 | Sierra |
| If the agent fails? | Escalation to a human: "in most cases, there's no charge." Bret Taylor: "If we do have to escalate to a person, that's free."1, 9 | Sierra |
| Secondary metric | Per conversation for routing or "greeter" traffic, where an outcome is hard to measure.1 | Sierra |
| Platform fee | Contracts reportedly have three parts: annual platform fee, per-outcome usage, and professional services.45 | Reported |
| Price level | About $1.50 per resolution. Minimum about $150K per year. Year one about $200K–$350K+ with setup.18, 30 | Estimate |
| Tiers? | None published. No free tier.17 | Sierra |
| Different by product? | Yes, in practice. The outcome definition changes by use case. Horizon bills multi-week goals such as a mortgage, a claim or a renewal. Voice reportedly has separate economics. Live Assist helps human agents, so it has no autonomous outcome. Its metric is not public.5, 17 | Reported Analysis |
| Base + paid add-ons? | Not published. The platform fee probably covers access. Outcome fees scale with value. | Analysis |
| Commitment discounts? | Not published. Annual commitments are required. Volume rate steps, prepaid outcome volume and multi-year discounts are normal in enterprise deals, but no source confirms Sierra's terms. New in Oct 2026: customers can buy through Google Cloud Marketplace and use their existing Google Cloud commitments.17, 7 | Sierra Analysis |
The job post for this team says the role "leads pricing strategy for major enterprise opportunities," builds "ROI analyses," and turns "new product capabilities into scalable commercial offerings."21 So expect three kinds of work: deal pricing (deal desk), value cases (ROI), and packaging for new products such as Horizon and voice.
What is outcome-based pricing, and what does it mean for Sierra?
What is it in the market? Does Sierra use it? What did Sierra signal? Pros and cons, considerations and implications.
- Outcome-based pricing: the customer pays only when the software delivers a specific, agreed result.
- Yes, Sierra uses it. It is Sierra's core model since launch, and Sierra talks about it in public often.1, 9, 10
- Many competitors copied it in 2025–2026. The model alone is no longer a big difference. The difference now is the outcome rate, the outcome quality, and the definition of an outcome.23, 24, 26
| Model | The customer pays for | Who carries the risk | Budget predictability | Example |
|---|---|---|---|---|
| Seat | Each human user | Customer | High | Microsoft D365 Contact Center, $110 per user per month34 |
| Usage | Input: tokens, minutes, actions, conversations | Customer | Medium | Google CES per request or per second; Agentforce Flex Credits, $0.10 per action33, 27 |
| Outcome | Output: a result | Vendor | Low to medium | Fin $0.99, Zendesk $1.50–$2.00, Agentforce $2 per resolution; Sierra25, 26, 23, 1 |
| Hybrid | Platform fee + outcomes or usage | Shared | Medium to high | Most enterprise AI deals, including Sierra's reported contracts45 |
What Sierra has said in public
let's pay for a job well done.He also said:
Changing your business model is really hardfor incumbents.10
reducing your token utilization for the same outcomes is your problem, not your customer's.He says outcome pricing is "pretty different than usage-based."9
We bear the burden of managing token spend.5
Pros for Sierra
- Aligned incentives. Sierra earns more only when the agent works better. This makes the sale easier.
- Larger budget. Sierra competes with labor and outsourcing (BPO) cost, not with software cost.
- Simple ROI story. "You pay a part of the cost that you save."
- Hard for incumbents to copy fully. Seat-based vendors lose seat revenue when AI replaces humans.1, 10
- Built-in expansion. More volume, higher resolution rate, and new outcome types all grow revenue without a new contract.
- Efficiency gains stay with Sierra. When model cost falls, Sierra keeps the margin. The customer still pays per outcome.9
Cons and risks for Sierra
- Sierra pays for failures. Inference and phone costs occur on every conversation, but only resolved ones earn money.16
- Revenue is less predictable. It moves with customer volume and seasons. Forecasts and revenue recognition become harder.22
- Definition disputes. "Was this case really resolved?" Each dispute costs time and trust.
- Long sales cycles. Each deal needs custom outcome terms and a custom value case.
- Price anchor can fall. The price is set against human cost. After human teams shrink, buyers compare with the $0.99 market rate instead.
- Copied by everyone. Salesforce, Zendesk and Fin now sell per resolution with public prices.23, 26, 25
Deep dive: the nine issues a pricing team must manage
Outcome definition and verification
Fin counts an "assumed resolution" when the customer leaves without more questions. Critics say this bills for unsolved cases. Zendesk now bills only "verified" resolutions, which a second AI model checks (May 2026). Fin reverses a charge if the customer comes back.25, 26
Write one standard definition per outcome type. Add a reopen window, a CSAT floor, and an audit log that the customer can see.
Attribution for revenue outcomes
A Horizon loan can take three weeks. Emails, calls and a human loan officer all help. Who gets the credit? Without a clear rule, every large outcome can cause a dispute.5
Use attribution windows, split-credit rules, and holdout (control) groups that measure the increase the agent causes.
Cost to serve and gross margin
AI software margins are often 50–60%, not the 80% of classic SaaS. Voice adds telephony and speech costs. If the resolution rate is low, cost can be more than revenue.43, 16
Use smaller or in-house models where possible (Sierra plans custom models to cut inference cost12). Add a per-conversation floor for voice. Track margin per outcome type.
Predictability for both sides
The customer's CFO wants a fixed budget. Sierra's CFO wants a forecast. Under ASC 606, variable fees are "variable consideration," which makes revenue recognition harder. Some customers also move from outcome pricing back to a fixed contract.22
Annual commitment plus prepaid outcome bundles. A floor and a cap (a "collar"). True-up at the end of the term.
Price anchor decay
Year one: "a human contact costs $8; we charge $1.50." Year three: the human team is smaller and the $8 number is gone. The buyer now compares with Fin at $0.99.24
Re-anchor on total cost per contact, resolution quality, and revenue gained. Lock multi-year rates with planned step-downs.
Mix shift: easy cases first
Agents automate easy intents first, for example "Where is my order?" Later intents are harder and cost more to serve. One flat rate under-prices hard tasks and over-prices easy ones.
Use an outcome rate card by value class: information, transaction, retention, revenue.
Incentive gaming
A customer can send more cases to humans to avoid fees. A vendor can close conversations too early. Both damage trust.
Joint governance with shared dashboards, quality scores in the billing rule, and regular business reviews.
AI agents on the other side
Personal AI agents now call businesses. Sierra built fleming-1 to detect them and launched the Personal Agent Protocol.8, 40 An agent-to-agent conversation is short and cheap. Is it the same outcome at the same price?
Define a separate price for agent callers, or a per-transaction fee for agent commerce, for example with Stripe payments.
Competitors copy the model
Outcome pricing became standard in 2026. Salesforce moved to $2 per resolution in July 2026. Zendesk and Fin also bill per resolution. The model itself no longer wins deals.23, 26, 24
Compete on outcome quality and scope: higher resolution rate, harder tasks, revenue outcomes.
Example inputs only. These are not Sierra numbers. Move the sliders to see how value is split between Sierra and the customer.
| Example vendor | Price per resolution | Resolution rate | AI cost per contact | Total cost per contact, with humans at $8 |
|---|---|---|---|---|
| Vendor A (higher price) | $1.50 | 70% | $1.05 | $1.05 + 30% × $8 = $3.45 |
| Vendor B (lower price) | $0.99 | 50% | $0.50 | $0.50 + 50% × $8 = $4.50 |
A strong answer shows both sides. Outcome pricing is a sales and positioning advantage. It is also an operating discipline. It needs clear definitions, good measurement, and margin control. Also show that you know the market moved: in 2026, "outcome-based" is the norm, so the new question is which outcome, at what price, verified how.
What do competitors offer, and how do they price?
How do they differ from Sierra? What can Sierra learn, or watch?
| Company | Type | Pricing metric and public price | Scale | Difference from Sierra |
|---|---|---|---|---|
| Sierra | AI-native, enterprise | Per outcome + per conversation for routing + platform fee. No public price. ~$1.50 per resolution Estimate18, 1 | ~$200M ARR; $15.8B13 | — |
| Decagon | AI-native, enterprise | Customer chooses per conversation or per resolution, plus yearly platform contract. ~$0.99 per conversation Estimate. Median contract ~$433K per year (Vendr).30 | $4.5B (Jan 2026); ~$35M ARR in Nov 2025 Estimate28, 29 | Gives the buyer a choice of metric. Customers edit their own agents with "Agent Operating Procedures." Earlier, more tech customers.30, 29 |
| Fin (was Intercom) | Help desk + AI agent | $0.99 per outcome, public. Self-serve. 50-outcome monthly minimum on other help desks. Seats extra on Intercom.25, 24 | ~$400M ARR, ~30K customers. Salesforce will buy it for ~$3.6B (signed Jun 15, 2026).24 | The market's reference price. Sells to SMB and mid-market. Wider outcome definition.25 |
| Salesforce Agentforce | CRM incumbent | Many models: $2 per conversation (2024) → Flex Credits $0.10 per action → per user from $125 → $2 per resolution for Help Agent (GA Jul 17, 2026).27, 23 | Very large base; adds Fin | Owns CRM data and routing. Bundles AI with existing contracts. Changed its pricing model many times.27 |
| Zendesk | Help desk incumbent | $1.50 per verified resolution (committed) or $2.00 (pay-as-you-go). Seats still required.26 | Large SMB and mid-market base | "Verified resolution" checked by a second AI model (May 2026).26 |
| Parloa | AI-native, voice-first | Per minute (voice) and per interaction (chat); some outcome deals. Entry ~$300K+ per year Estimate31 | $3B (Jan 2026)46 | Voice focus. Usage metric. Strong in Europe. |
| PolyAI | AI-native, voice-first | Platform fee ~$150K+ per year plus per-minute usage Estimate32 | — | Voice and multilingual focus. Usage metric. |
| Ada | AI-native, chat-first | No public rate. | — | Simpler chat flows and no-code tools.16 |
| Google CES | Hyperscaler | Public usage price: $0.007–$0.012 per chat request; $0.001–$0.002 per second of voice.33 | — | Cheap building blocks. The buyer builds the agent. |
| Microsoft D365 Contact Center | Hyperscaler / CRM | Per seat: $110 per user per month.34 | — | Seat model. Bundled with Microsoft agreements. |
| OpenAI Frontier | Model lab platform | Enterprise platform to build and run agents (Feb 2026). Sierra is a "Frontier Partner."35 | — | Partner today, possible competitor tomorrow. Bret Taylor chairs the OpenAI board.12 |
What Sierra can learn or watch
1 · Fin sets the reference price
Every buyer now knows "$0.99 per resolution." Sierra must explain its price in total cost per contact and in value per outcome, not only in rate per resolution.24
2 · Zendesk sells trust in the definition
"Verified resolution" is a trust feature that also works as a pricing feature. Sierra could publish its own outcome standard and audit method.26
3 · Salesforce shows the cost of changing models
Agentforce changed its metric about four times in two years. Buyers got confused. Sierra's steady message is an asset. Protect it.27, 23
4 · Fin shows the risk of a wide definition
Fin bills "assumed" resolutions: the customer leaves without more questions. Critics say this bills cases that are not solved. A narrow, clear definition protects renewals.25
5 · Decagon gives the buyer a choice
A choice between per conversation and per resolution can make procurement easier. Sierra could offer a fixed-price option for buyers who need budget certainty.30
What is good and bad about Sierra's current pricing?
| Part of the strategy | What works | What does not work well |
|---|---|---|
| Outcome as main metric | Clear value story. Aligns with the buyer. Strong brand position since 2024. | Sierra carries the cost of failures. Variable revenue. Competitors now offer the same model. |
| No public price | Sierra can price each deal to its value. Protects against direct comparison with $0.99 offers. | Long sales cycles. Hard to reach mid-market buyers. Analysts publish guesses that Sierra cannot control.18, 30 |
| Custom contract for every deal | Fits complex, regulated buyers (banks, health, insurance). Supports a high average contract value. | Hard to scale. Prices can become inconsistent across customers. Needs a strong deal desk and approval rules. |
| Blended model (outcome + conversation + platform fee) | The platform fee gives a revenue floor. Per-conversation fees cover traffic with no clear outcome. | More complex bills. The customer may ask why it pays three ways. |
| "Agents as a service" with a people-heavy team | High resolution rates. Fast go-live. Customer trust. | Service cost lowers gross margin. Ghostwriter and partners must reduce this cost. |
| Horizon "pay for business outcomes" | Moves Sierra into revenue budgets. Very high value per outcome. | Attribution is hard. Long cycles delay billing. Disputes about who caused the result. |
Overall strengths
- Pricing and product tell the same story: "pay for a job well done."
- The model targets the labor budget, which is much larger than the software budget.
- Revenue grows when the product gets better. This drives strong net revenue retention. Analysis
- Investors pay about 79× ARR. Part of that premium is belief in this model.13
Overall weak points
- Low transparency. There is no public rate, no tier and no self-serve entry.
- Margin depends on the resolution rate and on model cost.
- The price anchor (human cost) gets weaker over time.
- The model is no longer unique in 2026.
What is Sierra's outlook, and what can it do better?
- Growth is very strong. ARR went from about $20M (Oct 2024) to about $200M (May 2026). That is about 10× in 19 months.15, 13
- Sierra reported its "first-ever $50M quarter" in early 2026.2 This likely means about $50M of new ARR in one quarter. If it keeps about $50M of new ARR per quarter, ARR passes $300M by about the end of 2026. This is a simple projection, not a forecast. Analysis
- The main growth levers: Horizon (revenue work), voice, more countries, marketplaces and partners. The main risks: consolidation, price pressure, and margin.
Growth lever · Horizon
Moves Sierra from cost savings to revenue: loans, renewals, sales. Sierra calls it "the most significant expansion of Sierra since we launched." It also upsells the existing base.5
Growth lever · New channels
Voice is now the main channel. Stripe payments in the call. Personal Agent Protocol prepares for AI agents that shop and ask for service for people.15, 8, 40
Growth lever · Reach
Offices in about ten cities. Acquisitions in Japan and France. A partner ecosystem and Google Cloud Marketplace make it easier to buy.41, 39, 6, 7
Growth lever · Lower cost to deliver
Ghostwriter builds agents from documents. In-house models such as fleming-1 can lower inference cost. Both help margin and smaller deals.4, 8, 12
Pricing recommendations
These are our ideas, based on the public facts above. Analysis
Publish an outcome standard
Define each outcome type in writing. Add verification (for example, a reopen window and a CSAT floor) and an audit view for the customer. This answers the trust question before the buyer asks it.
Make an outcome rate card by value class
Use different rates for information, transaction, retention and revenue outcomes. This fixes the "easy cases first" mix problem and lets Sierra capture more value on hard tasks.
Add predictable commercial structures
Offer prepaid outcome bundles with volume steps, a floor and a cap, and a true-up. Give multi-year rates with planned step-downs. The customer gets a stable budget. Sierra gets a forecast and protection against competitor price cuts.
Package a standard offer for the mid-market
Use Ghostwriter and marketplaces for a lighter offer: fixed platform fee, standard outcome rates, little custom service. Sierra can keep the price private but make the structure simple.
Price Horizon as fee plus success fee
Use a base fee per active goal plus a success fee when the goal completes. Measure the increase with holdout groups. This splits risk and reduces attribution disputes.
Protect margin per outcome
Track cost to serve for each outcome type and channel. Add a per-conversation floor for voice. Move traffic to cheaper models when quality allows.
Prepare a price for agent-to-agent traffic
As personal AI agents call businesses, define the outcome and price for these conversations before buyers ask for a discount.
Signals to watch next
Interview prep
Questions that the interviewers may ask, with a short plan for each answer. Then good questions to ask them.
"A customer says Fin costs $0.99. Why should they pay us $1.50?"
- Change the comparison from rate per resolution to total cost per contact. See the example table in Q5.
- Show the resolution-rate gap, the harder tasks Sierra can do, and voice quality.
- Offer a structure, not a discount: volume steps or a multi-year rate path.
"How would you price Horizon for a mortgage lender?"
- Define the goal: for example, an application that is complete and submitted. Do not use "loan funded," because many other factors affect it.
- Price a base fee per active goal plus a success fee. Anchor the success fee on the cost of a loan officer's time, or on a share of the measured increase in completed applications.
- Use a holdout group for attribution. Set a cap for budget certainty.
"The CFO wants a fixed budget. What do you offer?"
- Annual commitment based on forecast volume. Prepaid outcomes at a lower rate. Overage at the list rate.
- A collar: a floor protects Sierra and a cap protects the customer. True-up at year end.
"How do you define a 'resolution'?"
"Voice or chat: same price?"
- No. Voice costs more (telephony, speech) and replaces more expensive human work. Use a higher outcome rate or a per-minute floor for voice.
"Build a quick ROI case: 5M contacts per year."
- Use the calculator in Q5. Example: 5M × 60% = 3M resolutions. At $8 human cost, that is $24M of avoided cost. At $1.50, Sierra earns $4.5M plus the platform fee. The customer keeps more than 75% of the value.
- Then add revenue effects: saved cancellations and upsells.
Good questions to ask the interviewers
- How do you resolve disputes about whether an outcome happened?
- How is revenue split between platform fees, outcome fees and services? How do you want this mix to change?
- How are you pricing Horizon in the first deals?
- Does Ghostwriter change how you price implementation?
- How do marketplace deals (Google Cloud) change discount and approval rules?
- How does the pricing team work with the deal desk and finance on forecasting variable revenue?
Glossary
- ARR
- Annual recurring revenue. Current monthly recurring revenue × 12.
- ACV
- Annual contract value. The yearly value of one customer contract.
- NRR
- Net revenue retention. Revenue this year from last year's customers ÷ their revenue last year.
- Outcome-based pricing
- The customer pays only when the software delivers an agreed result.
- Resolution / containment
- The AI agent completes the customer's request with no human help.
- Escalation / handoff
- The AI agent sends the case to a human.
- Platform fee
- A fixed fee for access to the platform, usually per year.
- Commit / true-up
- The customer agrees to a minimum volume. At the end of the term, both sides settle the difference with actual use.
- Collar
- A floor (minimum bill) and a cap (maximum bill) on a variable fee.
- Rate card
- A list of unit prices, for example a price for each outcome type.
- Deal desk
- The team that approves non-standard prices and terms in a deal.
- FDE
- Forward deployed engineer. An engineer who works directly with the customer to build the solution. Palantir made this model famous.
- COGS / gross margin
- Cost to deliver the product. Gross margin = (revenue − COGS) ÷ revenue.
- Inference
- Running an AI model to produce an answer. It costs compute money for each request.
- CSAT
- Customer satisfaction score, often on a 1–5 scale.
- BPO
- Business process outsourcing. A company that runs a contact center for another company.
- Variable consideration
- An accounting term (ASC 606) for a price that depends on future events, for example outcome fees.
- Holdout group
- A control group that does not get the AI agent. It shows the real increase that the agent causes.
References
The numbers in square brackets in the text link to this list. Sierra is a private company. Its own numbers come from its blog and interviews. Prices for Sierra come from third-party estimates. Check the newest sources before the interview.
- [1]Sierra · Outcome-based pricing for AI agents (Dec 10, 2024)Sierra
- [2]Sierra · Year two in review (Feb 6, 2026)Sierra
- [3]Sierra · Agent OS 2.0 (Nov 5, 2025)Sierra
- [4]Sierra · Ghostwriter: agents as a service (Mar 25, 2026)Sierra
- [5]Sierra · Horizon (Jul 16, 2026)Sierra
- [6]Sierra · The Sierra partner ecosystem (Oct 6, 2026)Sierra
- [7]Sierra · Sierra is on Google Cloud Marketplace (Oct 26, 2026)Sierra
- [8]Sierra · Blog: Personal Agent Protocol, fleming-1, Stripe (Oct 2026)Sierra
- [9]Cheeky Pint podcast · Bret Taylor transcript (Mar 2026)Interview
- [10]Sequoia Training Data · How AI is reinventing software business models, ft. Bret TaylorInterview
- [11]TechCrunch · Sierra reaches $100M ARR in under two years (Nov 21, 2025)Press
- [12]SiliconANGLE · Sierra valued at $15B in new $950M round (May 4, 2026)Press
- [13]Crypto Briefing · Sierra hits $200M annualized revenue in May (Aug 2026)Press
- [14]Bloomberg · Sierra valued at $10 billion (Sep 4, 2025)Press
- [15]Sacra · Sierra revenue, valuation and fundingAnalyst
- [16]Contrary Research · Sierra business breakdown and founding storyAnalyst
- [17]UsagePricing · Sierra pricing blueprintPricing tracker
- [18]ValueAdd VC · How Sierra AI makes moneyAnalyst
- [19]AgentMarketCap · Sierra acquires Takeoff (Jul 27, 2026)Press
- [20]Revelio Labs · Sierra employee countData
- [21]Built In SF · Sierra pricing role job post (GTM Operations, Pricing)Job post
- [22]SaaStr · The real question for outcome-based pricing is if it will stickAnalyst
- [23]Brian Letort · Outcome pricing goes GA (2026, week 29)Analyst
- [24]Drag · Salesforce buys Fin (formerly Intercom) for $3.6BAnalyst
- [25]CostBench · Fin (Intercom) pricing and resolution rulesPricing tracker
- [26]eesel · Zendesk AI resolution pricing, 2026 guidePricing tracker
- [27]Salesforce Ben · Complete guide to Agentforce pricing optionsPricing tracker
- [28]SiliconANGLE · Decagon raises $250M at $4.5B valuation (Jan 28, 2026)Press
- [29]Sacra · Decagon vs SierraAnalyst
- [30]Macha · Decagon vs Sierra AI (includes Vendr contract data)Pricing tracker
- [31]Harmony · Parloa pricing and packaging explainedPricing tracker
- [32]GetVocal · PolyAI pricing vs alternativesPricing tracker
- [33]Macha · Google Customer Engagement Suite guidePricing tracker
- [34]Microsoft · Dynamics 365 Contact CenterVendor
- [35]Fortune · OpenAI Frontier agent platform (Feb 5, 2026)Press
- [36]Gartner · Conversational AI will reduce contact center labor costs by $80B (Aug 31, 2022)Market data
- [37]Ringly · Call center statistics 2026Market data
- [38]Bloomberg · Sierra valued at $4.5 billion (Oct 29, 2024)Press
- [39]Complete AI Training · Sierra acquires Fragment, third purchase of 2026Press
- [40]CNBC · Meta joins companies on Personal Agent Protocol (Oct 6, 2026)Press
- [41]ZeroG Talent · Sierra hiring: agent engineers and forward deployed rolesJob market
- [42]Thrive Capital jobs · Sierra Strategist, Agent DevelopmentJob post
- [43]Second Order Labs · AI inference costs and the end of 80% SaaS gross marginsAnalyst
- [44]Sierra · τ-bench: Benchmarking AI agents for the real world (Jun 20, 2024)Sierra
- [45]eesel · Sierra AI pricing explainedPricing tracker
- [46]TechCrunch · Parloa triples its valuation to $3B with $350M raise (Jan 15, 2026)Press
- [47]FDE Academy · How Palantir invented the forward deployed engineer modelAnalyst
- [48]Shivanath D. · The difference between a forward deployed engineer and a consultantAnalyst
- [49]Business Wire · Palantir reports Q4 2025 results (Feb 2, 2026)Press release
- [50]The Strategy Story · Palantir business strategy 2026Analyst
- [51]Sierra · Careers page (job families)Sierra