Original practice question. The scoring rubric is our teaching guide. How we research and score answers
TL;DR: Seat pricing is simple and predictable but exposes margin to heavy users and to AI reducing the number of seats needed. Usage pricing tracks cost but makes customers anxious about spend. Outcome pricing aligns best with value but depends on a defensible definition of the outcome. Match the model to how value is created and to who carries the risk when the AI underperforms. A hybrid, a base subscription with included usage plus usage or outcome charges above it, is a common answer; work the numbers before choosing it.
How to approach it
Lay out each model's strengths and risks, then decide based on the product: how predictable usage is, how measurable outcomes are, what buyers expect, and what the cost per use looks like.
A strong answer
The three models.
| Model | Strengths | Risks | Fits when |
|---|---|---|---|
| Per seat | Simple, predictable for buyers and sellers | Heavy users can cost more than they pay; if AI does work that people used to, seats shrink | Usage per user is fairly even; AI is a feature in a broader product |
| Per usage (credits, tokens, tasks) | Tracks cost; scales with value | Budget anxiety; customers ration use; hard to forecast | Developer platforms; variable, heavy workloads |
| Per outcome (resolved ticket, processed document) | Aligns price with value; easy ROI story | Defining the outcome; disputes; revenue volatility | Outcomes are clear, measurable and attributable |
Outcome pricing in practice. Fin (formerly Intercom, part of Salesforce since September 2026) lists $0.99 per outcome on its pricing page, covering resolutions, Procedure handoffs, disqualifications and self-serve routings, and $9.99 for a qualified sales lead (checked November 2026). Check the full billing rules before comparing vendors. Zendesk solves the definition problem differently: only a "verified resolution", confirmed by a language model reading the finished conversation, counts against the allowance, while assisted escalations and contained resolutions do not (Zendesk, May 2026). The PM must define an outcome the buyer can audit, because that definition becomes the invoice.
The seat-compression problem. If an AI agent does work that people used to do, a per-seat price falls as customers need fewer seats. A fictional help-desk customer shows the size of it: 40,000 tickets a month, 50 agents on $115 seats, each agent handling 800 tickets. Our AI agent attempts every ticket at $0.12 and resolves 40% of them, so 16,000 tickets no longer need a person and the customer needs 30 seats.
| Pricing | Our monthly revenue | Margin on AI |
|---|---|---|
| Seats only, before AI | $5,750 | n/a |
| Seats only, after AI | $3,450 | AI cost unpaid |
| Seats plus $0.99 per resolution | $3,450 + $15,840 | 70% |
| Seats plus $0.40 per conversation | $3,450 + $16,000 | 70% |
Seats alone lose 40% of revenue as the product works. The two AI charges earn the same in a normal month and differ in who carries a bad one. If resolution falls to 25%, per-resolution revenue drops to $9,900 and our margin to 52%; per-conversation revenue holds, and the customer pays $1.60 per resolution instead of $1.00.
Our AI cost is 40,000 × $0.12 = $4,800 a month, or $0.30 per resolution. If the customer really removes 20 agents at a loaded $5 a ticket, it saves $80,000 a month in labour and $2,300 in seats against $15,840 of outcome charges. That saving exists only if headcount or overtime actually falls; a buyer who redeploys the agents sees the bill and not the saving. Outcome pricing also needs room for disputes: crediting 10% of billed resolutions for reopened conversations takes margin from 70% to 66%.
I would price this product per resolution on top of a reduced seat price, because the outcome is observable and the buyer can audit it. I would switch to per-conversation pricing if resolutions cannot be attributed (the customer's other channels are invisible to us) or if resolution rates vary so much by customer that we cannot carry the risk. Seats stay right for a copilot that makes people faster without removing work, because seat counts do not shrink.
A common hybrid, now on the rate cards. GitHub's Copilot Business moved on 1 June 2026 to "$19/user/month, including $19 in monthly AI Credits", with credits consumed by token usage (GitHub changelog): a seat price with an included allowance. Salesforce's Agentforce pricing puts three models on one page: $2 per conversation, Flex Credits at $500 per 100,000 (an action costs 20 credits) and per-user add-ons (checked November 2026; list prices, subject to change). The shape:
- Base subscription per seat or per workspace, including a usage allowance that covers most users.
- Usage or outcome charges above the allowance, or premium tiers for heavier use.
- Enterprise commitments with pooled allowances and predictable invoices.
Check the economics. Whatever the model, compute gross margin by customer segment, watch the heaviest users, and model what happens as model prices fall (margin improves if price is anchored to value) and as usage grows.
What interviewers probe next
"Customers say usage pricing makes budgeting hard." Offer committed plans with pooled allowances and alerts before overage, so spend is predictable.
"Our agent reduces the number of support agents customers need. How should that affect pricing?" Price per resolution or per task, so revenue grows with the work automated rather than shrinking with headcount.
How it's scored
| Signal | Weak | Hire | Strong Hire |
|---|---|---|---|
| Models | Picks one without trade-offs | Compares three | Strengths, risks and fit conditions for each, including seat compression |
| Outcome pricing | Not considered | Mentions it | Notes the definition problem with an accurate, attributed public example |
| Design | Single model with no numbers | Compares revenue under two models | Works seat compression, margin and the customer's saving for the chosen structure, says who carries a bad month, and names what would change the choice |
Common mistakes
Seat pricing for a product whose purpose is reducing seats.
Outcome pricing without a clear, auditable definition, or without a credit budget for disputes.
Counting the customer's labour saving as certain when they may redeploy the people instead.
Ignoring heavy users' cost under flat pricing.
See pricing AI products, defining 'resolved' for outcome pricing and agents will shrink our seat count.