Unit Economics

Unit economics describes the profitability structure per “unit” (user, task, contract). Traditional SaaS unit economics rests on the assumption of near-zero marginal cost — serving one more user has negligible added expense, gross margin stays at 80%+ year-round, and LTV simplifies to “monthly fee × retained months”.

Agent products break this assumption. Every inference is billed; gross margin is not a constant — it fluctuates with user behavior (task complexity, model selection, cache hit rate). Continuing to apply the SaaS formula produces severely inflated LTV.

Key formula adjustments

Traditional SaaS formula

LTV = ARPU × Gross Margin × Avg Retained Months

Gross Margin is typically taken as the company-wide constant (80%+), assumed independent of the user.

Agent product formula

LTV(per user) = Σ (revenue per task − cost per task) × avg monthly task count × avg retained months
              = gross margin per task × monthly task count × retained months

Key differences:

• Per-task gross margin replaces fixed Gross Margin — each task has a different margin
• Monthly task count replaces fixed ARPU — heavy and light users have vastly different unit economics
• Heavy users are the revenue engine in traditional SaaS (essentially zero cost), but in agent products may be a loss source (more tasks → higher cost → margin compressed)

Decomposing per-task gross margin

Per-task gross margin = revenue per task − Σ per-task cost
Per-task cost = token cost + sandbox / storage cost + HITL labor cost + refund / redo cost × failure rate

Dependencies:

• Token cost — see economics/cost-model; typical medium task $0.10-$0.30
• Sandbox / storage — typically < 1% of total per-task cost; negligible at unit economics level
• HITL labor cost — if tasks require human review, $0.5-$1 per minute (at support hourly rate); for workflows with high HITL intervention rate, this may exceed token cost
• Failure redo — failure rate × (tokens already consumed + human redo time)

Unit economics across three user types

The same product yields drastically different gross margin structures across three user types:

User type	Monthly tasks	Per-task cost (incl. HITL)	Monthly cost	Gross margin (subscription $40 / month)
Light (5%)	5	$0.15	$0.75	98%
Medium (70%)	50	$0.20	$10	75%
Heavy (25%)	300	$0.25	$75	−88%

(Synthetic example illustrating structural differences.)

Heavy users are loss-making under subscription — a pitfall unique to agent products. In traditional SaaS heavy users are “the most valuable customers”; in agent products they may be “the customers most needing to be guided to downgrade or upgrade”.

Response strategies

Four common approaches:

1. Usage caps — hard token limits under subscription, with throttling or suspension on breach. See the tier design in economics/controls-and-roi
2. Hybrid pricing — base subscription + per-token charge above limit. Heavy users automatically bear overage cost
3. Price discrimination — tier by usage profile (Lite / Pro / Ultra); heavy users guided toward higher tiers
4. Accept loss for growth — allow heavy-user losses in early stage, betting on cross-sell or network effects later

The trade-offs between strategies are the central topic of pricing.

Difference from traditional SaaS valuation

Investors typically value SaaS at an ARR multiple (10x-30x). Agent product ARR cannot be applied directly — at the same $10M ARR:

• If gross margin is 80% (mostly light users), valuation approaches traditional SaaS
• If gross margin is 40% (heavy users dominate), valuation moves toward “consulting / services company” levels (2-5x ARR)

Investors will drill down to unit economics to see the substance. Reporting ARR with gross-margin distribution split by user segment is far more credible than reporting a single number.

Cross-section connections

• Per-task cost decomposition: economics/cost-model
• Specific pricing mechanisms for response strategies: pricing
• Operational implementation of user segmentation: playbooks