Agent Pricing Committee

Agent pricing committee

Pricing decisions fail from a single perspective: the finance view raises until churn, the sales view discounts until margin vanishes, and the product view prices the build cost rather than the value. A committee forces the three arguments to meet before a number is chosen.

Team

• Value advocate (customer-interviews): argues from what the customer gains and what alternatives cost them.
• Margin advocate (saas-metrics): argues from unit economics, cost to serve, and payback.
• Competitive analyst (agent-competitive-analysis-team): places the option against real market alternatives.
• Synthesiser: writes the decision pack, including the case against the recommendation.

Shape: parallel advocates on a shared option set, converging into one pack for a human decision.

Method

1. Define the options before the arguments. Two or three concrete structures with numbers, not an open discussion. Advocates critique options; they do not invent new ones mid-debate (see saas-pricing).
2. Make each advocate one-sided on purpose. Balanced agents produce mush. The value advocate should genuinely push for the higher price and the margin advocate for the sustainable one.
3. Quantify the sensitivity. For each option, what happens at plus or minus twenty percent volume, and where does it stop working. A price with no break-even analysis is a guess with a decimal point.
4. Name the migration cost explicitly. Existing customers, grandfathering, contract terms, and the support load of a change are usually larger than the modelling suggests.
5. State the reversal path. How you would walk a price change back and what it would cost in trust, because pricing changes are far harder to undo than to make.
6. Deliver one pack with a recommendation and its strongest objection. A human decides. A pack that only argues one way has failed at its job (see devils-advocate).
7. Set a review date with the metric to watch. Conversion, churn, and average deal size after the change, checked on a stated date.

Run it

In Claude Code, define the option set in a file, spawn the three advocates in parallel each writing their own argument file, then run the synthesiser over all three. The human makes the call. Port to AutoGen as a GroupChat with fixed roles, or to CrewAI as parallel tasks feeding one synthesis task.

Signals it works

• Each option carries a break-even and a sensitivity range.
• The pack includes the best argument against the recommendation.
• Migration and reversal costs are stated, not discovered later.

Boundaries

Pricing is a business decision with legal and contractual consequences and belongs to a human. Agents must not change live prices, alter customer contracts, or communicate changes. Price coordination with competitors is illegal in most markets, so competitive analysis stays strictly to public information.