Measure the cost per outcome, not the cost per request
Per-request pricing is a trap for planning. It tells you what one reply cost, not whether it was worth it. The number that actually runs your business is cost per outcome: what you paid to qualify a lead, resolve a question, or book a meeting.
Why one model can’t win
Pick a cheap model and you’ll fumble the hard cases that matter most. Pick the most capable one and you’ll pay top-tier prices to answer “what time do you open?” Neither is a strategy.
Routing changes the math
When a fast, low-cost model handles the routine messages and a more capable model takes the tricky ones, your blended cost per outcome drops while quality on the hard cases goes up. The shape is simple:
- Simple messages, like opening hours, go to the lowest-cost model.
- Moderate ones, like comparing two packages, go to a balanced model.
- Complex ones, like moving two visits and adding a family member, go to the most capable model. They are the minority, so you pay for that capability only where it matters.
“What time do you open?”
“Compare your two packages for me.”
“Plan a 3-visit treatment around my travel dates.”
When the cap is reached: pause the AI in this chat and alert the team.
Optimize the blend, not the model.
What to actually track
Put three things on the wall: what each resolved conversation cost, how many conversations reached an outcome, and what share of traffic needed the most capable model. When routing is working, the first two improve while the third stays small.
And keep a safety net: set a spend cap per conversation, so a runaway chat pauses the AI and alerts your team instead of quietly burning budget. See how smart routing works.
Build your first AI employee
Pick the outcome you want, connect a channel, and let it start answering your customers.