Quick answer
Outcome-based AI agent pricing is worth it when the job has a clear, countable finish (such as a resolved support chat or a paid invoice), you can check the agent’s results yourself, and the price per outcome is lower than what that result costs you today. Before signing, get the vendor’s exact definition of an outcome in writing, ask how disputed or reopened cases are handled, and set a monthly spending cap.
Key takeaways
- You pay for finished results, not users or usage
- Best for simple, closed tasks with an obvious “done”
- The definition of an outcome decides what you really pay
- Compare cost per outcome with your current cost per result
- Always set a spending cap and check a sample of “successes”
📘 Main article: This guide follows our article Per Seat vs Per Token AI Pricing: Why It’s the Wrong Question (and What to Ask Instead), where we explain why paying per result is gaining ground and where it falls short.
In this guide
- What is outcome-based AI agent pricing?
- When is paying per result a good fit?
- When should you avoid it?
- How do you check what counts as an outcome?
- How do you compare cost per outcome with seats or usage?
- Which questions should you ask before signing?
- How do you test an outcome-priced agent safely?
- Frequently asked questions
- The bottom line
- Related posts
Some AI customer service and sales agents are now sold per result instead of per user or per message. On paper it sounds ideal: you only pay when the job gets done.
As we cover in our main article, AI strategist Nate B Jones thinks this model works well for closed, simple tasks and much less well for complex ones. This guide helps you tell which side of that line your offer falls on.
What is outcome-based AI agent pricing?
It means the vendor charges you each time the AI agent achieves an agreed result. Common examples include:
- A customer support conversation resolved without a human
- An appointment booked
- A bill or invoice paid during a call or chat
- A qualified lead handed over to sales
Conversations the agent cannot finish, or hands to a person, are usually not charged, but check this, as terms vary.
When is paying per result a good fit?
| Signal | Why it matters |
|---|---|
| The task has a clear end point | Easy to count and hard to argue about |
| You handle a steady volume | Per-result pricing scales cleanly with demand |
| You already know your cost per result | You can tell if the price is a good deal |
| Results show up in your own systems | You can check the vendor’s count |
| Mistakes are cheap to fix | Lower risk if the agent gets one wrong |
💡 Tip: If you can describe the outcome in one sentence a customer would agree with, such as “my question was answered and I did not need to contact you again”, it is probably a good candidate.
When should you avoid it?
- Complex or creative work with no single finish, such as strategy, design or long projects
- Low volume tasks, where a flat plan may be cheaper and simpler
- High-stakes outcomes such as refunds, legal or medical advice, where a wrong “success” is costly
- Cases where you cannot see the data needed to check the vendor’s count
⚠️ Warning: An agent paid per resolution has a reason to mark conversations as resolved. That does not mean any vendor does this, but it does mean you should check a sample yourself. Our guide on spotting reward hacking in AI agents explains the warning signs.
How do you check what counts as an outcome?
Ask the vendor for its exact definition in writing, then test it against real examples. Look for answers to these points:
- What triggers a charge? For example, the customer stops replying, says thanks, or confirms the answer helped
- What if the customer comes back? Is a reopened case within, say, 24 or 72 hours still charged?
- What about hand-offs? If the agent passes the chat to a person, is it free?
- Who decides? Can you see and dispute each charged outcome?
Then take ten recent conversations from your own business and decide, using their definition, which would have been charged. If you disagree with several, the definition is too loose.
How do you compare cost per outcome with seats or usage?
Work out what one result costs you today, then compare. Use this simple method:
- Total your current monthly cost for the task (staff time plus tools)
- Divide by the number of results per month to get cost per result now
- Multiply the vendor’s price per outcome by your expected number of AI-resolved results
- Add any platform or base fees
- Compare the totals, and remember you still pay for the cases the AI cannot finish
Illustrative example only: if you handle 400 support questions a month at a current cost of £1,200, that is £3 per question. An agent at £1 per resolution that resolves 250 of them costs £250, leaving 150 for your team. The saving only holds if those 250 are genuinely resolved.
For usage-priced tools, use our guide How to Work Out the Real Cost of an AI Task Before You Switch Models. For credit and minute-based plans, see Chirps AI Pro Credits vs Voice Minutes: What Costs Extra? as a worked example.
Which questions should you ask before signing?
- Is there a monthly cap or alert if outcomes spike?
- Are there minimum commitments or base platform fees?
- Can I export the conversation logs behind each charge?
- How do I dispute a charge, and how fast are disputes handled?
- Can I switch to a seat or usage plan later?
- Where is customer data stored, and is it used to train models?
📘 Main article: Not sure if any AI pricing model suits you yet? Start with Per Seat vs Per Token AI Pricing: Why It’s the Wrong Question (and What to Ask Instead), then measure your results using our AI ROI guide.
How do you test an outcome-priced agent safely?
- Start small: one channel or one type of question
- Set a hard monthly cap on charges
- Check 10–20 charged outcomes a week by reading the conversation
- Track complaints and repeat contacts from customers the agent “resolved”
- Review after 30 days: compare your cost per result before and after
Before you sign any AI agent deal: calculate your current cost per result, then follow our 30-day ROI method to prove the agent pays its way.
Frequently asked questions
Is outcome-based pricing cheaper than per seat?
Not always. It is often cheaper for high-volume, simple tasks, but a flat per-seat plan can be cheaper if volumes are low or the agent resolves few cases.
What does “per resolution” mean for AI support agents?
You pay each time the agent fully handles a customer question without a human. The vendor’s written definition decides exactly when that applies.
Can I dispute an AI agent’s outcome charges?
Usually, but check the contract. Ask for exportable logs and a clear dispute process before you sign.
Is outcome pricing suitable for complex projects?
Rarely. Complex work has no single, clear finish, so per-result pricing is hard to measure and easy to argue about.
How do I stop an outcome-priced agent from running up a big bill?
Set a monthly cap or alert, start with one channel, and review a sample of charged outcomes every week.
The bottom line
Paying per result can be a fair deal for simple, high-volume jobs, as long as you agree what a result is and can check it. Your next step: work out what one finished result costs you today, so you can judge any per-outcome offer in minutes.
Related posts
- Per Seat vs Per Token AI Pricing: Why It’s the Wrong Question (and What to Ask Instead)
- How to Spot Reward Hacking When an AI Agent Says the Job Is Done
- Chirps AI Pro Credits vs Voice Minutes: What Costs Extra?
