BlueprintEconomics
Talking to your CFO about outcome-based pricing for AI Agents
How to get budget approval for your customer service Agent
Outcome-based pricing has become increasingly common for AI Agents in customer service. In this model, customers only pay when an AI Agent delivers a defined result, keeping pricing and value closely linked. This helps customers easily understand what they are paying for while incentivizing vendors to constantly improve their product.
But for all its benefits, outcome-based pricing is still a variable line item you will need to explain to your CFO. They're used to approving annual fixed expenses, like total headcount or software seats, not Agents that may vary in cost month to month.
However, variability is not the same as unpredictability. Costs under outcome-based pricing in customer service are driven by factors you can forecast, allowing you to create grounded budgets and corresponding guardrails. This way, you can assure your CFO they won’t be surprised by runaway expenses while granting your Agent the bandwidth to serve as many customers as possible.
Let's look at what's required to build a forecast.
Build a baseline case
A simple baseline forecast of annual spend under outcome-based pricing comes down to four metrics multiplied together. If your AI Agent is already in production, you'll have easy access to these. If not, you can approximate them using past support data and industry benchmarks.
- Annual conversation volume: The total number of support conversations you see in a year.
- Involvement rate: The share of conversations your AI Agent participates in.
- Resolution rate: The share of involved conversations the Agent resolves.
- Price per resolution: Your unit cost.
Here's a breakdown of best practices for collecting this information, and some considerations to keep in mind while building your projected budget:
Conversation volume
When exporting the last twelve months of conversation volume from your helpdesk, make sure to note month-to-month fluctuations. If your Agent uses monthly billing and you experience significant swings in conversation volume throughout the year, costs will rise and fall accordingly, so communicate this to your finance team.
For example, Kalshi, a regulated prediction market in the United States where people trade contracts on real-world events, saw a record number of support conversations on Super Bowl Sunday.
Involvement rate
Your involvement rate can be impacted by a number of factors, like how many channels you deploy your Agent to or the topics you allow it to handle. If you operate in a highly regulated industry where an Agent is not permitted to answer certain queries, your involvement rate will be lower.
During initial deployment, teams typically run a phased rollout so involvement rate steadily climbs as the Agent’s scope expands. Consequently, total resolutions and costs may be lower in your first few weeks.
One in nine companies that use Fin have an involvement rate of 100%. To see an industry specific breakdown of involvement rates, head over to our benchmarking tool.
Resolution rate
Your resolution rate will be impacted by your query mix – the percentage of queries that are informational, personal, or action-taking – as well as how your Agent is set up.
The goal is for your Agent to handle queries across all three categories. But informational queries are the most readily handled out of the box, so long as your help center content is in good shape, while personal and action-taking queries typically require more configuring.
“I want to cancel my subscription” may sound straightforward, but it might require an Agent to check what plan the customer is on, look up their billing history, apply a retention policy, and trigger a cancellation through an API.
As a team’s deployment matures and their content and configuration improves, resolution rates rise. Fin customers see a median increase of 16 percentage points in the first 18 months.
You can approximate your query mix by analyzing your past conversations against the above definitions using an LLM. But be aware that this requires some dedicated effort, as building a reliable classification process takes good prompt engineering and validation. Fin users can run this analysis automatically using Insights.
If you're unable to complete a query mix analysis but still want a sense of what your resolution rate could be, here are the median first-month resolution rates by industry, based on Fin customer data.
| Industry | Median Resolution Rate |
|---|---|
| Agriculture | 66.1% |
| eCommerce | 52.1% |
| Education | 60.0% |
| Energy | 53.3% |
| Fintech / Financial Services | 53.1% |
| Gaming & Gambling | 43.6% |
| Healthcare | 57.3% |
| Hospitality & Tourism | 60.0% |
| Manufacturing | 63.3% |
| Media & Entertainment | 55.7% |
| Professional Services | 58.5% |
| Real Estate | 55.7% |
| Software & Technology | 54.0% |
| Telecommunications | 49.3% |
| Transportation | 51.4% |
Build your optimistic and conservative cases
Now that you have your baseline, create optimistic and conservative cases so you can present a budget range to your finance team.
Your baseline case assumes your conversation volume will remain unchanged from the year prior and your resolution rate stays flat after launch. Your conservative case will assume your conversation volume will fall while your resolution remains flat. Your optimistic case assumes conversation volume increases along with a higher resolution rate, reflecting the ramp-up you'd expect as your deployment matures. To keep things straightforward, the calculation applies that higher rate across the full year rather than modeling the ramp itself.
This approach is deliberately simple. More sophisticated methods exist – time-series models like ARIMA can account for seasonality and growth trends – but for most teams, the three-scenario framework outlined above is enough to get budget approval.
Get budget approval
Bring your conservative, baseline, and optimistic cases to your finance team to set expectations for the range of costs you may incur during the year, and flag any anticipated seasonal spikes in volume so budgets can be distributed accordingly.
Push for approval of the optimistic case, as it gives you the most room to invest in improving performance while absorbing increases in volume. It's worth reassuring your team that these costs can be capped with hard resolution limits so the budget they agree to won't be exceeded.
You should also take the time to make a contingency plan. If you hit your usage cap during the year, your Agent will be unable to help customers, placing sudden strain on your human support reps. Set aside time to routinely check in on total resolutions throughout the year, and keep finance leaders in the loop if volume is tracking toward the higher end of your range.
Exceeding your optimistic case would be unusual – it would mean far more conversations than expected, or a resolution rate well above your projections. But either way, it's worth being prepared.
Addressing common concerns
When presenting your budget, be ready to address finance leaders' concerns about the Agent and its pricing model. Here's what to highlight to help get their approval:
Variability vs. fixed costs: When volume spikes, an AI Agent scales instantly. The alternative is a backlog that hurts CX, followed by a slow hire-and-train cycle. And if volume drops back down, you're stuck with headcount you no longer need. Outcome-based pricing scales both ways and you only pay when the Agent resolves something.
This isn't runaway token burn: Stories of uncontrolled AI spend have made finance leaders wary. But outcome-based pricing is fundamentally different – costs are directly tied to results. Fluctuations in billing arise due to demand you already have visibility into.
Speed and adaptability: You can expand your Agent's scope (for example, expanding to new channels or query types) without procurement cycles or hiring timelines. Costs scales with output, not with setup.
What a strong forecast looks like
If you've worked through the steps above, you should have:
- An understanding of what factors drive AI Agent costs and how to forecast them.
- A baseline case built from your own conversation volume and price per resolution, plus realistic estimates of involvement rate and resolution rate.
- Conservative and optimistic cases that set the upper and lower bounds of your budget.
- An approved budget, with guardrails in place so costs stay controlled even if performance runs ahead of plan.
- A monitoring plan that tracks resolutions against your range and flags finance early if you're trending toward your cap.
Outcome-based pricing doesn't eliminate variability, but you can make costs predictable and controllable under this model. And with costs tied to outcomes, you can assure leadership of an Agent’s value to the business.
Eoin Gilligan Martyn is Senior Data Scientist at Fin. Connect with him on LinkedIn.
Stay up to date with us on LinkedIn
Follow Fin for the latest research, guides, and product updates on AI customer service.