Success starts with three things to bridge the organizational gap.

1. - **The translation table.** Guide the CFO through the metrics their infrastructure is already producing and what each one means in financial terms. The goal is not to explain the technology but to establish that infrastructure observability and financial reporting are currently describing the same business in two semantics that don't talk to each other.
2. - **One concrete gap.** Pick the question with the highest near-term relevance for your business from our samples. Or better yet, articulate one from your organization's current-year business plan. In our conversations, developer consumption of  LLM provider tokens or cost exposure in long-running agentic workloads tends to be of high interest. From the AI connectivity layer, show specifically what data exists, where it lives, and what integrating it into the other systems might require. Concrete is always more useful than comprehensive in a first conversation.
3. - **A metering proposal.** Not a full project plan, but an art of the possible. A clear statement of what instrumentation investment would produce what financial visibility, and what decisions that visibility could support. The CFO's job is to decide whether the investment is worth making. Give them what they need to make that call.

The conversation you're trying to have is not a "please look at all this data we already have" conversation. Instead, you're looking to say: "Our technical control plane is also a financial control plane we don't currently leverage. Here's what it would take to enable it, and here are the decisions it would let us make." That's a conversation any CFO is ready for.

Finance teams aren't equipped to find this problem on their own. They don't have access to the gateway metrics, and even if they did, raw infrastructure data doesn't arrive in a form finance can act on.

What they do know is that AI spending is growing, for most gross margin isn't growing with it, and nobody has given them a model for understanding the relationship between the two.

That's not a finance department problem. It's a data routing and reporting problem. The [AI gateway ](https://konghq.com/blog/enterprise/what-is-an-ai-gateway)AI gateway is already capturing everything the CFO needs to understand AI economics at the unit level: what each workload costs, what each customer costs to serve, where cost is being optimized automatically, and where it's running unchecked. The gap is that this data lives in infrastructure observability tooling, not in financial reporting.

When you walk into the CFO conversation, you're not asking for more budget. You're offering them visibility and insight. That's a different conversational dynamic and a much more productive one.

Every metric your gateway produces has a financial equivalent. You don't need to teach your CFO what p99 latency is. You need to show them how it maps to profits and losses. These are generic mappings that you can take and better align with your core business.

Every row follows the same pattern: a technical metric is a proxy for a business event. The gateway is producing signals about all of them in real time. The question your CFO will ask, once they see this table, is why they haven't been receiving these signals before. Have a good answer ready.

The table above covers operational metrics or simply describes how the infrastructure is performing. If this is all new to your CFO, this might be a good place to stop, but if there's capacity, we can pivot from AI utilization to AI productization.

The second set that goes deeper: monetization signals, agentic cost risks, and unit economics that sit closer to the revenue and margin conversations your CFO is already having. These are the metrics most directly connected to pricing decisions, contract compliance, and the question of whether your AI product is actually profitable at the customer level.

These are questions finance should be asking about AI economics that most aren't because the data hasn't been made available in a form that makes them natural to ask. Part of your job is to surface not the raw data but consumable data. Here are some example questions that can aid in the conversation. 

**Question 1: What is our gross margin per AI interaction, by user segment?**

Aggregate gross margin can look acceptable while specific user segments or feature workloads run margin-negative. You can't see it in the top-line numbers. You need cost and revenue attribution at the level of individual customers and agentic workloads.

Most organizations cannot answer this today. The token consumption data exists in the gateway. The revenue data exists in the billing system. Nobody has taken the time to connect them.

Put it to your CFO this way: “We don't currently know what it costs to serve each of our top accounts at the interaction level. If one of them is margin-negative at their current usage tier, we'd want to know before the next pricing conversation, not after.”

**Question 2: What is our cost exposure to LLM provider pricing changes?**

Every AI product routing through third-party large language models has direct exposure to repricing by those providers. Most finance teams have no quantified view of this because the token consumption data that would allow it doesn't route to finance. When a provider changes pricing, it shows up as a margin compression event in the quarter because the organization never had the data to forecast it.

The gateway captures every token consumed, by model, by workload. That's enough to model the P&L impact of a pricing change from any given provider before it happens.

Put it to your CFO this way: “If a major LLM provider repriced tomorrow, we'd find out what it meant for our margin at month-end close. With two weeks of instrumentation work, we could model that scenario in an hour.”

**Question 3: What margin is the gateway already creating, and is anyone measuring it?**

Model routing, semantic caching, and request optimization reduce inference costs without changing the user-facing pricing. The same revenue, at lower cost, with no degradation in the product. This is the margin the gateway is producing automatically, right now.

Most organizations can't quantify it. Which means it's happening outside any financial accountability structure. Nobody is getting credit for it. Nobody knows if it's worth investing in further.

The framing here is different from the first two questions. This isn't about a gap that needs closing instead it’s about value that already exists and isn't being measured. Put it to your CFO this way: “Our gateway is doing cost optimization automatically. We know it's working. We can't tell you by how much, because we've never used it as a financial metric. We should and easily can because the answer will inform every conversation we have about pricing and infrastructure investment.”

**Question 4: Do you know which internal teams are driving AI spend, and do these teams also know it?**

External customer economics get most of the attention, but internal developer usage is where unattributed AI spend quietly accumulates. A developer running an agentic workflow in a shared environment has no reason to optimize it. The cost is invisible to them and lands somewhere else on the P&L. Finance sees an aggregate LLM bill. The platform team sees traffic. Nobody sees which internal team's inefficient agent is responsible for 40% of token consumption this month.

This is the AI equivalent of shadow IT. The spend is authorized. The attribution isn't.

The gateway is the only place where that changes in a unified way for every vendor and model. The gateway can tag every request by team, cost center, workload type, and environment. The same metering schema that creates customer-level visibility creates internal chargeback capability. Without it, there's no basis for team-level AI budgets, no signal for identifying inefficient workloads, and no accountability structure connecting developer behavior to cost outcomes.

Put it to your CFO this way: “We don't have a consistent cost center attribution for internal AI usage. We can't do chargebacks, can't set team-level budgets, and can't identify which internal workloads are cost-efficient and which aren't. That data exists in the gateway. Right now it goes nowhere.”

The most important technical point to communicate in non-technical terms is this:

Most AI products and platforms have engineering-grade observability. They can tell you how many requests were processed and at what latency. Very few have finance-grade metering: the ability to attribute workloads and cost to serve a specific customer, on a specific workload, at a specific model utilization level, compared to what that customer is paying.

At Kong, we 've designed enterprise observability and finance-grade metering into the core of our platform. No need to retrofit multiple solutions into a live production system. No need to wait months before converting raw data into business insights. No gap in governing your AI connectivity in alignment with business outcomes. 

You already have the right tools. Now is the time to enable them.

The ask for the CFO is not budget for a new system, but leveraging the system you already have in place. It's alignment on what questions the metering infrastructure needs to answer, so the engineering gets to the right problems from the start. That requires the CFO to say what financial visibility they actually need and they can't do that if nobody has told them the data is already available.

That's the initial conversation. Start it.

*For more about AI cost control, watch *[*Agentic AI Cost Control for the CFO: Attribution & Visibility*](https://konghq.com/events/webinars/agentic-ai-cost-control-for-the-cfo-attribution--visibility)*Agentic AI Cost Control for the CFO: Attribution & Visibility**.*

**1. Why present AI metrics to a CFO in financial terms?**

CFOs speak the language of risk, ROI, and cost predictability. Raw technical metrics like tokens per second create confusion, not confidence. Translating AI performance into financial impact shows how your AI initiatives connect to the company's bottom line.

**2. What is an AI gateway, and why does it matter to finance teams?**

[Kong AI Gateway](https://konghq.com/products/kong-ai-gateway)Kong AI Gateway is the management layer between your applications and LLM providers like OpenAI or AWS Bedrock. For CFOs, it delivers a unified point for cost attribution: finance teams can see exactly which business unit or product is driving AI spend, shifting the conversation from "why is this bill so high?" to "which team is generating the most value?"

Beyond visibility, Kong AI Gateway actively governs AI spend. Semantic caching reduces redundant LLM calls by reusing responses to similar prompts. Token-based rate limiting enforces per-team or per-product consumption quotas before costs escalate. And semantic routing directs requests to the most cost-effective LLM provider without sacrificing output quality. The result: AI FinOps built into your infrastructure, not bolted on after the fact.

**3. What is token efficiency, and how does it connect to budget?**

Token efficiency means getting the best AI output for the lowest token cost. For a CFO, this is an operational margin story. Improving token efficiency through semantic caching and semantic routing lets your organization handle more queries and process more data without a linear increase in LLM spend.