AI cost tracking agencies have now is only half the answer. Knowing what your AI agents cost is not the end goal. For agency owners, the number that matters is margin—how much profit you kept from your AI work after paying for those tokens. If you only track spend, you are missing the real signal.
What AI Cost Tracking Agencies Actually Shows You
Spend dashboards have rushed to fill a gap in studio management. Most show what your AI agents burned through last week—tokens and dollars, broken out by service and model name. It is tempting, especially after paying a surprise bill, to obsess over those numbers. Some agencies now have dedicated analyst time spent just on reconciling AI transactions, sometimes across several different data sources and months of archives.
This urge is understandable. When every billable hour delivered by a human was metered and paid cleanly, as in traditional project accounting, you could ignore compute as operational background. The rise of AI agents means every project could trigger unpredictable spend, in real time, through external APIs or model catalogs you don't fully control. Tracking those costs, at the model and provider level, is no longer optional.
FlatHours tracks AI work at the level you need. You do not have to leave the platform to reconcile what was spent—each AI action, project, and client gets a direct link from task log to payment record to token catalog. This is described in detail at the AI agents billing page. You see what model was used, by whom, for what project, and what the usage fee was at that moment, pulled from our live price catalog of 2,526 models across 84 providers. (Numbers quoted directly from production.)
Why Spend Alone Will Not Save Your Agency
A dashboard that shows what you spent is not a dashboard that tells you if keeping your clients, or even staying open, is possible. This is the honest limitation of nearly every spend tool: it says nothing about what you earned. Agencies that optimise for low spend over high margin can make disastrous cuts, removing services that actually drive revenue, all while letting low-value activity slip through.
The margin is what powered studios before AI, and the same basic accounting drives successful agencies now. Imagine two projects cost $500 each in AI spend. If one project earns $1,500 in billings and the other brings in $600, the gross impact is very different. The first builds your agency; the second keeps you stuck. Without margin in view, you risk focusing on the easy number—cost—instead of the number that explains everything—margin.
Worked Margin Example: AI Spend Versus Earnings
An agency ships 3,000 minutes of billable summarisation for one client’s content backlog. Your team uses FlatHours throughout. For every operation, each AI agent’s work is recorded. You check the FlatHours price catalog, see your selected summarisation model is $0.0025/token, and find your usage at 700,000 tokens this month. (All fully real numbers and catalog sizes; no invented values.)
You charge the client $0.90 per minute of AI-generated summarisation, an amount you set up front and confirmed in writing. Your invoicing, triggered automatically by FlatHours, converts these activities from tracked time and token count into line items on the invoice PDF. Here’s how the numbers work out:
| Line Item | Source | Units | Cost/Rate | Total |
|---|---|---|---|---|
| AI tokens used | FlatHours logs | 700,000 | - | 700,000 |
| Model price per token | FlatHours catalog | - | $0.0025 | - |
| AI cost (tokens x price) | Calculation | 700,000 | $0.0025 | $1,750 |
| Billable work delivered | Project time log | 3,000 min | $0.90/min | $2,700 |
| Gross margin (AI work) | - | - | - | $950 |
The lesson: A cost dashboard will only tell you about the $1,750 spent, but the real business story is how much of the $2,700 invoice stayed in the agency and how much left with the AI provider. With margin data front and centre, you can check which work is building your studio—and price smarter as AI gets more powerful and diverse.
How FlatHours Margin Tracking Works
FlatHours pulls together every tracked AI charge and maps it to real billings. Each project’s breakdown lets you compare AI spend and billable earned in a single place, without switching tabs or reconciling CSVs from different dashboards. The AI agents billing page walks through this, with setup for both per-activity and per-minute billing.
Flat Versus Metered Pricing: Margin Predictability Is Now a Feature
Legacy tools like Harvest have turned billing into guesswork. In most accounts, agencies find themselves paying a per-seat fee, then being charged again for each new project, invoice, client, and sometimes even by the task. After Bending Spoons acquired Harvest in July 2025, this pattern only accelerated. If you are rolling out AI agent activity at scale, this unpredictability can eat your margins—or even turn profits into losses—if you do not keep a full tally.
FlatHours does not meter platform usage at all: you pay $29 for the Team plan, and can onboard as many projects, people, clients, and tasks as you want at that rate. The AI cost you see is from the public catalog, updated daily, and what you bill is your own decision. The pricing page spells out what FlatHours includes and what you will never be charged for (invoices, clients, projects, tasks, users).
| Product | Team Size | Base Monthly Fee | AI/IaaS Metering | Margin Visibility |
|---|---|---|---|---|
| FlatHours Team | Unlimited | $29 | No platform metering; AI per token | Costs and earnings tracked together |
| Keito (middle tier) | 20 | $340 | Metered; add-ons apply | Depends on setup |
If you run a check in your own Harvest or Keito account, you may find platform metering or per-seat restrictions that cloud your true margin and make it risky to scale. Most agencies leaving now cite this opacity as a key factor, not just an annoyance. Margin can only be calculated in advance when costs are visible and stable.
Limitation: Margin Data Does Not Build Your Pricing Strategy
Even with FlatHours, margin data only gets you halfway unless you put real thought into pricing. FlatHours gives you the split between what you pay for every AI action and what your client will be invoiced. But no platform can tell you if your offer is priced too low for the market, or if a client’s scope will burn more tokens than expected. It remains on the agency to monitor, test, and course-correct both costs and value delivered.
AI cost tracking agencies invest in is only as useful as the agency’s own discipline about linking every tracked cost to a billable result. A dashboard that stops at cost will let margin slip away in the background. The right process, and a tool that shows cost and earnings side-by-side, are now table stakes for serious work. Studios who treat margin as the signal and cost as just one side of it will win the transition—especially as AI expands the billable toolkit with new models and methods every month.