The dashboard has five main numbers and one panel that actually tells you what to do about them. Here's what each one means, and the order that gets you to something actionable fastest.
1. Total Cost, Total Requests, and Avg Cost per Request are the headline metrics — Estimated Savings is the one that tracks what you've actually recovered.
2. Cost-by-feature and cost-by-model are where the real diagnosis happens.
3. The recommendations panel is the fastest path from "here's what's happening" to "here's what to change."
The Top-Line Metrics
Total Cost current period
Your aggregate spend across every provider and model, for whatever time window you've selected. The headline number — start here to know if you're trending up or down versus the prior period.
Total Requests
Raw request volume. Useful mainly in combination with Total Cost — if cost is rising faster than request volume, something about the mix (which model, how long the prompts are) is changing, not just how much traffic you're getting.
Avg Cost per Request
Total Cost divided by Total Requests. This is the number to watch for silent regressions — a system prompt that grew, a routing rule that quietly stopped working, a cache hit rate that dropped. If this number climbs without a corresponding change in what your app is doing, something's worth investigating.
Estimated Savings the one that matters most
The running total from recommendations you've implemented — not a projection, an actual recovered amount based on what changed and what your traffic looked like before and after. This is the number that answers "is this tool paying for itself," directly.
Where the Diagnosis Happens
Cost by feature breaks spend down by whichever part of your product triggered the request — this is usually the fastest way to find where money is concentrated, because "total cost went up" is a symptom and "the summarization feature's cost went up 40%" is a diagnosis you can act on.
Cost by model shows the split across whichever models you're routing to. If a high-volume, low-complexity endpoint shows spend concentrated on your most expensive model, that's usually the single highest-leverage thing to check first — it's the pattern behind the most common waste source in production LLM traffic.
Want to see these numbers on your own traffic?
Your dashboard updates in near real time as requests come through.
Open Your DashboardHere's what each number means and where to focus.
The Recommendations Panel
This is where the dashboard stops being a report and starts being a to-do list. Each recommendation is ranked by projected dollar impact and comes with the specific pattern that triggered it — which endpoint, which model, what to change. Working through this panel top to bottom, starting with the highest-impact item, is the fastest path from looking at the dashboard to actually reducing your bill.
The Order to Check Things In
Once you've been through this sequence once, checking it again takes under two minutes — which is exactly the point of running it as a regular habit rather than a one-time audit.
Frequently Asked Questions
What does 'Estimated Savings' mean on the Preto dashboard?
Why is my Avg Cost per Request higher than I expected?
What order should I check the dashboard in?
How often does the dashboard update?
Now that you know what to look for.
Open your dashboard and run through the four-step order — Total Cost, cost by feature, cost by model, recommendations.
Open Your DashboardQuestions about a specific number? Reply to your welcome email.
