Knowing what you spent last month is useful. Knowing what you're on track to spend this month — by group, before the billing cycle closes — is what actually changes behavior.
Spend targets are LLM Cost Tracker's answer to that gap. You set a monthly dollar target per group, and the dashboard shows you where each group stands today: how much it's spent month-to-date, what it's projected to hit by end of month at the current run rate, and how far off the target it is. No alerts required, no threshold to configure — just a clear view of whether you're on track.
The problem with monthly totals
Most LLM cost visibility stops at the total: here's what you spent this month. That number is useful for invoicing and for spotting that something went wrong. It's not useful for managing spend proactively.
The issue is timing. By the time you see that your summarize group spent $800 this month instead of $400, you're already halfway through the next cycle. You can investigate, find the cause, and fix it — but the $400 overage is already billed.
What you actually need is to know on May 12th that your summarize group has spent $220 in 12 days and is on track to hit $550 by month end, against a target of $400. That's actionable. You still have time to do something about it.
How spend targets work
Spend targets live in the dashboard under a dedicated page. For each group tag you're tracking, you can set a monthly dollar target. The dashboard then shows three numbers per group:
The projection is intentionally simple — a linear extrapolation from your current run rate. It doesn't try to model seasonality or usage patterns. That simplicity is a group: it's easy to reason about and easy to sanity-check. If your summarize group spent $220 in 12 days and there are 31 days in the month, the projection is $220 ÷ 12 × 31 = $568. No black box.
Groups without a target set still appear in the list ranked by MTD spend — so you always have a full picture of where your budget is going, even for groups you haven't gotten around to targeting yet.
Setting targets that are actually useful
The most common mistake when setting spend targets is anchoring to last month's spend. If your search group spent $180 last month, setting a $200 target feels safe — but it doesn't tell you whether $180 was reasonable in the first place.
A more useful approach is to work backwards from your unit economics:
- Start with your group's value. If search is gating a $29/month subscription, how much is it reasonable to spend per user per month on inference? If you have 200 Pro subscribers using search, a $400/month target means $2 per user — probably fine. $1,200 means $6 per user, which starts to compress margin meaningfully.
- Set targets at the group level, not the total. A single $1,000/month total target tells you nothing useful when you have four groups. Four targets of $250 each tell you exactly which one is out of line.
- Treat the first month's target as a baseline, not a goal.Set targets based on your current run rate, let a month run, then tighten. The goal is visibility first, optimization second.
What to do when you're tracking over target
Seeing a group projected to overshoot its target is the start of an investigation, not the end. A few common causes and what to look for:
Token count increased. Check the average tokens per call for that group over the past few days vs. the previous period. If input tokens went up, something changed in what you're sending — a larger context, a longer system prompt, more retrieved documents. The call log will show individual calls; sort by input tokens to find the outliers.
Call volume increased. Sometimes spend goes up because usage went up, which is good news. Check total calls per day for the group. If volume grew proportionally to spend, your cost per call is stable — the target just needs revisiting to reflect the new usage level.
Model changed. If a deploy switched a group from a cheaper model to a more expensive one — even as a fallback or retry path — cost per call will jump. Filter the call log by group and look at the model column for any new entries.
Failed or retried calls. A retry storm on timeouts can double your call count without doubling your useful output. Look for calls with zero or near-zero output tokens — those are the wasted ones.
Spend targets vs. spend enforcement
Spend targets and spend enforcement solve different problems and work at different layers.
Spend targets are a planning and monitoring tool. They live in the dashboard, they're set by you for your own visibility, and they don't affect how calls are handled. A group tracking 140% of its target will still make API calls normally — the target is informational.
Spend enforcement is an operational control. It's configured in the SDK, it runs at call time, and it can block or warn before a call is made. Enforcement is what you use when you need a hard guarantee — a free-tier user cannot spend more than $5 this month, full stop.
The two work well together: use targets to monitor your own group spend and spot trends early, use enforcement to put hard limits on end-user spend that you're responsible for. Most teams start with targets, then add enforcement for the groups where overspend has a direct cost-of-goods impact.
Getting started
Spend targets are available on all plans including Free. If you're already tracking calls with the SDK, your groups will appear in the targets page automatically — no additional instrumentation needed.
- Go to Spend targets in the dashboard sidebar.
- Your groups appear ranked by month-to-date spend. Click the target field next to any group and set a monthly dollar amount.
- The projected spend and delta columns populate immediately based on your current month's data.
If you're not tracking calls yet, the quickstart walks through adding the SDK in about five minutes. Once you've tagged a few calls with a group label, they'll appear here automatically.