Spend Limits and Individual Controls
Why limits matter
Without spend controls, a small number of power users can consume a disproportionate share of your contract budget. A single engineer running large code-generation tasks or a researcher synthesising long documents can generate ten times the spend of an average user. Without visibility and limits, this is invisible until the invoice arrives.
Spend limits give you a cost governance layer: you set a ceiling for a group or individual, and when they approach or hit it, Claude stops processing requests until the next period begins or you raise the limit.
Two levels of control
Claude Enterprise provides spend controls at two levels:
1. Group-level spend limits
Groups (configured in Admin Console → Groups) can be assigned a monthly spend cap. This is the most common control used by organisations.
Example:
- Engineering group: £2,000/month
- Marketing group: £500/month
- Legal group: £300/month
- Leadership group: no limit
When the Engineering group's cumulative spend for the calendar month reaches £2,000, members of that group receive a message in Claude that their group has reached its limit. They cannot send further messages until the limit resets (start of next month) or an admin raises the cap.
Setting a group spend limit:
- Admin Console → Groups
- Select the group to edit
- In the group settings, find Spend limit
- Enter the monthly amount in your contract currency
- Save
2. Individual (per-seat) spend limits
Some Enterprise contracts allow spend limits at the individual user level. This is less common in practice — most organisations manage at the group level — but useful in specific scenarios:
- A small group of very high-volume users (e.g., a research team doing intensive synthesis work) where you want per-person accountability
- Pilot programmes where you are testing how much different roles actually use Claude before setting broader limits
- Contractor or temporary staff who should have tighter limits than permanent employees
Setting per-user limits (where available):
- Admin Console → Members
- Click the three-dot menu next to the user
- Select Edit member or Spend settings
- Set monthly limit
- Save
Note: individual limits may not be available on all Enterprise contract tiers. If you do not see this option, contact your Anthropic account manager.
Budget alerts
Rather than waiting for a group to hit its limit and lose access, set alerts at a threshold below the limit. This gives the group (and you) a warning.
Common alert thresholds:
- 60% — early warning, time to investigate which use cases are driving high spend
- 80% — action threshold — decide whether to raise the limit or ask the group to moderate usage for the rest of the month
- 95% — near-limit — prepare for potential disruption; have the raise-limit process ready
Setting alerts: Admin Console → Usage → Budget settings → Add alert → enter threshold percentage → assign notification recipient (typically the admin and the group's department head).
Alert notifications are sent by email to the designated recipients.
What happens when a limit is reached
When a group or user hits their spend limit:
- Claude returns a message to the user explaining that their group/account has reached its monthly limit.
- New conversations cannot be started; in-progress conversations may also be blocked.
- The limit resets automatically at the start of the next calendar month (not 30 days from now — the calendar month).
- An admin can manually raise the limit at any time to restore access before the reset.
There is no grace period and no partial access. It is binary: under the limit, full access; at the limit, no access.
This is intentional — it prevents overspend — but it means a group hitting its limit mid-month loses access until either the month resets or an admin intervenes. Plan your limits conservatively in the first few months until you have reliable spend data.
The right way to set limits
Don't guess at limits. The right approach:
Phase 1 (months 1-3): observe
Run without limits initially, or with very high limits that are effectively uncapped. Use the Usage Dashboard to gather real spend data by group.
Phase 2 (months 3-4): analyse
Look at the Usage Dashboard (M65-L1) and identify:
- Average monthly spend per active user by group
- P90 spend (the 90th percentile user in each group)
- Peak months vs. quiet months
Phase 3: set limits with headroom
Set limits at 130–150% of your observed average monthly spend for each group. This gives enough headroom that normal usage variation does not trigger the limit while still providing a ceiling against unusual spikes.
Example: Marketing group has averaged £320/month over 3 months with a peak of £420. Set the limit at £600 (roughly 1.5× the peak). This accommodates a busy campaign month without being uncapped.
Phase 4: review quarterly
Revisit limits every quarter as usage patterns change. A team that starts using Claude for a new intensive use case (e.g., Marketing starts drafting long-form SEO content) will see spend rise; adjust limits before they hit a wall.
Communicating limits to your team
Nothing frustrates employees faster than Claude going silent mid-month with no explanation. Before you set limits:
- Tell each group what their limit is and why.
- Explain what happens when the limit is reached.
- Give them a process to request a limit increase (typically: message the group's admin or IT).
- Publish the limits in your internal AI usage policy.
A user who knows their group has a £500/month limit and sees it is running low will moderate their own usage. A user who has no idea limits exist and suddenly cannot use Claude will escalate to IT as an outage.
Priority users: exemptions and higher limits
Some users legitimately need higher limits than their group default. Examples:
- A product manager running competitive intelligence research monthly
- A data analyst synthesising long reports
- A department head who also uses Claude for strategy work outside their group's core function
Handle these by:
- Creating a sub-group or separate group for high-volume users with a higher limit, or
- Assigning individual user limits (where available) above the group default, or
- Moving the user to a group with a higher limit if their role justifies it
Do not simply remove all limits for these users. If you have 47 users with no limit and 3 with a limit, the 47 uncapped users represent uncapped cost risk.
Monitoring spend in real time
The Usage Dashboard (Admin Console → Usage) shows:
- Current period spend by group — see which groups are tracking toward their limits
- Individual spend leaderboard — top 10 spenders for the month
- Spend trend — daily/weekly trajectory; useful for projecting end-of-month spend
Check this weekly, not just when something goes wrong. A group tracking at 110% of last month's spend in week two is going to hit a limit in week four — you have two weeks to raise the limit or investigate the cause.
Scenario: Finance team mid-month shutdown
A financial services firm sets the Finance group limit at £400/month based on early usage data. In March, the Finance team is preparing for year-end and the FD asks the team to use Claude intensively for commentary drafting across 12 business units. By March 19th, the group hits £400 and loses access — 12 days before month end.
The team raises an urgent ticket. The admin logs into the Admin Console, raises the Finance group limit to £800, and access is restored within minutes.
Lessons:
- Year-end was a predictable spike — the Finance group limit should have been raised proactively.
- A 60% alert (£240) and 80% alert (£320) would have flagged the pace of spend by March 10th.
- Quarterly limit reviews would have caught that March = year-end months ago.
Key takeaway
Set limits based on observed data, not guesses. Run without limits initially to gather a baseline. Set limits at 150% of peak observed spend. Set alerts at 60% and 80%. Communicate limits to your teams before they take effect. Review quarterly. The goal is cost governance, not rationing — limits should be high enough that normal usage never hits them, but low enough to catch genuine anomalies.
📖 Official Documentation See this in practice in Anthropic's live support docs: