How do I cap AI spend in a CRM?
Set a soft alert at 80% and a hard stop at 100% of a monthly allowance. Cap the AI only, so manual work never stops. Meter usage per feature and per agent, so the cap is explainable, not only enforced.
The short answer
- Set two thresholds on a monthly AI allowance. A soft alert around 80% gives whoever owns the budget time to react. A hard stop at 100% prevents further spend rather than only warning about it.
- The hard stop should apply to AI features only, never to the CRM as a whole. AI drafting failing gracefully at the cap is a reasonable trade-off. A rep unable to open a record or send a hand-written email because AI usage hit a ceiling is not.
- Meter usage per feature and per agent, not only as one combined total. When the cap is approaching, whoever owns it can see which feature or agent is driving the spend and make an informed call about what to trim.
- A cap without per-feature visibility only tells you spending stopped. It does not tell you why. The next decision, adjust the allowance, cut a feature, or tighten an agent's guardrails, becomes a guess instead of a choice.
Why the AI-only distinction matters more than the threshold numbers
It is easy to focus on getting 80% and 100% exactly right and miss the more important design decision underneath. What stops when the cap is hit? A CRM that ties its core functionality to the same allowance as its AI features creates a scenario where hitting an AI usage ceiling takes down record access, manual sending or basic CRM operations along with it. That turns a cost control into an outage. It punishes the team for exactly the situation the cap was supposed to handle gracefully.
The fix is architectural, not only a threshold choice. Separate the AI allowance from the product's core operation entirely. Hitting the cap then degrades or disables AI drafting, summarizing and scoring, while everything else keeps working as before. Viewing records. Writing and sending an email by hand. Updating a deal stage.
Cap types and what happens at each
| Cap type | Behaviour | Who is notified |
|---|---|---|
| Soft alert at 80% | Nothing stops; a warning is raised with usage remaining | The account owner or whoever manages the AI allowance |
| Hard stop at 100%, AI only | Features that rely on AI pause or fall back to a manual path | The account owner, with a visible notice in the product |
| Per-feature metering | Tracks which feature (drafts, summaries, scoring) drove the usage | Available to the account owner for review, not a separate alert |
| Per-agent metering | Tracks which agent's runs consumed the allowance | Available to the account owner for review, not a separate alert |
Making the cap explainable, not only enforced
A hard stop that arrives with no context, "AI usage limit reached", leaves the account owner guessing at what to do next. Per-feature and per-agent metering turns that into an actionable moment. If one agent's runs are consuming most of the allowance, that agent's guardrails or schedule can be tightened specifically. The whole team does not lose AI drafting because of one high-volume automation.
Disclosure: SalesCrew is our product, and its AI caps work this way. A soft alert at 80% and a hard stop at 100% of a plan's monthly allowance, for AI usage only, never for manual CRM operation. Every model call is logged with the feature and the agent that triggered it. An account owner approaching the cap can see exactly what to adjust rather than shutting AI off entirely while investigating.
A hard stop on the whole product is a hostage situation
Questions
- What should happen when the hard stop is hit?
- Features that rely on AI should stop or fall back to a manual path. The CRM itself, manual data entry, sending an email by hand, viewing records, should keep working without interruption. A hard stop that takes down the whole product is a design mistake, not a spend control.
- Is 80% and 100% the right threshold for every team?
- It is a reasonable starting point, not a fixed rule. A team with unpredictable usage spikes might want an earlier soft alert, like 60%, to leave more room to react. The specific numbers matter less than having both a warning and a hard limit rather than one or the other.
- Who should get notified at each threshold?
- Whoever owns the budget or the AI rollout, typically an admin or the person managing the subscription. Not the whole team. A soft alert at 80% is a heads-up to that owner. A hard stop is worth a more visible notice, because it changes what people can do that day.