Ghost Seats Are Costing You 35% of Your AI Budget

The most expensive AI seat in your organization isn't the one your heaviest user is burning through. It's the one nobody is using at all — billed at full rate every month, generating zero value.

The industry calls this a Ghost seat. We define it precisely: a provisioned AI licence with token consumption below 5% of plan capacity over a rolling 30-day period.

The industry average: 35% of provisioned AI seats qualify as Ghost or severely Underutilised. For a 100-seat Claude deployment at $100 per seat per month, that's $3,500 per month — $42,000 per year — generating no measurable output.

How does this happen?

IT provisions seats for everyone during an AI rollout. Adoption is uneven — some teams embrace the tool immediately, others ignore it. The heavy users consume 80-90% of their plan capacity and occasionally hit limits. The light users barely touch their allocation. Both bill at the same rate.

The problem compounds with model tier mismatches. A seat provisioned on Claude Max 5× ($100/month, 250M tokens) for a user who runs five casual conversations a week is consuming perhaps 2-3M tokens — less than 1.5% of their plan capacity. The same work costs $20/month on Pro. The delta is $80/month per mismatched seat. Across a 20-person team where 12 users are similarly mismatched: $1,920/month recoverable with zero capability impact.

The classification framework we use:

Ghost: under 5% of plan capacity. Immediate downgrade or deprovision candidate.

Underutilised: 5-20% of plan capacity. Review within 30 days.

Normal: 20-80% of plan capacity. Monitor quarterly.

Power User: above 80% of plan capacity. Upgrade risk — monitor for limit hits.

At-Risk: Approaching plan ceiling. Proactive upgrade conversation needed.

The key insight: rightsizing is not about taking capability away from your team. It's about matching the plan to the actual usage pattern. A Ghost seat moved from Max 5× to Pro loses nothing — because it wasn't using the Max capacity to begin with.

The question to ask your IT team this week: How many of our AI seats consumed less than 20% of their plan capacity last month? If you can't answer that question, you don't have the visibility to manage the spend.