Go through the company card statement at month-end and the same lines keep showing up: Claude, ChatGPT, Claude again. The statement doesn't tell you who uses each account, or whether it's a team plan or someone's personal subscription. The dev team says they "bought one more because they kept hitting the limit", while another team paid for a seat two weeks ago and hasn't touched it. And when someone uses AI on a company PC with a personal account, you don't see it at all. 🤔

The AI menu (beta) in SELLEASE just got a major update. It pulls together the AI subscriptions used on employee PCs and your company's API key usage into four views, by account · by member · live by team · API (tokens and cost), and it can show a warning window on any PC where an unapproved AI account is used. Here's what each view shows and how to put it to work.

1. 🧾 Usage by account: buy more, or cut back?

AI › Usage by account lists every AI subscription account used in your company in one table. It currently tracks Claude and Codex/ChatGPT subscriptions separately, and for each account you see:

• Plan and monthly cost: Pro, Max, Plus, Free and so on, plus the monthly cost your company enters
• Output tokens · requests · active time: how much work the account actually did
• Current limit: how much of the 5-hour and weekly limits is used, and when they reset
• Users and last used: whether several people share one account, and when it was last used

The most useful column is the verdict. Accounts are labeled Hitting limits · Unused for a while · Room to spare, and the default sort is by action priority, so the accounts that need attention come first. 💡

• Hitting limits: often above 80% of the limit, or has reached 100%. A candidate for a higher plan.
• Unused for a while: paid for but not used. A candidate to cancel or reassign.
• On hold: when nothing has been collected for over 48 hours, the verdict is held back instead of guessed.

👀 Who actually uses a shared account

Select an account in the list to open Account detail. For a shared account, Member breakdown shows who uses it and how much (share of the account, active time, requests and output tokens), and which assigned members aren't using it at all. Usage by model shows which models the account was used with. It gives you the evidence to decide whether to split a shared account that keeps hitting its limit, or drop unused members and move to a lower plan.

2. 👥 Usage by member: who uses AI, and with which model

AI › Usage by member regroups the same data by person. You see how many members use AI (adoption rate), each member's active time, requests and output tokens, their top model, and how many subscriptions and devices they use.

It's a good answer when leadership asks to see what AI adoption is actually delivering. If one person holds several subscriptions, there may be room to consolidate. Note that usage from PCs with no assigned user isn't counted here, so assign a user to each PC in asset management first.

3. 📡 Live team usage: what AI is doing across the company right now

AI › Live team usage shows AI work collected from employee PCs in near real time (up to a 5-minute delay).

• AI work time: hours over the last 4 weeks scaled to a month on weekdays, and the daily average per PC
• AI cost: total subscription cost and cost per AI hour
• Right now: which departments are running how many AI tasks on which models, shown as a flow
• Today's flow: AI work time since the workday started, compared with the same time yesterday
• AI hours by department: daily average and total per department over the last 4 weeks

Cost per AI hour divides subscription cost by actual AI work time, so the next "should we buy more seats?" conversation can be about numbers instead of hunches. 📊

4. 🔢 API usage: tokens and cost by project and model

If internal tools or automations call the OpenAI or Claude APIs, AI › API usage shows what they cost. Once your company's API connections are registered, it reports for any period:

• Cost for period (USD) and the change versus the previous period
• Requests and the daily average
• Tokens, split into input, output and cached
• A trend chart and breakdowns by project/workspace · API key · model

When API spend suddenly jumps, find the spike in the trend chart, then narrow down the cause in the project and model tabs. The connection status also shows how many API keys are going unused.

5. 🚨 Unapproved AI account warnings: catch personal accounts

This is the one security teams will like most. When an AI account your company hasn't approved (for example, one created with a personal Gmail address) is used on a company PC, a warning window appears on that PC. You set it up under Unallowed Policy at the top right of Usage by account.

• Two warning targets: "All accounts not approved" (warns unless the account is approved, registered by the company or in an approved domain) or "Unapproved accounts only" (warns only accounts marked unapproved or in an unapproved domain)
• Domain rules: register up to 50 approved domains (such as your company email domain) and up to 50 unapproved domains. Personal email domains like gmail.com can be marked unapproved too.
• Warning interval: the same account is warned again only after 1, 4 or 24 hours, whichever you choose.
• Message added to the warning window: up to 200 characters, such as "Please sign in with your company account. Questions: IT Support".

The most common setup is to approve only your company domain and warn on everything else. Set the warning target to "All accounts not approved" and add your company email domain as an approved domain, and any AI account signed in with a non-company email gets a warning. Adding personal mail domains as unapproved domains also makes them easy to spot in the account list.

Accounts are evaluated in this order: account setting → company-registered account (approved) → domain rule → unclassified. The warning informs rather than blocks, and usage collection and reports continue as normal. The goal is to cut down on company code and documents going into personal AI accounts without interrupting work. 🔒

6. ✅ What to do in your first week

1. Check PC user assignments: member and department totals are only accurate when each PC has a user assigned in asset management.
2. Enter monthly subscription costs: cost per AI hour and total subscription cost only make sense once each account's monthly cost is entered.
3. Add approved domains → watch for two weeks → turn on warnings: see how many accounts come up as unapproved before switching warnings on. It causes less confusion.
4. Review at month-end: move Hitting limits accounts to a higher plan, and treat Unused for a while accounts as candidates to cancel or reassign.

AI subscription data is collected by the SELLEASE agent installed on each PC, so PCs with the agent are the ones counted. For a broader software review, see the SaaS and software license checklist before IT budgeting, and for agent-based IT asset tracking overall, see SELLEASE IT Asset Management. 📌

💬 FAQ

Which AI services are tracked?

Subscription accounts are currently tracked separately for Claude and Codex/ChatGPT. API usage is based on the OpenAI and Claude API connections your company registers.

Does using an unapproved account get blocked?

No. A warning window appears on that PC to let the user know. Usage itself isn't blocked, and usage collection and reports continue as normal.

Do you collect what employees type into AI (prompts)?

These views show usage and cost: accounts, plans, tokens, requests and active time. If you'd like details on what's collected, we'll walk you through it during an enterprise consultation.

🌱 Talk about AI spend in numbers

The next time someone asks for one more AI subscription, you can answer with that account's limit usage and cost per AI hour. To see your own company's AI usage, send us an Enterprise inquiry. We'll cover agent rollout and security policy too. 🚀