Key takeaways
- Most companies now pay for several AI assistants at once, because teams add whatever works and rarely turn the old one off.
- Every AI assistant bills differently, from seats and credits to spend limits and per-person licenses, so no single admin console shows total assistant spend by team.
- Tracking AI assistant costs means bringing GitHub Copilot, Claude, ChatGPT Enterprise and Microsoft 365 Copilot spend into one view by person and team, next to the rest of your AI, cloud and GPU spend.
- With seat activity next to spend, IT can reclaim idle seats before renewal and finance can split assistant cost with the same rules it uses for cloud.
Best for
- FinOps leaders
- Finance leaders
- IT leaders
- Teams running AI assistants
How finance and IT can keep up with AI assistant spend while teams keep adding new tools
Standardizing on a single AI assistant would be too easy. Reality is most companies now pay for several AI assistants at once. That means disparate bills, budgeting, forecasting, and more. Tracking their cost requires bringing GitHub Copilot, Claude, ChatGPT Enterprise and Microsoft 365 Copilot spend into one view by person and team.
Picking a favorite AI coding assistant (also referred to as SDLC or ADLC tools) is a little like picking a favorite kid. You’re not supposed to, and the answer changes depending on the week.
That’s the AI assistant horse race. Every few weeks there’s a new tool or release worth trying: Copilot, then Claude Code, then Cowork, then the latest ChatGPT. Your teams try them, as they should. They add whatever works for their job, switch when something better shows up, and rarely turn the old one off.
So most companies end up running several at once. In September, 43.8% of businesses on Ramp’s platform paid Anthropic and 40.6% paid OpenAI, and Ramp notes many pay both (Ramp AI Index, October 2026). LayerX puts the average enterprise user at 2.24 AI apps (LayerX, State of AI Usage 2026).
Mixed AI stacks are here to stay. That’s good news for the people doing the work. It’s harder on whoever owns the bill.
Why is AI assistant cost so hard to track?
Every assistant charges differently. GitHub Copilot mixes seats with usage-based credits and overage. Claude sets spend limits per person. ChatGPT Enterprise runs on credits or tokens, depending on your contract. Microsoft 365 Copilot is a per-person license with on-demand usage on top. Each admin console shows its own slice, in its own units.
So when finance asks what you spend on AI assistants and which teams drive it, the honest answer usually starts with “give me a few days.” Someone exports four reports and matches names across them in a spreadsheet. Then renewal comes around with a second question nobody can answer quickly: how many of these seats are people actually using?
How do you see AI assistant spend by person and team?
Mavvrik puts AI assistant spend in one place, next to the rest of your AI, cloud and GPU spend. We’re spotlighting four of the most common assistants here, with the essential signals you want access to. These signals impact business decisions, everything from product pricing to margin management.
- GitHub Copilot: cost and usage by person, team and cost center, overage included, plus cost by model.
- Claude: Claude Enterprise spend and spend-limit use per person, plus Claude Code and Cowork spend by person and session, down to the tokens and tool calls in each session.
- ChatGPT Enterprise: cost by person, day, model and product, including Codex by type of work. Cost follows your contract: your credit price when you enter one, OpenAI’s estimate otherwise.
- Microsoft 365 Copilot: license cost and on-demand usage by person and department.
Each vendor stays the billing source. Mavvrik shows you who’s using what (individuals or agents, teams, departments, etc.), how it’s allocated, and how close each person is to their limit.
How do you find unused AI assistant seats and split the cost?
- Catch unused seats before renewal. For Copilot and Claude, assigned, active and idle seats sit right next to spend, so IT can reclaim seats or move people to a smaller plan before the renewal count is set.
- Split the cost the way you split everything else. If engineering pays for Copilot and Claude Code and sales pays for Microsoft 365 Copilot, the same rules you use for cloud cost can split assistant cost by team, department or product. More on that in Chargeback vs. Showback.
- Find the heavy users. When a team keeps hitting its spend limit, you can see whether it’s a few people or the whole group, and adjust limits with them.
- Budget by team. Finance gets assistant spend by department in one table for 2027 planning.
Is it secure?
Every connection is read-only and set up once by an admin, with nothing to install and no per-person setup. History comes in on day one (12 months for Copilot and Microsoft 365 Copilot, full history for ChatGPT Enterprise) and data refreshes daily. Mavvrik reads billing, usage and license data. For ChatGPT Enterprise, that means no conversations, prompts or files.
How do AI assistants fit into total AI cost?
Assistant seats are often the most visible AI line item. The same teams also drive model API spend, agents, and the cloud, data and GPU capacity underneath them, across public cloud, private AI and on-prem. When assistant spend follows the same allocation as everything else, finance gets one answer to what AI costs each team, whichever assistant is winning that week.
Which AI assistants does Mavvrik track?
We’re spotlighting GitHub Copilot, Claude (including Claude Code and Cowork), ChatGPT Enterprise and Microsoft 365 Copilot here. Coverage keeps growing across AI assistants, coding tools and agent platforms with each monthly release.
Can I see unused AI assistant seats?
Yes, for GitHub Copilot and Claude, right next to spend.
Does Mavvrik replace vendor billing?
No. Each vendor stays the billing source. Mavvrik adds who’s using what, allocation and reporting across all of them.
