A budget review has a fixed date on the calendar. An OpenAI or Copilot bill doesn't. Token usage shifts by the week, seat counts creep up quietly, and none of it pauses to wait for the meeting where someone has to explain what changed and why.
This isn't a general argument for why AI cost governance matters. That case is made in full here. It's a specific sequence for finding real reductions in the time before a specific meeting, in the order that gets results fastest.
Start with which model is doing the work, not how much it's used
The instinct before a budget review is to go looking for waste: idle seats, forgotten pilot projects, accounts nobody logs into anymore. That instinct isn't wrong. It's just not where the fastest money usually is.
Model choice is. Most AI tools default to whichever model is most capable, not whichever model the task actually needs. A support ticket summary and a legal contract review don't need the same model, but they're often billed at the same rate because nobody went back and checked. A workload that started on a frontier model in early 2026, when that was the only option worth using, is frequently still running on it now, even though a cheaper model would return the same quality of answer for routine tasks.
Checking this takes an afternoon, not a quarter. Pull usage logs by task type for the last 60 to 90 days and flag anything routine, summaries, simple lookups, first-draft replies, still running on the most expensive available model. That gap is usually visible immediately, and closing it changes the very next invoice, not the one after.
Find the seats and subscriptions nobody's using
Once model choice is checked, the next fastest win is what's already paid for and sitting idle. Not a company-wide audit. A short, specific list: Copilot seats assigned to people who left the team, ChatGPT or Claude accounts opened for a pilot that ended months ago, and API keys still active from a project that shipped and moved on.
A second list catches what usually gets missed, because no single item on it looks expensive by itself. Multiple teams each paying for their own separate AI subscription instead of one consolidated account. A developer tool with a Copilot seat bundled in that nobody uses because the team switched editors. Trial-tier API access left running past the trial, quietly converting to paid usage. Individually, each one is a rounding error. Together, they're not. Left unreviewed for six months across a growing AI stack, they add up to a number worth checking before the review, not after.
None of this requires a full IT audit. It requires pulling the seat list and the API key list side by side against who's actually logging in or making calls, and flagging anything that hasn't been touched in 30 days. A Copilot seat with zero activity for two months isn't an edge case. It's a subscription nobody's looked at since it was assigned.
Get your AI vendor spend benchmarked before review
CostRoom compares your OpenAI, Copilot, and AI vendor spend against structured market data, not list price.
Check pricing against market, not just against last quarter
Model right-sizing and seat cleanup both answer "is the usage efficient." Neither answers "is the price fair." That's a different question, and it's usually the one nobody in the building is positioned to answer.
An engineering or IT team that manages AI tools well can absolutely execute the first two steps above. What a technical team, however capable, doesn't typically have is live market intelligence: current benchmark pricing on enterprise agreements, seat bundling terms, and volume commitments, informed by structured relationships across a large number of vendors at once. That's not a gap in the team. The team isn't the issue. It's a gap in the system behind the team, and it's the same distinction whether the tool in question is OpenAI, Copilot, or anything else on the technology bill.
This is where a sourcing and vendor network earns its place in the process. CostRoom's network covers more than 100 partners globally across SaaS, cloud, and AI, and it exists specifically to bring benchmark pricing into a negotiation that would otherwise run against list price as the only available reference point. It carries no OEM agreements, no commission arrangements, and no bias toward any vendor in it. The recommendation reflects the client's own data and current market conditions, nothing else. No exceptions. How this network works in full is covered here.
Benchmarking without a negotiation behind it is just a number on a slide. Negotiation without benchmarking is guesswork dressed up as a strategy. The budget review needs both, and most companies walk in with neither.
What to bring into the budget review
Most AI spend conversations stall on one thing: nobody can say precisely where the company actually sits. A useful way to check, borrowed from how CostRoom assesses new engagements, is four levels. Visibility: can you see what's being spent, on which model, by which team. Usage tracking: can you see which tools are actually being used, and how often. Procurement management: is there a process for approving new AI tools and seats. Optimization: is spend actively being reduced, benchmarked against market, and renegotiated on a schedule, not just watched.
Most companies walking into a budget review have Level 1, sometimes Level 2. Rarely more. Level 3 is the one almost everyone assumes they already have, and almost nobody actually does, since AI tools tend to get approved informally, one manager at a time, rather than through a real process. Very few have Level 4, and that's not a criticism of anyone in the room. It's just where most companies' AI spend naturally settles without a structured process layered on top.
What actually lands well in a budget review isn't a narrative about all of this. It's a specific list. Which models got switched for routine tasks, and what that changed starting next cycle. Which seats or subscriptions got cut, and the dollar impact. Where current AI vendor pricing sits against market benchmarks, and what's being renegotiated as a result. And which level the company is actually operating at today, stated plainly, with what moving to the next level would require.
That last point tends to be the most useful thing in the room. It reframes the conversation from "did we save money this quarter" to "where does this process actually stand," which is a harder question to answer with a spreadsheet alone.
None of the three steps above require waiting for a bigger initiative to get approved first.
Start with an AI spend analysis
CostRoom maps your full AI vendor footprint before any optimization work begins.
Frequently asked questions
How can I reduce my OpenAI or Copilot bill before the next budget review?
Start with model right-sizing, checking whether routine tasks are running on the most expensive available model by default. Then clean up idle seats and forgotten subscriptions. Both show up in the next billing cycle. Benchmarking current pricing against the market is the slower but larger lever, and it's the one most companies skip.
What's the fastest way to lower an AI vendor bill?
Checking which model is handling routine, low-complexity tasks. A workload running on a frontier model by default, when a cheaper model would return the same quality of answer, is one of the most common sources of avoidable AI spend, and fixing it doesn't require switching providers.
Do I need an engineering team to reduce AI costs?
For model right-sizing and seat cleanup, yes, someone technical needs to execute the change. For knowing whether the price itself is fair, no. That requires market benchmark data across vendors, which is a different resource than engineering time, and most technical teams don't have it on hand regardless of how capable they are.
What is a sourcing and vendor network, and how does it help with AI costs?
A sourcing and vendor network is a set of established relationships across many vendors that brings current market pricing into a negotiation. CostRoom's covers more than 100 partners globally, carries no OEM agreements or commission arrangements, and recommends only based on client data and market fit.
What should I present at a budget review about AI spend?
A specific list: which models were switched for routine tasks and the dollar impact starting next cycle, which seats or subscriptions were cut, how current AI vendor pricing compares to market benchmarks, and which of the four stages (visibility, usage tracking, procurement management, optimization) the company is actually operating at today.
Related reading:
