AI cost visibility: what the KPMG survey shows.
KPMG found 53 percent of large enterprises have AI agents deployed and 26 percent can see what their AI costs in real time. What that gap means, and how to close it.
- Topic
- AI cost
- Product
- Novie Control
- Format
- Essay
- Reading time
- 5 min
- Published
- 8 October 2026
- Updated
- 29 October 2026
In KPMG's Q2 2026 AI Quarterly Pulse Survey, 53 percent of large US enterprises had AI agents deployed and 26 percent had full, real-time visibility into what their AI systems cost to run. The distance between those two numbers is where AI budgets drift. Closing that distance starts with knowing which of the two groups your own organisation sits in.
What the survey found.
KPMG surveyed 204 US C-suite leaders at companies with more than $1 billion in revenue and published the results on 24 June 2026. Four of its numbers belong side by side.
- 53 percent have AI agents deployed.
- 18 percent are orchestrating multiple agents across workflows, double the previous quarter.
- 36 percent have direct token or usage controls in place.
- 26 percent have full, real-time visibility into what their AI costs to operate.
The same companies plan to spend an average of $202 million on AI over the following twelve months.
Read in order, the numbers describe companies that have moved faster on deployment than on the instruments around it. Agents are in production at half of these firms. The ability to see what those agents cost, at the moment they cost it, sits at a quarter.
TakeawayHalf of large US enterprises run agents and a quarter can see what those agents cost as they run.
What real-time AI cost and spend visibility means.
Real-time AI cost visibility is the ability to see the cost of each model call as it runs, attributed to the team, project and workflow that caused it, without waiting for a provider invoice.
The invoice arrives monthly and bills the credential. A shared API key serving five teams produces one total and hides who spent it. Real-time visibility moves the record from the invoice to the call, so the cost is known while the work is still happening.
TakeawayThe record moves from the monthly invoice to the call, so cost is known while the work is still happening.
Why agents widen the gap.
A chat assistant costs roughly what a person types into it. An agent decides for itself how many calls to make. It plans, retries, calls a tool, and asks a second model to check the first. One request from a user can become dozens of model calls across more than one provider, and the cost of a single business outcome ends up spread across invoices that never meet.
That is why the 18 percent figure matters. Multi-agent orchestration doubled in one quarter. Every agent that hands work to another agent multiplies the calls, and with them the spend that a monthly invoice reports only as a total.
TakeawayEvery agent that hands work to another multiplies the calls, and a monthly invoice reports only the total.
Planning $202 million with a quarter of the view.
At an average of $202 million in planned AI spend, timing becomes the whole question. An agent that starts looping on the first of the month keeps running until someone reads the bill. With cost recorded at the call, the same loop shows up on the first afternoon, against a named team, and a cap stops it before the money moves.
Controls are slightly ahead of visibility in the survey: 36 percent have direct token or usage controls, 26 percent see cost in real time. A cap works best when the people setting it can see what it protects. Live cost data turns a budget from an estimate into a limit someone can defend.
TakeawayCost recorded at the call turns a budget from an estimate into a limit a cap can hold.
How to close the gap: five steps.
- Map every route AI spend takes into the organisation. Provider APIs, cloud marketplaces, AI features switched on inside software you already license, tools on team cards. Ask engineering, operations and finance separately; the differences between their lists are the finding.
- Give every call an owner. Either one credential per team, or a layer that stamps team, project and workflow on each request before it reaches the provider.
- Record cost at the call. Model, tokens, cost, team, project and workflow, written at the moment the call runs. This is the ledger; every report afterwards is a query against it.
- Put caps where the call happens. Set them at organisation, team and project level, and route each increase to a named person.
- Test it with one question. What did a given team spend on AI last week, and on what? Aim to answer it in minutes, while the meeting is still running.
TakeawayMap the spend routes, give every call an owner, record cost at the call and put caps where the call happens.
What to look for in AI cost visibility tools.
- Every provider in one view. Model APIs, cloud platforms and seat products such as coding tools, in one ledger.
- Cost at the call. Tagged to a team, a project and an agent within minutes of the call.
- Your providers' own data brought in. Most now report usage by project, key and model, which is a check on the ledger.
- Native limits beside the spend. Which accounts have a hard stop, and which only send an alert.
- Caps and approvals in the same place as the view, so what you see can be acted on.
- A report finance can read without asking engineering.
Our comparison of AI spending limits shows what each provider lets you cap natively, which is where most visibility work starts.
TakeawayA visibility tool earns its place when it shows every provider in one ledger, at the call, beside the limits that act on it.
What changes when you can see it.
With cost visible at the call, the AI budget becomes something a company can plan against. Finance writes next year's line from this year's ledger. The teams building agents can scale the ones that earn their cost, with the evidence already in hand when they ask for more.
The companies in KPMG's 26 percent are showing the rest where this goes. Deployment is already moving. Visibility is the part each organisation can build this quarter.
TakeawayWith cost visible at the call, finance plans next year from this year's ledger and scales the agents that earn their cost.
How Novie Control does it.
Novie ControlAvailable now
Full cost visibility.
See spend by team, workload and model, before the invoice arrives.
$5,440
spent this cycle.
Example figure.