The problem: 18 AI services, one unexplained bill
One of our software companies runs a platform that helps galleries and other arts organizations manage calls for entry, while giving artists a place to find opportunities and submit their work. Behind the platform, 18 AI-powered services handled tasks such as researching prospects, validating payment receipts, and cleaning up listings.
Those services connected to AI providers through 11 API keys. Each service had been set up separately and recorded its usage differently. Some logs showed token usage, while others stored the information in a different format.
As a result, the monthly bill showed how much the company had spent on AI, but not which service had spent it. The team could not easily compare the cost of different features, identify unusually expensive activity, or determine whether a service was producing enough value to justify its cost.
Answering a simple question such as "What did the prospect research service cost last week?" required an engineer to search through separate system logs and narrow the activity down to a few possible services. Most of the time, the team did not investigate because finding the answer could take an entire afternoon.
The lack of visibility became a serious business problem when the company noticed a series of unexpected AI charges.
"The charges were coming through in the middle of the night, when nobody was using the application. We knew that an automated process was running, but we could not see which service was responsible."
Partner, Third South Solutions
To fix this situation, an engineer eventually traced the charges to a process that had become stuck in a loop and was repeatedly sending the same request. Overall, this lack of visibility cost the company:
- 2 hours of engineering time spent searching through logs
- 3 additional charges generated during the investigation
- Nearly $400 spent without producing anything useful
The investigation exposed a larger business problem. As the company expanded its use of AI, a single total on the monthly bill was no longer enough. The team needed a more granular view of its AI spending, with costs broken down by service and request, so it could investigate unusual activity faster and make better decisions about its AI budget.
To address this problem, we applied the same focused methodology we use for Third South Solutions projects, turning fragmented AI usage data into a clear, actionable view of where its money was going.
The solution: Labeling every AI request with the service that sent it
Following the incident, we focused on one clear business requirement: the team needed a reliable way to connect every AI cost to the service that created it.
The goal was not necessarily to reduce the company's total AI spending. It was to make that spending clear enough for the team to understand its bill, investigate unusual activity, and determine whether each AI-powered service was producing the expected return.
That requirement shaped the architecture of the solution. Most AI cost analytics tools explain spending after it has occurred, while the company needed to see which service was responsible during an active cost spike. Also, a full observability layer or separate proxy would have added unnecessary infrastructure.
For that reason, we chose a shared gateway that could route the company's AI requests through one place and connect each request to the service that sent it.
The final setup included a custom module that labeled each request passing through the gateway with the name of the service that sent it, along with other request details. This created one consistent record of the company's AI activity, even though the 18 services had previously used different API keys and logging methods.
Once we had chosen the gateway approach, the project moved quickly: we built the first working version in 1 day, then spent the next 10 days testing and refining the setup as it was rolled out across all 18 AI services.
This gives the company one place to see its AI spending:
- By service
- By individual request
- By token
The rollout did not require the company to rewrite its prompts, change which AI models it used, or rebuild how individual services handled errors. The services kept working as before, and only the way their requests were sent and recorded changed.
Once the rollout was complete, the company continued to maintain the gateway internally. The dashboard supports granular access permissions, although the cost data is available to the broader team. The gateway doesn't send automatic alerts, so someone still has to notice when spending rises, but once they do, the team can quickly trace the source and act before the cost continues to grow.
The result: The next AI cost spike was traced in 2 minutes
When the next unexpected cost increase appeared, the team opened the dashboard, filtered the activity by service, and identified the source in 2 minutes.
The same investigation that previously took 2 hours of searching through separate logs became a quick dashboard check, as every cost could now be connected to the service, request, and tokens behind it.
The dashboard also shows how costs are spread across individual sessions, so unusually expensive activity stands out.
"It's impossible to know exactly how much we saved by catching the spike early, but if it had kept running for a couple more days, it would have used up our entire monthly AI budget."
Partner, Third South Solutions
The initial incident prompted the project, but the everyday visibility became more valuable. The team can now compare the cost of different AI-powered features, investigate expensive requests, and determine whether a smaller, less expensive model could perform the same task.
This information also supports budget planning. The company can set spending caps and decide when to increase its overall budget as AI use grows, rather than reacting each time credits run low across different providers.
This reflects Third South Solutions' approach to AI: start with a clearly defined business problem, build a focused solution with a measurable result, and make sure someone owns it after launch.
Your AI bill should show what each tool costs
If your team runs several AI tools and the bill only shows a total, we can help you connect each charge to the service that created it.