How OpenTeams freed a leading ad analytics firm from a black-box trap.
The firm’s business—AI-powered logo detection, brand valuation, and audience measurement—was taking off. More clients meant more to analyze, more logos to detect, and more data to crunch. But their team discovered a flaw. Their system, a third-party SaaS solution, cost a fortune, and did not give them the control they wanted.
They were trapped. Paying high fees for a “black box” system they didn’t own, couldn’t customize, and couldn’t scale. They needed a way out. They needed OpenTeams.
The client needed its own model and its own data pipeline. The OpenTeams crew dove in.
The team laid out a new blueprint, strategically blending the best of open source with the scalability of a public cloud.
The core of the system was the object detection model. Instead of building one from scratch, they chose a PyTorch Open-Detection Model. Why? Because the open-source community had already done the heavy lifting. It came with pre-trained weights that could be fine-tuned for the client’s specific needs—identifying over 2,000 distinct brand logos and 34 types of assets, from jerseys to jumbotrons. This decision alone saved months of development and leveraged the collective intelligence of thousands of developers.
The primary challenge was scale: processing 50 million distinct video frames a month without prohibitive costs. The team integrated their PyTorch model into a massively parallel AWS infrastructure, which provided virtually infinite on-demand computing power. Crucially, this architecture could also scale to zero, automatically shutting down compute resources when idle. This ensured costs were contained. The team paid only for the processing power they actively consumed.
For the database, they stuck with the trusted open-source workhorse PostgreSQL. It provided the robust, reliable data storage they needed without the licensing feeds of proprietary alternatives.
Locked into an expensive, inflexible third-party SaaS vendor.
Built a custom, in-house platform blending strategic open-source tools (PyTorch, PostgreSQL) with cloud infrastructure.
Total Ownership & Reduced Cost: The client now owns its model, data, and software stack. This provides a cheaper, more customizable, and faster system.
The client is now truly in control of their architecture. They own their own intellectual property. They can customize their models, add new brands on the fly, and tweak the logic. Best of all, they could scale. Onboarding a new client who needed to analyze thousands of hours of video was no longer a cause for panic. It was just a matter of spinning up more resources in the cloud.
Every January, the global economic conversation moves to a small alpine town in Switzerland. Heads of state, founders, and technologists gather in Davos to debate the future. This year, beneath the familiar headlines about geopolitical tension and economic uncertainty, another theme dominated nearly every private conversation: artificial intelligence.
