Crafting Custom AI Solutions: A Success Story

In today's dynamic digital marketplace, the ability to create custom AI solutions is crucial. This necessity becomes increasingly clear in light of the major challenges outlined in our Blog 1, which underscore the cost and complexity often associated with developing such bespoke solutions. Reflecting these broader industry challenges, we present a case study demonstrating our journey in tailoring an AI model to meet the unique needs of a high-profile client.

The Challenge

This client needed a custom Video KPI solution, reflecting an unmet need in the market. Both the availability of data and the inherent problem complexity, key challenges we discussed in Blog 1, became forefront issues. Advertisers had predominantly optimized towards Completion Rate or CpCV, and the client's unique requirement presented a scalability problem, mostly because of limited support for Vpaid or advanced Vast formats by most premium video inventories. The task for our team was to build a scalable, multi-advertiser solution that was both cost-effective and quick to implement.

Our Approach

Our approach to tackling this intricate challenge blended the principles of quick consensus and solution reuse, effectively mirroring the MLOps and CI/CD methodologies discussed in Blog 2 and Blog 3, respectively.

Quick Consensus

Effective communication was crucial for consensus and project acceleration:

1. Our team initiated the project by presenting three intuitive custom KPI examples to provide the client's team with a clearer understanding of potential solutions.
2. Instead of detailed written proposals, clients described their KPI in Python pseudo-code, offering precise, code-based clarity.
3. Live working sessions replaced lengthy email threads, improving iteration speed through immediate feedback and timely decision-making.

Solution Reuse

Our second strategy involved using existing solutions to streamline development:

1. By delving into the impression data and investigating the KPI definition, we were able to ascertain that a custom algorithm would produce adequate signals using volume control alone, guiding spending towards the high-performing inventories with respect to the custom KPI.
2. This breakthrough allowed us to fully reuse a Volume control-based SPO pipeline and associated algorithms already developed for previous projects, saving us time and cost.

Both these strategies, quick consensus and solution reuse, successfully addressed the significant cost factors and challenges introduced in Blog 1, primarily by ensuring maximized use of available data and lowering the problem complexity.

Results:

Our end result was a custom AI solution that significantly achieved and surpassed the client's initial goals, underscoring the power of custom AI solutions, when combined with efficient MLOps and CI/CD practices.

1. Trial campaigns saw a 2x increase in the Custom KPI Score in test ad groups compared to control groups.
2. Satisfied with the results, the client prepared a branded case study, applied the solution to other campaigns, and requested a similar solution for another KPI.
3. Our dedication and performance strengthened our relationship with the client, leading to increased spends and greater trust in our services. The success story even attracted interest from other markets and global entities.

Our commitment to automation and efficiency, emphasized in Blogs 2 and 3, greatly contributed to the success of the custom AI model project. By harnessing automated processes and consistently efficient workflows, we were able to eliminate potential errors and redundancies, accelerating the project's overall momentum and delivering the solution in an optimal timeframe.

Moreover, the success of this project extended beyond mere execution. The results cultivated stronger client relationships, expanded potential markets, and validated the use of our MLOps and CI/CD methodologies in delivering high-impact results. This further amplifies the role these principles can play in overcoming the key challenges faced in machine learning projects, as outlined in Blog 1.

In conclusion, our approach to building a bespoke AI model involved understanding a client's unique needs and rapidly developing a custom, scalable solution that not only met but exceeded expectations. This case study illustrates our agility, vision, and determination, reinforcing our reputation as a reliable partner in delivering high-quality, tailor-made AI models.