Client story Alltold Scales Inclusive Media Measurement Model Months Ahead of Schedule With Vertex AI

Business challenge

Alltold needed a way to measure representation and bias in media at scale. Traditionally, measuring bias in content — like ads or movies — relied on a slow, expensive, and manual process. People would watch the content, pause it often, write notes in a spreadsheet, and then start playing the content again. This manual method presented three major problems:

• Long waits for results: Projects often took seven months to finish, making the results too dated to use for quick business decisions. “These types of analytics aren’t very actionable,” says Morgan Gregory, CEO at Alltold. “If seven months have already passed, you can’t be very responsive to market conditions.”

• Limited growth: The manual process only worked for short-form content like ads. Analyzing long-form content, such as a 90-minute movie or a TV show, would take too much time for a human annotation team, holding the company back from working with film studios and streaming platforms.

• Too much work: While Alltold first used a human team to create a crucial training dataset, they knew that to grow, they had to automate the work. Manually training and evaluating AI models without a pipeline also created a very burdensome process of trying to manage data versions and track accuracy in documents.

Not building AI models posed the biggest risk

Alltold understood that advanced AI models were the key to solving this challenge. They needed to boost their operational capabilities to rapidly develop, deploy, and test multiple models. “AI models are so core to what we’re doing, the biggest risk would be not building them,” says Kree Cole-McLaughlin, CTO at Alltold. “There are open-source projects that do very coarse gender and age classification. We evaluated them, but they’re not very good or usable.”

They just needed the right infrastructure/pipeline to build their core models and scale their business quickly. “The longer it took to create these models, the more time it would take until we could work with companies in TV, film, and generative AI,” says Gregory.

Solution

With their background in working on Google Cloud infrastructure, Alltold also wanted to build their AI/ML models on Google Cloud. To move quickly and build the necessary AI infrastructure, Alltold worked with Insight a multiple-time Google Cloud Partner of the Year. Working with Insight was strategic. By collaborating with Insight on the foundational work, Alltold’s team could focus on their core expertise: unique AI models and training data.

Their goal was to create a minimum viable product (MVP), which meant building the smallest, most essential version of the platform. Insight designed and built a fully automated, end-to-end AI/ML pipeline using Google Cloud’s Vertex AI platform and integrated it with their existing Google Cloud services like Cloud Storage and Dataflow as well as third-party services. This pipeline let Alltold overcome their challenges rapidly developing, deploying, and testing multiple models in conjunction with following MLOps best practices to manage the pipeline.

Key solution components

• Vertex AI pipeline: This was the most important part of the project: the MLOps-enabled pipeline that automatically manages every step, including pulling the data, creating the data set, starting model training, logging results, and deploying the model. Modular design makes it simple to swap in new models or add different evaluation steps later. 

• AutoML for the baseline model: The team chose Vertex AI AutoML to train the first model, focusing on the “person’s age” attribute. This gave Alltold the best model architecture for the data and a dependable, high-quality performance baseline to measure future, more complex models against.

• Expert help: Insight’s specialized knowledge dramatically sped up the work. “Insight’s AI experts, who know how to organize code in Vertex AI, helped Alltold avoid months of work,” says Cole-McLaughlin.

• Discovery: The initial discovery phase helped both teams stay laser-focused on quickly building out the capabilities Alltold needed. “The discovery phase was invaluable and set us up for success,” says Cole-McLauglin. 

Collaborating for project success

In the development of the pipeline, Insight data engineers received constant input from the Alltold team as they integrated the existing inference and data preparation code bases. This collaborative style was key to the project’s success. 

“Insight was really an extension of the Alltold team,” says Gregory. “The weekly meetings Insight held with Alltold made the project handoff a non-event because the teams were so integrated and worked together so closely.”

Impact

The new AI pipeline removed the huge time and cost constraints of manual work, completely changing Alltold’s business model and growth possibilities. The collaboration not only solved the immediate technical challenge but also secured Alltold’s future ability to innovate quickly.

Quantifiable benefits

Insight’s collaboration delivered significant, quantifiable benefits for Alltold:

• Major time savings: The time required for infrastructure deployment dropped dramatically from an undefined number of months, if Alltold had attempted it on their own, with many unknowns, to a measurable seven weeks for the initial MVP. This efficiency was largely due to implementation of a fully automated, end-to-end Vertex AI pipeline, replacing the previous manual model training and evaluation process. 

• Scalability: The pipeline gave Alltold the scalable laboratory they needed for training AI models and tracking of all of their experiments. “The new pipeline enables us to pull data sets, slice and augment them in different ways, and figure out what’s going to work the best,” says Cole-McLaughlin.

• New markets: Insight helped Alltold expand their addressable market opportunity that had been limited to short-form media like ads to the far greater potential of long-form media such as film, TV, and generative AI content.

• Transformed the ability to experiment: The automated pipeline transformed Alltold’s ability to experiment with AI models from being impossible to possible. “In quick succession, we trained all the AI models we could,” says Gregory. “For gender expression, age, skin tone, and body size, with this pipeline in place, we now have fully operational production models that work at scale. For us, that’s huge.”

Qualitative improvements

The most important improvements centered on speed, confidence, and internal focus:

• Faster growth: The successful AI pipeline allowed Alltold to begin having conversations with executives at streaming platforms and film studios, a market they couldn’t enter before. It also let them evaluate content from the quickly growing generative AI space.

• Confidence in the future: Insight’s upfront scoping and clear, tight development schedule gave Alltold a high degree of certainty, which was crucial for them. “Just having clarity and confidence in the timeline let us plan for when we’d complete the project, which was very helpful,” says Gregory.

• Team focus and efficiency: Pipeline automation cut down on the need for human annotators for those four key identity attributes. Most importantly, it freed the technical team from the heavy burden of manual model training, evaluation, and version control, letting Alltold repeat model training to find the best models.

• Excellent documentation: The project gave Alltold a set of high-quality, comprehensive technical design documents. “The quality of Insight’s documentation made the pipeline so easy to use and a no-brainer for me to pick up and start working with,” says Cole-McLaughlin. 

Future outlook

The Vertex AI pipeline acts as a modular foundation for Alltold’s AI future. By combining automation, reproducibility, and scalability, the MLOps-enabled pipeline transformed Alltold’s ability to develop, test, and operationalize multiple models rapidly — creating a strong foundation for data-driven decision-making and AI innovation. 

The company can now easily test new model types or change parts of the process, such as data sampling or evaluation metrics. This sets them up to build on top of Insight’s foundational work to achieve a more scalable environment.

The success of the project confirmed Alltold’s belief in Insight’s collaborative approach for AI infrastructure development. “I wouldn’t hesitate to recommend Insight to another AI-native company,” says Gregory. “We found working with Insight to be extremely valuable.”