Vista Group have worked with data and AI for years, finding the best ways to incorporate cutting-edge technology into our workflows to provide insights to the cinema industry. Our Data Science and Engineering teams work tirelessly to create new innovative ideas, and here we dive into the creation of Oneview’s AI-generated podcast of insights.
Cinema circuits rely on monitoring key metrics to measure their daily performance, and to inform decisions to achieve the greatest efficiency in their operations. Truly actionable insights are hugely valuable, and with cinema executives needing to focus their time as effectively as possible, delivering those insights in the best way possible is key.
Oneview, Vista Group’s data visualisation app, provides cinema chain leadership with the best capabilities to stay connected and informed about their circuit’s performance.
One of our most recent innovations for the Oneview app is our AI-generated Podcast, which collates benchmarked insights on highly curated key metrics for cinema executives on the go.
Working with new and exciting technology to produce innovative ideas like this is no small feat, and our Data Science and Engineering teams worked tirelessly to bring this concept to fruition. Here’s how they crafted a fully automated production pipeline with end-to-end verification to deliver great insights in a new and innovative format.

The best of both worlds
Current AI technology is still not able to automatically derive meaningful insights by interpreting data intelligently, but Vista Group has been working with cinemas and their data for decades. As such, our Data Science team possessed a wealth of experience and understanding of the key performance metrics that reflect cinema performance.
It was essential in designing the Oneview podcast feature that we harness both our human expertise, and the strengths of Large Language Model (LLM) technology to its fullest potential.
With our Data Science team’s knowledge, we identified tried and tested metrics that belonged at the heart of this podcast report, and used the greatest strength of current-generation LLMs: their humanised interface. That is, their ability to deliver metrics and insights in a natural-language format. By combining our human expertise and the capabilities of LLMs, we harnessed the best of both worlds.
Anatomy of the Oneview Podcast
To achieve fully automated end-to-end delivery of the Oneview podcast, we divided our process into four phases:
- Daily data extraction.
- Fact generation.
- Script generation.
- Text-to-speech.
Within fact and script generation, and the text-to-speech conversion, we conducted rigorous verification to test for errors when moving between phases. Multiple languages are also supported for both text and speech, and the final two phases incorporate data verification subroutines for language translation.

Fact and Script Generation
The first phase, data extraction, aggregates key insights from Horizon, Vista Group’s business intelligence and data warehouse offering.
The most critical element to get right in this process was to ensure the numbers we reported are exactly what is received in the data feeds from our Horizon clients. So, the insights are hand-crafted and tested by our Data Science team, and supported by our engineers.
Once we have derived numerical values, the fact generation phase sees an LLM create a natural-language summary of the metrics, placing them in context with meaningful comparisons and benchmarks.
Next, the script phase collates these individual facts and composes narratives on box office, concession sales, performance by location, and upcoming movies, along with an introduction and outro. Translation is applied afterwards for supported languages, currently including Spanish, French, Dutch, and English in several regional variations.
Finally, speech synthesis is applied to the scripts to create the final podcast, such as the example below:
Welcome to the Oneview Podcast, your AI-generated, personalized cinema performance summary tailored for the executives and managers at the Oneview Demo cinema chain. In this episode, we're diving into the weekend's trading figures from March 3rd to March 5th, 2023, and comparing them to our benchmark period of February 24th to February 26th, 2023. Stay tuned for a concise analysis that will keep you informed and ahead of the curve. From March 3rd to March 5th, we saw a notable uptick in box office admissions, with a 10 percent increase from 138,000 to 152,000. Revenue also climbed by 14 percent, reaching two million six hundred thousand dollars from the previous two million two hundred ninety thousand dollars.
The importance of verification
Accuracy is paramount in every form of reporting, and the Oneview podcast was no exception. It was critical we ensure at every step that no errors or changes occurred in data between phases. If a single error was detected at any stage, the process began again from the beginning with data extraction. The loop continues for several iterations, and if a perfect result cannot be achieved, the phase ultimately alerts the Data Engineering team for further investigation.
Even at the end of each phase, facts are individually extracted from the script and compared with the original database data, closing the loop of fact generation through verification. This process has proven extremely robust, with every Oneview client’s pipeline executed daily without any uncorrected error.
Text-to-speech generation
The final stage of the process is to turn a script into a podcast: the generation of audio. We use several different technologies in this process, striking a balance of ensuring accuracy of the text-to-speech translation, quality of the audio output (specifically how ‘human’ the voice sounds), and outputting English speech in multiple accent options for our clients.
Once more, text is extracted from the audio file, and in turn reported numbers in that text is compared to the original script to verify at this last stage that no errors took place during the speech synthesis.
Language support
Vista Group operates globally, so supporting a large number of languages is essential. However, strong translation likewise relies on human expertise for accuracy compared to automated translation.
In the initial stages of developing a new language for our translation, the Data Science team works with human translators to translate example scripts and gain understanding of the grammar and sentence structures of these languages. Simple translation using an LLM often results in the LLM being unable to capture regional specifics and nuances of language, or misinterpreting local dialects and location names.
Along with translators, custom grammar rules are developed for each language. This approach has proven very successful for the Data Science team, and has seen great appreciation from our clients using Oneview.
Valuable insights from a new way to provide insights
Creating the Oneview podcast feature has proven a valuable experience in combining the best strengths of Large language Models, traditional data science, and data engineering into one seamless, fully automated production pipeline. There is a spectrum of misconceptions around the value LLMs provide—from detractors who denounce them due to hallucination issues, to those who consider LLMs a one-stop shop that will magically deliver insights and business value. Neither is true of course, and LLMs are hugely valuable when used for the purpose for which they were developed.
Working with new and exciting technology like this, supported by human expertise and verification, fuels our innovative efforts.
The Oneview podcast as seen great reception, proving that carefully curated metrics and facts developed through diligent business analysis and industry expertise are highly valuable. In fact, they are essential to delivering the very best insights that will provide tangible value to business operations around the world.