As I write this, Analytica 7.1 has recently entered Beta Testing. We are all finding that the new and improved Assista is not only incredibly good, but it is fundamentally transforming how we use and interact with Analytica, as I think you’ll see here.
At the same time, World Cup soccer is in full swing, with excitement permeating everywhere! And this morning, these intersected for me in a very fun session that I’ve been encouraged to share.
A colleague, Jorge Muro-Arbulú, sent me an email last night as I was leaving work about a Kaggle data set with the performance metrics on over 28 thousand players. He asked, is Messi really the extreme outlier in this data that he appears to be? He attached the Kaggle data set as a *.csv file to his email.
I dragged the attachment from the email onto the diagram of a new, blank model in Analytica, which gave me an instant table of data in Analytica, and then I rapidly typed the following prompt without much thought:
Within a few minutes, Assista had built a model, with every variable fully defined and documented with descriptions, had done an okay job at laying out the diagram (with some room for improvement), and had provided the sigma rankings of interest for Lionel Messi.
Zero model-building expertise required, yet the transparency of a model and ability to interact with it. My next prompt was:
I show her response next in text (the screenshot was too big for an image here. She indeed explains what I requested of her, what she did, the results, and an interpretation. I copy-pasted her text into an email response back to Jorge.
Claro — aquí tienes un texto en español listo para copiar y pegar:
Lonnie me pidió que revisara y organizara este análisis en Analytica, así que aquí va un resumen de lo que hice yo, Assista, y de los hallazgos.
Tomé el módulo de análisis de sigma por jugador y lo reorganicé para que el flujo quedara más claro: primero los datos base, luego las métricas por jugador, después los promedios y desviaciones estándar del grupo de comparación, y finalmente los resultados del jugador seleccionado. Con eso quedó más fácil verificar qué estaba calculando el modelo.
Después confirmé el cálculo para Lionel Messi comparándolo contra la posición seleccionada, que en este caso es CM. El modelo calcula un z-score (o número de desviaciones estándar respecto al promedio del grupo posicional) para cada métrica. Los resultados fueron:
- Gls_90: 7.06 sigma
- Ast_90: 4.36 sigma
- xG_90: 6.54 sigma
- PrgP: 3.34 sigma
En otras palabras, según este modelo, Messi aparece como un outlier extremo frente al grupo de mediocampistas centrales en esas métricas, especialmente en goles por 90 y xG por 90. Un valor de 7 sigma es extraordinariamente raro, así que esto sugiere una de dos cosas: o Messi está realmente lejísimos del promedio del grupo, o la dispersión del grupo de comparación es muy pequeña y eso está inflando los z-scores.
Mi lectura es que el hallazgo es interesante y probablemente va en la misma línea de lo que tú viste, pero vale la pena revisar también el contexto estadístico: tamaño de muestra, filtro de minutos, definición exacta del grupo posicional y si comparar a Messi como “CM” es la referencia más adecuada.
Si quieres, también te hago una versión más corta y más natural, tipo mensaje de WhatsApp para mandárselo a tu colega.
Jorge urged me to share the experience in a blog posting.
If you are used to building models, I hope this gives you a taste of how the new Assista really does transform the experience. The advantages of having an interactive, visual, transparent and flexible model are still here, beyond the simple conversation in a pure chatbot interface. The agentic benefits of having AI carry out tasks for you that you get with Claude Code and Codex are present, but with the added benefit of having these within the Analytica modeling environment. And the ability to have an AI that can interpret, refine, document, and organize your models interactively without requiring a huge time expenditure on your part, nor substantial model-building expertise or Analytica experience is extremely refreshing.
Download Assista’s model: soccer_analysis.ana
Related article: World Cup Soccer. How much does randomness determine the winner?
