For the past two decades we’ve seen an enormous growth in the data space. As data moved to the cloud, platforms like Google, Amazon, Microsoft surged, and companies like Snowflake and Databricks built multi-billion dollar businesses in a short amount of time. Venture capital dollars were poured into the "modern data stack", building a new generation of tooling that enabled data to flow from various systems to a central data warehouse, be cleaned and modeled and delivered in business intelligence interfaces.
However, the modern data stack is already in for a serious overhaul. It was developed around the perception that the final end-user is someone sitting behind a computer screen, clicking around in data tools and graphical presentations of insights. That is no longer the case, as AI is increasingly becoming a major consumer of the data stack.
The person now working in business intelligence tools can finally get full insights, conclusions and suggested actions from interacting with AI. No more clicking on filter after filter or creating a dashboard for top management. AI assistants are quickly becoming the predominant end-user of data, and that changes how we should build data platforms and products.
Redefining data sources
Even today most businesses are used to thinking of tools like CRM and ERP systems as the main data sources for their data infrastructure. The mantra is to extract all the raw data from these sources and load it into a central data warehouse. But with AI models and their ability to interpret any piece of information you throw at it, we need to rethink what our source data is.
SaaS systems like CRMs are essentially built to ensure a company gets standardized and structured data. However, with the current AI models, you can already skip many of these steps and get the required insights from scattered and unstructured data, without the strict workflow regimes of a typical SaaS product. Look at what Klarna is doing here for example.
LLMs are already really good at processing text, tables, images and even video into reliable information. And, in organizations, far more valuable data is stored in the many Word, Powerpoint, Excel and PDF files that teams produce and edit every single day, than in the structured data in CRM, ERP and other systems. Combine semantic search with LLMs and you get rich context about customers, projects and solutions that traditional data sources miss. Add the growing ability to interpret almost any kind of information, and it is a strong combination.
A new way of data modeling
Until now, data modeling was all about transforming raw data into clean, structured, and readable formats, typically for human consumption. Today, many AI use cases can bypass this step, as models like LLMs have different requirements for data preparation, focusing on structuring data so AI can access and interpret it instead of formatting it for the user's consumption.
Data modeling now faces different challenges. The work is finding the best way to extract as much detail as possible from documents using LLMs, and that turns out to have little in common with writing database queries to reshape data and run calculations on it. Companies with the best performing AI systems understand this well, and have built entirely new teams for it.
Business Intelligence will no longer be about dashboards and charts
For years, business analysts have built Business Intelligence through dashboards using tools like Tableau, Power BI, and Looker, relying on filters and drill-downs to create charts and tables.
AI now allows business users to ask questions in natural language, getting answers from all accessible documents in the organization, and even requesting charts with a simple query. Business Intelligence tools as we know them will change drastically in the next few years, as AI models become multi-modal and enable interaction through voice or messaging, and the traditional filters fall away.
This doesn’t mean that BI will disappear, as the daily insights will still remain relevant and user behavior isn’t changed overnight, but over time, more and more analysis will happen on the fly without the need for an analyst in between.
Data has a new end-user, and that’s very exciting
AI is rapidly becoming the primary consumer of organizational data. That changes how we process, analyze and present data, and how we get to insights in the first place. We think it is a very exciting shift.
At Mozaik we are already building data platforms for our customers that look nothing like the ones we would have built two years ago, and we keep finding new things data can do for us along the way.
