Business analytics is mostly broken.
Traditionally, insights derived from data were delivered via business portals and dashboards. Data science teams developed dashboards in response to requests from business leaders. Since very few business leaders understood data science, and very few data scientists understood business, the dashboards were never perfect and often outdated.
With the emergence of Gen AI, the situation is dramatically changing for the better. To reflect this shift, we introduce a new term – Generative Analytics – a different approach to how we interact with data, making it more conversational, dynamic, and accessible.
Democratization is the most important benefit of the new approach.
Consider a simple question that a marketing manager might ask: “What would happen to our customer engagement if we increased our ad spend by 15% in Q4?”
With traditional analytics, the marketing manager would need to engage the data science team, submit a request to a queue and wait for the turn. A data scientist would then
- Interview the marketing manager to understand the question
- Identify the relevant data set and understand the data
- Select a model type that can answer the question based on the data available
- Prepare the data: cleaning, augmentation, feature engineering
- Train and validate a model on historical data
- Evaluate the model on the most recent data
- Visualize the results and present them to the marketing manager
The entire process may take days or even weeks. In the end, the marketing manager may discover that the real question they wanted to ask was “How can we increase our customer engagement by 15%?”, and we are back to square one.
These days, a large language model equipped with access to data and a Python sandbox can go through all the steps in minutes. Business leaders can state their problems in plain English and collaborate with AI in real time to get the desired insights and forecasts.
Generative Analytics also greatly expands the range of data available for analysis.
Business analytics is mainly limited to structured historical data: sales numbers, website visits, etc. Extracting features from unstructured data, such as earnings reports, product reviews, audio recordings, and images, requires complex data preparation and modeling.
With modern multi-modal large language models, these capabilities come “out of the box” and can be used in many scenarios with little to no customization.
For this to work, the user interface presented by a Generative Analytics solution must allow self-service multi-modal data intake, augmentation, transformation, and analysis.
It turns out that the good old spreadsheet paradigm works surprisingly well. Instead of numbers, strings, and formulas, the cells in a multi-modal spreadsheet can contain chunks of text, images, documents, videos, and audio recordings. They can also contain prompts and calls to AI agents transforming, enriching, and augmenting data.
At AnyQuest, we coined a new term for such multi-modal spreadsheets driven by prompts, large language models, and AI agents – Smart Datasets.
Users can upload raw datasets and define prompt or agent columns to extract, enrich, and transform multimodal data. Agents can use internal and external APIs - CRM, web search, semantic search - to enrich data. AnyQuest takes care of scheduling and running thousands of jobs in parallel to perform the calculations and update the datasets.
What does this mean for your business?
Companies that embrace Generative Analytics will dramatically increase the pace of innovation and the quality of decision-making. They can free up their data scientists for strategic projects and address their busy work with business user self-service and AI automation.
Achieving this will require some training for the business users, mostly in prompt engineering and agentic automation.
Companies will also have to enhance their data and business analytics foundation with Generative Analytics capabilities: data fabric, Python code sandboxes, chatbots, multi-modal spreadsheets, and AI agents.