There’s a lot of buzz in the media now about the data privacy concerns associated with using LLM chatbots. How did you address these issues when building your GenBI features?
All that buzz is based on highly valid concerns. People have good reasons to recoil from the idea of sharing their extremely sensitive corporate data with OpenAI’s engines and other similar solutions. At the same time, they want to tap into the power of AI to get answers to their data-based questions.
At Pyramid Analytics, we reconcile the two motivations by turning the whole issue upside down. We’ve structured things in such a way that the AI model can never actually access your data. Basically, once you send your prompt to Pyramid, we send a data-agnostic version of it on to the large language model (LLM), together with a basic description of the data, and receive in exchange a clear direction of the best way to ask your question in the form of a prompt for Pyramid to run internally on your actual data.
We like to think of this as sending the LLM the ingredients for analysis, and receiving a recipe in exchange. We’re the ones who apply the recipe to your dataset in situ. Your data never leaves your server or cloud, either to go to our platform or to an LLM. We serve as the interface between your data and the LLM, providing you with insights, analysis, and/or the best ways to visualize and explore your data.
Before natural language AI prompts came into the picture, was true “self-service business intelligence” an unfulfilled promise?
I have to admit that self-service BI never really delivered on its promise. All the same data-related problems that we were grappling with in the 1990s, like “garbage in garbage out,” are still with us, but they metastasized because the amount of data ballooned so rapidly. What’s more, the idea of self-service created new problems, like multiple versions of the truth, because now everyone has their own copy of the data.
The main problem is that people don’t think in pie charts or scatter plots. They don’t actually know what visualizations or formats to apply to their data, but self-service BI only works if you have some level of expertise in data manipulation and exploration. People really just want a system that delivers answers to their questions.
That’s where Generative BI, or GenBI, like what we’ve developed for Pyramid, comes into the picture. It allows people just to ask a question in natural language, and receive a dashboard or newly sliced chart that provides the answers you seek. With GenBI, BI is now truly self-service, because anyone can ask a question in natural language, without any expertise in data exploration, and receive the insights they need in under a minute.
You talk a lot about the power of data to drive better decision making. What do you think holds business leaders back from making the leap from using data in a descriptive manner to using data in a prescriptive manner?
First, the insights extracted from data should be available at any time, as this is key for tactical decisions. Second, I believe that AI is the technology that gets us to prescriptive use cases.
I think that we all appreciate the power of data to enable better decision-making that’s based on solid facts and evidence. The trouble has always been accessing the insights locked within the data, and in a timely manner. Until now, business leaders didn’t always have the expertise to analyze data and produce meaningful insights, and if they had the tools and the capabilities, they couldn’t get those insights in time to make tactical decisions at the speed of business.
We’ve already come a long way towards bridging the gap to prescriptive data management by simply delivering actionable answers to business questions, in under a minute. We’re bringing the same kind of automated magic to corporate data that Google Maps brought to journey-planning. In the same way that you enter your destination into your navigational app and follow the instructions, we’re hoping that GenBI will soon be able to provide strategic advice based on data trends and projections.
What are some of the coolest ways you think app developers can implement embedded GenBI into user-facing experiences?
The coolest ways probably haven’t even been imagined yet. The sky’s the limit.
But so far, we’re looking at use cases like a banking app inviting customers to ask “How can I save more money?” A financial advisor might deliver automated smart suggestions about how clients can optimize their portfolios. Or manufacturing factory management apps might provide generative, interactive supply chain efficiency insights.
Other BI platforms claim to support natural language queries. What makes your approach different?
For starters, Pyramid can take even very vague questions and turn them into workable data queries that deliver meaningful insights. Most BI platforms won’t be able to handle something as imprecise as “analyze sales in the Northeast of the US over the past year” and then break that all down in visualizations that show per-branch category leaders and trends over time. Pyramid also supports spoken requests, not just textual inputs.
Another significant difference lies in the results you’ll receive. Current BI platforms produce only basic pie charts from simple datasets. Pyramid serves up highly complex visualizations and a whole range of graphs, charts, reports, and formats.
What’s more, all our results are fully dynamic, so you can manipulate them to zoom in on specific subsets of data, adjust the colors, and more. Most apps that promise natural language data management are based on Python. They produce a static image that can’t be tweaked or explored any further. All of our visualizations can be adjusted and sliced however you like, using follow-up natural language prompts, just like a conversation.
Where do you see GenBI’s capabilities improving in the next year?
We’re definitely focused on taking GenBI to its next stages. I can’t elaborate at this point too much, but you can expect more surprises and disruptive technologies from Pyramid. I can tell you that we are addressing the next major obstacle for GenBI tools: scalable personalization.
People want to be able to submit very specific questions about personalized entities, but LLMs require full and detailed knowledge of the data involved in order to deliver a meaningful response, which, as we already explained, isn’t practical for privacy reasons, but it also isn’t practical because these are often huge data sets that change dramatically throughout the day, every day. We’re exploring ways to get over that hump of situation-specific data investigation, and we’re very close to success.
At the same time, we’re working on enabling users to apply their own formulae and select their own mathematical interpretations of the data. This is challenging because it’s so difficult to understand the user’s fundamental concern and push forward the right syntax. Enacting a formula, once you’ve understood which one to use, is not particularly difficult. We think that solving these issues will bring us across the next frontier in BI and data management.
