What GenAI Did to Data Analytics with Yael Lev at AIE 2024

Good morning and a afternoon, everyone, wherever you might be today. I'll share how this technology is not just enhancing analytics, but revolutionizing it, making the enterprise not just intelligent intelligence, but let's say artificially in intelligence, am I'm ya, Lev, the AI and data science product leader at Sisense. I started off as a computational neuroscientist and later spent many years as a data scientist and leading data science teams.

Today I'm a product manager at Sisense. Sisense is an analytical platform specializing in embedded analytics. My focus is on advancing the integration of AI, machine learning and advanced statistic tools within our, within our analytical solutions, making complex data easier to handle and more accessible and actionable for businesses, businesses of all sizes.

Imagine a world where decision making is as easy as asking a question in plain English. No need to dig through databases, no complex queries required. No need to reach out to an analyst to Fitch or update the latest report.

Just ask for the information and trust the results. Imagine, imagine that the answers not only draw on existing data, imagine the answers. Also, also integrate newly generated insights and data sources from your com company's unstructured data and the world, and are based on optimization algorithms, um, run on the fly to enrich your results.

This is not a setting for a futuristic novel. This is the business, um, reality where stepping into today, thanks to generative ai since the release of chat g PT in late 2022, we've seen a significant expansion of the variety of generative AI applications. Countless innovation solutions surfaced across different sectors.

Tailoring AI to meet unique industry needs. Jasper acts as a co-pilot for marketing teams assisting in content creation and strategy. Amazon Q developer is designed to boost developer productivity by automating coding tasks.

TAMA supports marketing and sales teams by helping and them craft compelling presentations quickly and efficiently. These are just a few example virtual. Every, virtually every major industry has now developed its own specialized AI assistant.

From enhancing customer interactions with AI powered chatbots to assisting doctors diagnosing diseases, AI is proving to be a powerful enhancer of productivity. This is merely the starting point. Given the rapid pace of development in new AI applications, we can anticipate even more profound changes in the coming years.

This slide presents a vivid map that highlights the rapid development and release of larger language models or LLMs. These advanced AI systems are designed to understand, generate and interpret human language models by open ai either are the most famous, including the latest GPT-4 oh that just came out. They've played a pivotal role in powering platforms like chat, GPT and Microsoft Co-pilots.

In 2022, we observed a significant increase in LLM development peak peaking with the release of chat GPT. This momentum continued into 2023 with the introduction of other LLMs lama by Meta Palm Bard, and Gemini by Google, Jurassic by I 21 Labs Titan by Amazon Rock by Xai. Each major cloud provider is now either hosting these models or providing frameworks for their training.

This highlights the intense competitiveness in this com. In this market, the latest breakthrough by OpenAI, GPT-4 oh is a multimodal model, meaning it can understand and generate not only text, but also images, audio and video, but it runs that in real time. It makes the AI aware of your surrounding and provides high level insights information based on that.

Looking into the future, the pursuit of artificial general intelligence or a GI represents the next big thing in AI research and we're all excited to see what that brings. A GI, um, aims to create systems that have the ability to understand, learn, and apply intelligence across a broad range range of, of tasks mimicking human cognitive abilities. Leading organ organizations like Open ai, deep Mind, and ARO Anthropic, um, are leading these race to a GI and striving to achieve what could truly be a transformational breakthrough in artificial intelligence.

As we witnessed the rapid advancements in AI and get closer to the realization of artificial general intelligence, the impact on digital transformation is massive Across every industry. Companies are accelerating their innovation efforts propelled by the dynamics of today's market. Every organization is becoming a software company.

Creating and consuming data organizations reco, uh, recognize more than ever the value of personalized data-driven decision making, the potential of data to drive monetization usage and market innovation is being reimagined and it doesn't stop there. The way we create software itself is undergoing a sh a fundamental shift. Sat Satya Nadela has highlighted a platform shift of a magnitude similar to, to the transition from desktop to MO mobile computing.

The sh the shift is now towards AI driven tools that significantly lower the barriers of entry for software development, enabling a broader range of people to build and innovate. As companies digitize more rapidly than ever before, the critical role of data escalates and the methodol methodologies of software development evolve to meet this new digital era. We're not just using ai, we're integrating it deeply into every fabric of, um, digital transformation.

Let's dive into a fun example with the open AI data analyst. GPTI uploaded my data, or you can even ask it to simulate it, and I gave it a short explanation and what I wanted to do and how I wanted the results to be presented. This AI data analysts assist with the data exploration provides both data visualizations, textual explanations, and the underlying code necessary to generate it.

This is beyond just generating text and images. The AI is generating Python code that gets executed in real time in case the code breaks. It will try to fix itself based on the errors and repeat the analysis.

I have to pause here. This is a glimpse to the future where AI is autonomous fixing itself, adjusting its response planning, what it needs to do, what tools it needs to perform the task improving with self feedback. These type of feedback loops are what will take us to the next level from copilots to automation.

The data in the chart and the chart itself aren't just generated images like you would get from ge generative models like Dolly. The numbers there are calculated by a program and you can read the Python, um, you can drive, uh, dive into the code for an extra and very important layer of explainability. This is a huge productivity boost, but it still requires a human in the loop there to evaluate the analysis and navigate the exploration.

Nevertheless, this achievement by OpenAI poses a significant transformation in the analytics domain and completely changes traditional analytics. Traditional analytics typically typically follows a business intelligence use case. bi it's built by BI developers requires a company to have a dedicated BI team with experienced BI developers, data engineers, and analysts to get all the data in one place, process it, clean it, and create analytics reports and dashboards.

The primary insights generated through traditional analytics are mostly descriptive, aggregated data presented in charts. BI tools allow limited self-service exploration by allowing users to drill down into additional data dimensions. If you're lucky, you might get a glimpse of pre-pro pre-processed predictive tools like churn analysis, lifetime value predictions, forecasting, all of which require data science expertise to set up users of traditional analytics include both the creators, the BI developers and business analysts and the end users who are usually business unit owners.

These users interact with the analytics through BI tools, which can range from sophisticated platforms to Excel files the most. LI Excel is the most widely used BI tool out there. Insights are usually delivered to the end user users with emailed reports or interactive dashboards, users can further slice, slice, and dice their pre-made analytics limiting them to only ask about what was prepared in advance by the BI team.

This approach has been the cornerstone cornerstone of data-driven decision making for many, many years. But lately this is, this is just feeling outdated. Users want more.

They wanna self-serve and not wait for the analytics to be created for them. They just wanna ask for it. They want access to the information beyond the dashboards in context of their daily daily workflows.

If it's Slack, an application they're using, asking Alexa or chat pt, they wanna consume the numbers from the comfort, comfort of their work environment. Traditional analytics come with several significant challenges that makes it complex to implement data silos. Merging data from diverse sources often presents significant difficulties limiting the ability to gain comprehensive insights across domains.

Traditional insights struggle to adapt to growing volumes, variety and velocity of data. As data complexity increase increases, these system systems become less efficient and more cumbersome to manage lengthy data processing cycles and high infrastructure costs weigh heavily on the organization's agility to adopt analytics. The reliance on specialists for data operations create bottlenecks.

Dependence on the BI team reduces accessibility and slows down the decision making process. End users from all kinds often have problems in interpreting the data. Data literacy.

Not everyone is an expert and they need support to understand what the data means. Traditional reports and dashboards often lack interactivity, preventing users from performing on the fly analysis and deeper data exploration. When they hit a dead end and the data isn't available, they need to wait sometimes months for data to be prepared and accessible for them.

Each one of these challenges presents an opportunity for AI to transform analytics and significantly enhance the user experience. AI can make data more accessible throughout the data stack from data connect connections through data modeling, all the way to actionable insights. Next, I wanna deep dive into data maturity.

You might be familiar with this curve. Do I have the expertise in my organization to be data-driven? What value can we get from the data?

How hard is it to get there? Understanding the data maturity level is crucial to understand the value value data analytics can bring to an organization. Data maturity is analytics journey.

It can be broken down into distinct stages, each representing a higher level of complex complexity and technological maturity. Each stage brings with its increased value and intelligence. The base level is descriptive analytics.

What happened, A table chart or textual explanation explaining the historic data. In hindsight, it usually involves generating simple reports and visualizations and is the foundation of traditional analytics. Next level is about diagnostics.

I wanna start asking why did it happen? Why did my revenue drop? Why didn't I meet my sales goal last quarter?

Why are the hospital's beds in full capacity? Why is there an increase in support tickets? I wanna understand the underlying reasons for past events and starting and, and start uncovering hidden insights from my data.

Next in line is predictive analytics. What will happen? What will my sales next, uh, be?

Next, uh, quarter. What will the availability be in the hotel next weekend? What is the likelihood for a user to turn this?

This stage involves using historical data and statistical models, uh, to predict future outcomes Techniques such as regression analysis, machine learning and forecasting models are used. This usually requires specialized predictive analytics software, machine learning frameworks and data scientists. As, as you can see, we're going up in the value and intelligence AEs, but the complex, the complexity to get these insights is going up as well.

Predictive analytics is not accessible to most organizations. Data scientists and ML infrastructure are expensive and require a lot of expertise. Lastly, the holy grail of the most advanced stage of the analytics maturity, prescriptive analytics.

What should I do? These questions are with of much higher value. What should I do to be profitable?

What should I do to be more efficient? What should I do to lower my costs? Prescriptive analytics is all about recommended actions.

It's not only predicting future outcomes, but also suggests actions to achieve desired results. This involves optimization, algorithms, simulations, and complex event processing. Something that requires a lot of work.

Today, very few organizations leverage automated prescriptive analytics. It is usually up to the decision maker, business strategist and operations manager to come up with a course of action to achieve the goal. But analytics maturity is now closer than ever and is being accelerated by ai.

Just as I was rehearsing my session, OpenAI announced an another upcoming improvement, um, to their offering, uh, an a new AI data analyst, no setup. Just connect to your Google Drive or OneDrive and ask analytical questions. Get insights directly on your data with your visualization tools.

Wow, what, what a productivity boost for anyone using those um, tools, bringing state-of-the-art algorithms and analysis to the fingertips of anyone with a chat GPT subscription or who knows, maybe by the time you watch this presentation, it'll be open for everyone. Gen AI increases the insights, complexity, boosts productivity and value increases the explainability of the result with textual descriptions and visualizations. All of that wa value while lowering the required expertise level to get to those insights.

Another interesting use case in the data maturity journey is easily dealing with complex uh, data. Complex data has large val volumes, velocity, and variety of data. AI not only generates the insights but can sift through extremely massive amounts of, uh, insights and surface the most interesting ones.

The, um, process can even be customized by specific user need or business context. Increasing the relevancy and value of the insights natural language interfaces make it this super simple to give a set of instructions, some background information about the business context and, um, those easily provide the basis for personalized insights and recommendations. Finally, in the analytics maturity journey, there is an additional dimension that generative AI uncovered.

Explainability. Generative AI shines in any task that has to do with text generation summarization and key insight extraction. It has a critical role in improving data literacy, empowering users to understand the meaning of the insights and understand how it can help them or how they can make use of it.

This is a huge value for companies because it lowers the expertise levels of the user. The username may not be familiar, familiar with Excel or histograms, not to mention ML based predictive analytics. Adding to all the value AI assistance introduces a new kind of analytics conversational analytics.

Not only can the AI run the analysis, but it can hold your hand while you're doing it and help you explore your data. Generative AI truly caused an accelerated evolution in analytics and it's exciting to think what's around the corner next quarter, next year in five years. It is accelerating the analytics journey from hindsight to insights to foresight, from dashboards, to chat bots, and in context analytics.

It plays a role in transforming where analytics is con consumed by who and the value it brings. It sets the stage for a higher level insights and recommendation that support decision making and makes it easy to self-serve data exploration with conversational analytics. On the right, you can see a simple illustration encapsulating all of that.

This is a conversation between a shop owner and its AI analytic assistant. The assistant has access to their inventory management system, sales transactions, marketing promotions, and a loyalty program data. The shop owner can easily ask for insights on our data without any additional preparation.

Asking questions like, when will I run out of stock, will trigger, trigger a cascade of analysis, fetching data from your trusted data sources, running the analysis, choosing the most valuable ones, and providing insights and recommendations based on that. Clicking on an insight opens the insights panel, keeping a human in the loop, allowing the user to dive deeper into the insights and understand how they're calculated and what they mean before she approves them. The assistant then suggests, suggests, um, some recommended actions and with a click of a button, the shop owner can estock.

Generative AI breaks the barrier for self-service exploration by assisting with what to ask. Increasing data literacy and explainability, providing high value insights and recommendations, and by democratizing the data analyst, making data available to everyone on The AI powered analytics is not restricted to conversational interfaces. Let me introduce you to the next generation of embedded analytics, what we call at Sisense Micro analytics.

Analytical applications can be so much more than traditional analytics, so much more than dashboards. This is what the future looks like in the context of your workflow. AI generated insights and recommendations in the right place at the right time.

I believe that with time and technology advances, the descriptive data layer will start, be, start to become invisible, concentrate, concentrating on the actionable insights and not the data. But to get there, we are missing one very critical piece. We need to be able to trust the insights if you wanna base our critical decision making on it.

Leading me to my next slides, Trust your data. So important. Hallucinations is a real problem, and machine learning 1 0 1 lesson says garbage in, garbage out for enterprises or businesses of all sizes.

This poses a great challenge and risk. In order to provide high quality insights and recommendations on complex data, we need an analytics platform to handle our data. This is the way to truly democratize the exploration experience in an enterprise environment.

Analytic platforms provide a stack of tools to handle the data. A very important part of the stack is the semantic layer. A semantic layer is an abstraction layer that translates complex data from multiple sources into a user-friendly format enable easier access and understanding for end users.

It provides a consistent and comprehensible view of the data, facilitating accurate and efficient data analysis to increase trust in data. When question is validated, this, uh, that is, can be fully mapped to real data that exists in the data sources, and this is very, very important because you, the answer you're getting is coming 100% from your data sources. When a business user refers to these metrics, they don't need to know how they are calculated.

Calculated, rather refer to the metric by name. What is my revenue? The formula behind revenue can be a very complex formula that's up to the data owner to define.

On the other hand, the data owner can trust that the user is getting access to the right data. Lastly, analytical platforms have a layer of data security and governance, making sure that the right people get access to the right data and there are no data leaks. Before we wrap up, let's take a moment to reflect on everything we've discussed.

Today. We've explored the transformative impact of AI on the analytics journey from traditional methods to advanced AI driven insights. Now, as we approach the end of the journey together, I wanna summarize the key takeaways and look at how AI truly revolutionizes the world of analytics.

As we we've explored today, AI transforms analytics journey across the entire data analytics stack. In order to provide enterprises with high value analytics on complex data, you need a robust analytics platform to handle comp complex data relationships, and deliver scalable high performance queries and generate the highest value insights from connecting DA data sources. Through data preparation and modeling to generating insights, AI significantly boosts the capability of the analytics creators.

AI empowers developers of all skill levels to quickly create sophisticated data products by simply simplifying complex data tasks into user-friendly experiences. It democratizes the traditional analyst role putting powerful insights directly into the hands of the users. So whether you're a seasoned data scientist or a business user just getting started, AI is here to make your life easier and your insights sharper.

Thank you for joining me on this journey and buckle up with AI in your toolkit that the sky's the limit to what you can achieve.