Introduction

Thank you for having me here.

About R&B Technology Group

So we are R&B Technology Group, a AI company developing enterprise AI solutions in the platforms for a couple of years.

From Causal AI to Agentic AI

And we started from something very niche and emerging, which is causal AI. And over time, since this year and since the agentic AI is becoming dominating in the market today. So we are embracing agentic AI and integrate them with causal AI.

Focus of This Talk: Enterprise Decision-Making

So what I'm going to present next would be a little more technical and mostly about decision making of enterprise applications.

Defining Enterprise Decision-Making

So firstly, let's define what is enterprise decision making. So take some examples for defining that.

Example: Restaurant and Retail Operations

The first example is about restaurant business we are recently expanding to. So the restaurant owners and the managers, they are keen to understand. And I believe not only restaurant business, any business owners are keen to understand why and how their business

in terms of growth is impacted, and what are the real driving forces to drive the growth in terms of revenue acquisition, in terms of gross or net margins, and customer basis, so how they are causally impacted by, say, in today's heavy digital operation. I believe retail business owners, they usually use multiple digital platforms from administration to sales and marketing to inventory management.

Making Sense of Complex, Multi-Platform Data

So customers come to us, and they want to figure out different insights out of these complex data ecosystem, which they use or maintain for even 10 years plus. So what meaningful and actionable results can be added? by AI, because humans have no way handling historical data over 10 years for, let's say, five digital platforms, including social media, including CRM, POS system, and let's say, booking systems.

Example: Renewable Power Generation

And another example is like, I will

Identifying Generation Loss and Root Causes

Retouching the use case next, say the renewable power generation. So for solar wind farms, asset managers will have to deal with generation loss.

Probably, for example, in a solar farm, probably due to the dust, due to the fault or failure of components. How much power generation loss can be identified, especially in terms of solar radiation and the weather conditions that varies from time to time?

And what are the root causes for these detected generation loss? So this is the customer leads we have been identifying for enterprise users.

The AI Market Gap

And on the other hand, if we are looking at the AI market today, and if you search the top 30 or top 50 AI companies from Google, you will see most of them are agent companies, and 99% of them are handling with generation AI, generation of speech, write-ups, video, image, and many of them are becoming multi-billion dollar companies already.

Why Generative AI Falls Short for Decisions

However, few companies, if not none companies, are touching enterprise applications, especially on decision makings. This is because for generation AI handling business problems, there is a missing piece. The missing piece is the ability of handling data, and particularly numerical data or quantitative data, and the ability of reasoning.

Limits of LLMs with Large Numerical Datasets

Yeah, as I recall, this gentleman asked if he can upload a .csv file to something like a chat GPT. And well, chat GPT can output something for that data set.

But if that .csv file, let's say, become a file with thousands of rows and thousands of columns, I don't think ChartGP can output anything meaningful. So at this point, you need some additional tools handling this numerical data set.

Causal AI Fundamentals

So a couple years ago, When the market, there is still no something called generative AI or large language models in the market. Actually, we started developing this causal analysis.

What Is Causality in AI?

So have you guys heard about causal analysis or causality? OK, so this is a very special, very niche technique in data science and in the AI. Fundamentally or mathematically, it's differentiated from large language models, which is based on transformer or tokens.

From Cause–Effect to Causal Graphs

Causal analysis is something to identify or work out the cause and effect out of data. Could be both structured and unstructured data. And today, multiple techniques to build up causalities out of structured or unstructured data.

Integrating Causal AI with Agentic LLMs

And so, What we have developed over time is to integrate causal analysis or causal AI with super reasoning ability and also numerical data skills with large language models and its extended applications such as agents. And we put them together to work as a gigantic decision-making platform for for enterprise or business applications.

Platform Capabilities

So the key differentiator or key features of this agent platform is advanced modeling and supported by global search and the knowledge base. And we all know If you want to adopt artificial intelligence, working as a human experts or like senior managers, support the decision making, you need skills of math, skills of reasoning, and domain knowledge.

Skills Needed: Math, Reasoning, and Domain Knowledge

You went to MBA, so you must be trained with reasoning. quantitative ability as well as business ability. So business is equivalent to domain knowledge, actually. So if we want to build up artificial intelligence in a serious way, we need to develop models or tools for these three categories.

Causal Discovery and Inference

So a little more knowledge about causal analysis. Usually, it's composed of causal discovery and inference.

Building Causal Graphs from Enterprise Data

So for example, for a retail business owner or manager, they usually use these kind of digital platforms for CRM, POS, social media. And they are usually stored in a structured way in data tables. If it is a small data set, it could be like a spreadsheet or .csv file.

And a causal analysis engine is able to turn that into a graph like this, which is so-called a causal graph. And there are arrows point with directions pointing to core effect. And the node that's starting the rows is called cores.

Quantifying Impacts and What-Ifs

And with further training with machine learning, it can be quantitatively tell you how one particular variables impact the effect.

Pricing and Promotion Uplift Analysis

So again, a great example of retail business, let's say if we offer 10% discount for a particular item we sell or we promote over some social media. So I believe every business owner are eager to know how many or how much additional sales this 10% discount is able to boost.

or able to acquire, or how much customer basis this 10% discount or any discount, any percentage of discount can create for the business. 1So a causal analysis-based tool is able to provide the answer in a perfect way.

Conversational Analysis with Agents

So a genetic causal platform is like we want to perform the causal analysis and any deep, deep analysis with serious models in a conversational way. So it's like we are telling agents, hey, run the causal model based on some particular data we are extracting, say, from CRM.

some additional tools in the powers working together with AI agents.

Use Cases

So next, I just want to showcase a couple of use cases from assets management to renewable energy to risk management in transportation and oil and gas.

Predictive Maintenance and Anomaly Detection

So this is an anomaly detection use case in predictive maintenance of any mechanical or industrial system. So with the causal engine, it is able to map or turn the data points of any mechanical systems.

Usually it's a scanner system for industrial manufacturing, for example. And then it is able to run this causal analysis and generate this causal graph with this random dot as a root cause decided by the inference engine.

And then the AI agent is able to interpret in parallel this color graph in the natural language or in an easier way for users, mostly, let's say, O&M teams or technicians to understand what's going on with the anomaly or with the system.

And we developed a way for users to embed the knowledge into these decision-making models.

Solar Farm Inverter Alerts: Root-Cause Isolation

So a very typical example we got from a real solar farm operation. So one day, the operation team encountered with 14 errors generated by the inverter. And obviously, they were overwhelmed and had no idea of how to react and what actions to take.

So with causal AI, This table can be turned into a graph, and we identify the root cause on the left-up corner. And actually, it is this root cause that caused the other 13 alerts from this problematic component or equipment, so that the operation team can can react correspondingly to handle this root cause. And if they have their handbook or operation manual turned into a knowledge base on the platform, so they can immediately understand and get the answer to handle this issue.

Business Impact and ROI

So this is a business result we got from this particular case study. So I just want to mention, so decision making of enterprise always have a dollar sign. And obviously, we can make ROI and the dollar sign visible to decision makers.

Transportation Risk: Traffic Crash Prediction

Okay, so next two use cases are about risks of transportation systems. So the first example, I believe it's a very interesting one, about predict the probability of traffic crash on a highway or whatever road.

So we took a public data set, which contains six types of data here from location to weather to point of interest severity and the date of time. Point of interest is mostly about the feature of the roading structure. If it is a stop sign or traffic light or a ramp from one highway to another.

Automatic Causal Graphs from Multi-Source Data

And If we run the causal analysis in an automatic way, the platform is able to generate automatically a causal graph like this illustrating the cause and effect of these six types of data or variables to the probability of a traffic accident.

And this is just a visualization of traffic accident map in Texas. And here is Houston, San Antonio, and Dallas. This should be Austin, sorry. Austin, Dallas, and Houston.

And here we also, in this project, we defined a level four traffic accident. So level four is easily said it's the most terrible traffic accident in terms of the length of the line and the time for law enforcement to clean the site.

Location- and Time-Specific Risk Scoring

1So the model predicts if we import some location, a time, for example, the location of a ramp from IE 610 to IE 45. And on that particular time, the probability of a ramp Most terrible traffic accident at that particular location is like 13 times of the average traffic accident of Houston area.

And then we need a model to tell us the causal factor. So the top three factors that caused traffic accident. So location, the ramp from I-610 to I-485, and the time, so that is Saturday night. However, the weather on that night is good, actually.

Actionable Insights for Agencies

So understanding the deep insights like this of causal factors of traffic crash, so the governmental agency or regulation department can take corresponding actions to mitigate the risk of a terrible traffic accident.

Transportation Risk: Insurance Premium Drivers

So next a use case about the transportation is about how to understand the causal factors of insurance premium.

And so I think everyone here have some common sense, like for young people, they usually have to pay high premium monthly for their vehicle insurance. And this is in perfect alignment with the causal analysis generated also by a public data set on our genetic platform.

And also you can take deeper insights about different groups with different age range. and which have different risk factors for their insurance premium.

Retail: Membership Purchasing Decisions

So this is a use case about a retail business, how to analyze the purchasing decision making of half a million members of a sports club in California.

So taking master Salesforce and this just a portion of digital platform they are using for their operation.

So based on the integration of this data platform, we run the AI that generates a causal graph for further causal factor analysis to understand why people purchase or make the decision to purchase their seasonal ticket and any other items, particularly through their digital marketing.

Subsurface Exploration and Geoscience AI

So the last use case is about subsurface exploration. As we are in Houston, the oil capital, and over time, we just came to realize subsurface exploration actually generated tons of data.

And so to... To an AI company like us, we love this.

We are eager to have tons of data and with the same process running on the Agile platform, we are able to achieve technical analysis for specific operations such as seismic well-tight and also enable automatic horizon tracking and and decision-making faults in the subsurface.

Industry Engagement and Outlook

And I think two weeks ago, we were just with Image Conference in Houston.

And we believe this is going to be the next generation subsurface decision-making software compared to all the

Conclusion

Lexi software which is still used in the industry So this is the end of my presentation today