In this episode, Geo George, Account Executive at Salesforce, and Amit Sehrawat, Principal Solution Engineer for Data Foundation at Salesforce, discuss how Salesforce uses autonomous data management to creates trusted, accessible environments. Discover how transportation organizations can use data-driven insights to improve operations.
[Anthony Jimenez]
Welcome back to Carahcast, the podcast from Carahsoft, the trusted government IT solutions provider. Subscribe to get the latest technology updates in the public sector. I'm Anthony Jimenez, your host from the Carahsoft production team.
On behalf of Salesforce, we would like to welcome you to today's podcast, focused around transportation data. Geo George and Amit Sehrawat will discuss how agencies can move from reactive reporting to proactive data-driven operations.
[Geo George]
Good afternoon, everyone. Thank you for taking the time to join us here today. Before we begin, I just want to take a moment to thank all our transportation customers for all the incredible service that you do during this summer holiday season, and also the upcoming 250 celebration, making sure that we safely get to our destinations.
I'm Geo George. I'm with Salesforce. I work in the transportation industry, focusing on local government transportation organizations.
I get the opportunity to work with many transportation organizations across the country. What does Salesforce do when it comes to transportation and supporting transportation organizations? We help organizations leverage technology to enhance customer experiences, improve operational efficiency, and drive innovation.
Enhancing customer experience is basically by providing personalized, real-time information to transportation organizations so that they can improve customer satisfaction and loyalty with their customers. To improve operational efficiency, it's by leveraging data automation so that transportation organizations can optimize their operations, reduce costs, and improve that service delivery. And then finally, driving innovation.
By leveraging AI and analytics, transportation organizations can identify new opportunities and develop innovative solutions. So where do we help with a lot of these organizations? If you look at the customers, we are really incredibly proud to partner with trailblazers across the entire transportation sector.
Our customers include massive transportation networks like the New York MTA, BART, SEPTA, Amtrak. We also support a lot of major airport hubs such as Heathrow, DFW, Changi Airport. And one of the common themes across all these organizations is the amount of data that's generated.
Every single day, a transportation organization generates terabytes of data every few hours. In fact, a single connected vehicle or a transit asset can produce over 25 gigabits of data in an hour across an entire network, which is incredible. That's an incredible amount of data that's generated.
So where is all this data coming from? It's coming from streaming in real time from IoT sensors on engines, tracks, or it's also flowing from GPS trackers. It's also coming from traffic cameras.
It could be coming from maintenance logs, weather feeds, passenger mobile apps. Essentially, at every point, there's a digital footprint that's created. So we all know that the potential of AI is well recognized from predicting to scheduling, automated scheduling, real-time route optimization.
But all of that lies in the effectiveness of your data. And without a data strategy, vast quantities of information are really fragmented across different data silos. And really, the critical thing here is establishing a data strategy.
So today, we're going to really show you how to establish a data strategy and build that foundation of data so you can actually really leverage AI capabilities. So I'm going to turn it over to my colleague, Amit Sarwat, to kind of walk us through. Thank you, Geo.
[Amit Sehrawat]
Before we begin, I do want to extend the gratitude that Geo shared as well, which we have for our partners at transit authorities. I guess I'm a pretty good example of that commuter who's in New York City right now, who had to take a train and a bus to get over here, and with everything going around with Knicks games and then the World Cup, having access to reliably move from place to place with that kind of event is amazing. So thank you for doing that.
And as Geo said, today's session is going to be pretty much about how do we bridge the data gap, which we normally see in any transit authority, and how can we reduce that gap but also use it to engineer our systems so they could be reliably consumed just not by humans but also by AI agents. First of all, thank you for your time and attention today. And today, we're going to speak about what is really happening, what is the opportunity out there for us from an AI perspective, but then what is it going to take to really have that AI, really grab that AI opportunity and provide a consumer experience to reduce our operational cost, to be better at managing our assets, what are those use cases, how does the data foundation play a critical role, why does it matter, right? And some final thoughts on next steps, what it may look like.
So the first question, I guess for all of us, which we are looking at is that AI, we have a lot of tools which can sift through a large amount of data, a huge amount of volumes, Geo spoke about terabytes of data being built, which is really good to have, but is that data really worth it if we can't really understand it, right? And is that data really error-free, and you spoke to it about data quality is absolutely one of the key things. And then as we start introducing these insufficient or not unified or bad data to an AI model, what's going to happen, right?
Is it going to make the entire system less reliable? And if it does, then does it really provide us the very value which we are seeking from AI itself? And we know that bad data doesn't happen by accident, it does happen by accident as well, but in reality what really happens is that we don't have enough controls or governance or standards today because we have built things in different timelines, right?
We started somewhere in 70s and 80s and we kept building in different platforms and different technologies, but we also built in silos. We built things with one system integrator here and then built something else there with a very different framework and all together. So it's by design that we have got to a point where our data is not exactly as clean as somebody else might have it, like Airbnb and Uber's organizations we have been built on cloud natively, they're going to have much higher quality data because they just came to it right now.
They don't have to carry on the legacy data systems as we have to, right? And these are the statements coming from US three-letter agencies to Congress on why they can't adopt AI as soon as Congress really wants them to, right? But this is also a question for us, from our constituents.
Why are you not giving us access to amazing experience which somebody else could provide, right? Primarily because they don't have the same problems as we do. So let's talk about it and let's talk about a future of it.
So we are in 2026. Believe it or not, there were no LLMs three years ago. There was no chargeability three years ago.
It came as a storm and it absolutely took us with it as well. And now you can't have a conversation in the industry itself which doesn't somehow gravitate towards AI and LLM. So let's actually think through this.
Let's think through this and where are we really going and what it's going to look like. So in 2028, like a good couple of years from today, how are commuters going to talk to agencies? I'll give one simple example.
Yesterday I was in New York City and I had to take a bus, something I'm not very used to being from Boston. We don't have a lot of public transport, not as huge as what we have in New York City. So when I had to take the bus, I needed to know when is the last bus which I can take.
Can I leave after the dinner or do I have to leave before the dinner? And first I asked that question to chargeability and I didn't get a really reliable answer because of what's going on with FIFA and everything else. And then I asked the same question again to Claude before I did anything else.
And after not getting an okay-ish kind of answer, I ended up calling somebody I know who's using these bus systems for a long time. So as you can see, the way we interact, the way we interact with our transit agencies is changing. And there are so many different channels that you put yourself out there today to talk to your commuters.
You give them an option to call you. There is a chat option at a lot of transit agencies' web pages. There's a self-service portal where you can go.
Some agencies have opened up themselves with different text and text messaging services like WhatsApp. There are kiosk station at MTA over here where you can go and you can figure things out. There are smart speaker integration, which some agencies have built now.
But you could ask a question to Alexa or Google and you can email, you can use Facebook. And now there's a new novel brand new channel, which is live agents. But what does that mean?
Having so many channels, how do we get exactly same data? How do we get exactly same answer? And how do we become ready for the world where we know our commuters are going to talk to us?
Primarily through their own LLM applications like ChargeBDs and Clouds and Gemini's of the world, right? And how do we make sure that we don't have to stretch our existing stretching budgets to make that as a reality? Well, there is an architecture out there.
And that architecture is very well supported by the entire Salesforce data foundations. But it starts with having a reference architecture where at the very top you have a single agent. And this is AI agent.
And if some of you don't know what AI agents are, let me explain what they do. They are an implementation of LLM, which are creating agents, AI agents like personas, which answer very specific questions in a very specific manner. So they are very narrow in the scope.
They have a very tight classification. And they have an instruction set. So they use LLM to have some conversation capabilities.
But the agents are very much grounded in your data, in your specific data. And they have a very narrow scope. So you're going to look at some of them, so don't worry.
But in this scenario, they're going to be a single concierge agent, which has only a few jobs. It is not a specialist in fare collection. It is not a specialist in figuring out what is the mean time to prepare for a specific asset.
No. Its primary job is to figure out what is the intent of the question. To do identity resolution successfully of the person who is asking a question, if it needs to.
To route the prompt which is coming in to a very specific agent which can do the job. To understand the context behind the prompt and do a handoff. And to handoff that conversation to specialized worker agents.
And this is where the world is coming in. This happens today. If you call MTA Helpline and if you have a question around lost and found, they're going to route you.
They're going to route you to lost and found department. And then you ask them, hey, I was on this train a couple of days ago. I lost a camera bag.
Could you help me in locating it? This is exactly what a concierge agent is going to do. It's going to understand your intent and route you to a different agent.
But not a real routing. You're not going to be on hold for five minutes until some agent picks up. No.
It's going to be seamless. You're going to keep talking to the same agent you started with. It's just that a special agent is going to answer that question because that agent is built to answer that question.
It has a very narrow scope of doing that. But how is a lost and found agent is going to work, right? It in itself is going to connect to existing applications.
The application where you log lost and found data. The application where you do your refund and billing. Application where you do your fair equity programs.
Now, the question is, how does it do that? Well, it happens through MCB. And if you haven't heard a lot about MCB or you have heard a lot, but you want to know more about what it is, how it works.
MCB is basically, in generic terms, it's a USB-C kind of thing. It's a USB cable. It's a good analogy for an LLM to connect to your systems.
And it could be any system. It could be any external system. It could be a case management solution.
It could be your own docket application, your Java application. It could be a mainframe. It could be Salesforce.
It could be anything. The intent over here is that there is a very secure, very transparent, and very auditable way to connect an agent, an AI agent or an LLM, to your existing systems. And that's called MCB.
And we're going to talk a bit about it. But then even beyond that, we know that our data is not clean. You yourself voted for that when we asked you that question.
The data quality is still a big issue. We may be able to bring a great LLM, but that LLM output is not going to be that great primarily because it is consuming the data, which in itself is not very clean. And this is going to be powered primarily with some kind of MDM capabilities or master data management kind of capabilities where we could figure out that how do we identify our commuters?
How do we converge the profile of multiple people into the same person? Because they were always the same person. They have changed their email address perhaps.
They have changed their addresses in the last couple of years. They have maybe changed their phone numbers a bit here and there. But at the end of the day, they are the same person, and we can unify them under a single unified profile for commuters.
And we can do the same thing with our assets and so on. This is going to be pivotal because that is the answer to the data quality problem. And all of this data is going to come from your existing system fricates, your real-time feeds, ERP systems, asset management systems, scheduling systems, and so on.
And it's all going to use any model because we know the models are rented. You are not going to build your own LLM. We are not building our own LLM.
Maybe we are at Salesforce, but that is not your core business. Your core business is providing the services to commuters, but also your core business from an IT perspective is managing the data, understanding the context behind it. And the models are going to be rented from OpenAI or Entropic or Google or whoever is building them.
And they're going to handle all the knowledge base, and you're going to do all the orchestration over there. But none of this is going to happen where you give a huge amount of weightage to data governance. And that governance comes from these guardrails.
How do we identify and authenticate somebody as a commuter so we can take an action on their behalf, which makes sense? How can we identify and authenticate a caseworker or a mechanic who's working on a bus, right? All of that needs to happen, and it needs to be passed down from agent to agent as we do that.
And then how do we take that consent and store that consent that somebody wants to create a case, we want to create a case on somebody's behalf? And how can we be purposeful about it? How can we make sure that there's no PII data leakage?
How do we make sure that we're grounding in our data so we don't have hallucinations going on? How do we make sure that there is a log available for every interaction between every human through an agent and between agents as well? And how it's accessible?
All of that is, I believe, how the transit agency is going to look like only in a couple of years. So the opportunity is there. A reference architecture is there.
The tooling is there, and the technology is there, and I'm sure there's a will there to make it happen. But let's see what it would take to make that happen, right? So we're going to talk about it.
And the primary question is, do you have a data foundation to execute this? Do you have ability to provide the right information to the right person, the right time, the right format, the right quality to support a right agent to ultimately make that decision? Because in general, as we speak about this future, we forget one thing, which is that humans are very versatile.
I call this term human cushion, which basically means that humans provide cushion between good decisions and bad data. Because we have the acumen, we have the tribal knowledge, we have the domain knowledge to make sure that we still do that. Can a probabilistic or non-deterministic agent do that?
Probably not. That's going to reduce your efficacy by a huge, huge margin. So where do we start?
Well, we start by acknowledging that what do we have today? We have our data and silos sort of disconnected, some deduplication as well. And all of these are very common characteristics.
And this is something we have seen a lot, right? So where do we start? Well, we have to keep our goal in mind.
Our North Star is always same. We want to save money and we want to simplify our operations and we want to provide a great service to our constituents while building AI readiness. Because we may not be doing AI tomorrow, but we're definitely going to do it and not day after tomorrow, maybe in months or weeks or a year, right?
So we have to get ready for that. So as we get ready for it, we have to think about our data gravities. What are the things we care about?
We care about asset management, right? We care about commuters. And all of this is just going to be a precursor for building a complete agent in government.
And then as you think about it, on the left, what do you have? What we're doing today and we'll continue doing. We're going to have legacy forms and fields and papers and siloed data.
And then somehow we need to get it to a new agent in government because constituents are going to expect it. I'm going to expect, and I know about how complex transit agencies are from their IT landscape, but I still don't feel well. I can ask this question to Uber.
How can I not ask this question to a transit agency's agent, right? Why does it take so long for me to go to the labyrinth of a website and get an answer? The question is, which we're going to answer today, how do we do these things?
We start with data. We start with the data because we want to make the data visible. We want to make sure that we know what we have.
We can govern it. It is high quality that we can trust it. So if you take an action on the data and somebody calls you and asks you, hey, do you completely trust the data?
And you should be able to say yes, because if you can't say yes, the AI agent is not going to think twice. It's going to take an action on it. So it all starts with the data where we need to get to a point where it's completely actionable or at least actionable to a layer where we can say, okay, we are good.
This is not very different than self-driving cars. Most of us have some or other level of a self-driving car, either lane assist or keeping you in a lane at the center or even taking turns for you. Who knows, right?
There's so many levels of AI already being used today on the roads. We did not get to a point where we said, okay, I'm just going to trust it. We tested it a bit, right?
We came through phases to a point where we started trusting it and then started letting it take actions on our behalf. This all starts with the data. I think more and more of our organizations are going to start thinking about how do we make it natively accessible by LLMs to do this?
Because we know humans don't speak HTTP, right? But we use HTTP to get to a transit agency's web page and find the data. And we are very visual, right?
We like to get our information in a very specific way and ask those questions to LLMs. And LLMs are only going to have two options. Either they can just go and look at your website and scroll it and scan it and crawl it and get you the data. Or we can build a meaningful way for those LLMs to connect to us.
Something like an MCP. We are still in a very early stage of LLMs, I believe. But we have a fifth version of ChatGPT out there, right?
We have Opus 4.8 out there yet right now. And Fable is in the corner as well, right? So I want to bring the focus back on the data.
How are people going to consume us, LLMs? And what is going to make those answers better? LLM prompts as well as the data.
So let's start with the data because that's something we can control. We know that we can't control building LLMs. That's somebody else's job. That's outsourced out there, right?
We know that we can't change the prompts completely as well because that's a consumer behavior. We can decorate a prompt as it comes in. Absolutely, we have those capabilities.
But what we absolutely own is our data. And how do we unify that data in a way that we can bring the commuter at the very center of it, but then we can bring this data from all of these different systems, from like tap-to-pay, what we have at Omni, at MTA, and account data, case data, lost and found data, benefits, everything else. How do we unlock it so it is available to different agents?
How do we trust it through cataloging and lineage and all of those things, right? And how do we make it consumable for every application out there, like disruption recovery, fare equity, paratransit, fare evasion, personalized trips? How do we do all of that and activate that data across our agentic enterprise, our analytics layer, as well as our own applications?
Well, at this point in time, you're going to have a question from me. You're going to say, Amit, this is all good, but my data is not clean, ready to be a part of a commuter 360. Well, I don't think anybody says, but there's a way to do that.
And that way is actually quite simple. It is a multi-step process. And when I say simple, like simple to have a vision and build a technology landscape around it, we need to have an ability to ingest the data as is where it is.
And then we need to have a capability to standardize and enrich the data with data quality rules for addresses, phones, emails, and fill the gap with the low confidence data by basically pulling and unifying the data based on a fuzzy logic and a rule. Then we need to match and merge it into a single record, a single golden record, and then have a golden record available to us, which we can govern, and we can expose to our experiences and publish it back to the source system. So every source system have exactly same unified profile.
The addresses are not different. The emails are not different. The phone numbers are not different.
And it's completely, again, consumed by your existing application. So there is a way of getting to your commuter 360. And exactly the same way you can get to any 360, you can do that with ACID 360.
You can do that with any other large data gravity programs for you. And what this allows you to do is to understand, like, how you can do a predictive maintenance, because now you know far more about your assets than you ever did. What is the state of good repair capabilities?
What is the uptime for your existing trains and buses and escalators and elevators? And why did you have an incident? Did it happen because of a safety issue or operation issue or maintenance issue?
Is there a way to, again, go back and predict that? Can we reduce our mean time to repair across our different assets, right? So this is the value.
And when you think about the value, you can imagine that value today in the existing enterprise. Imagine the value, going back to the previous hypothesis, that a trusted data context is going to be absolutely pivotal to have great AI outcomes. And it has to be a precursor to that agenting journey itself.
So now we're going to talk about opportunity a bit. And this is something which we did a while ago. We asked the CIOs, what is their priority focus in the government today?
And primarily, we heard things which we have been hearing for a long time, like cloud, cybersecurity, controlling the budgets, making sure the government is more digital. But AI, AI was absolutely the top. It's top of the mind.
And what that means for us is that we are going to go digital completely. And we're going to modernize. I mean, we can't go completely digital because we know that we, as a transit agency, have to provide a paper form and we have to provide each and every way of communication.
But we can still digitalize as much as we can. We can have data management and analytic strategy. And all of that is going to power AI because our agents are just not going to be FAQ agents.
They're going to create a task for themselves. They're going to do things which some humans are doing them today. And they're going to do it faster.
And because of a trusted data context, they're going to be more accurate. And they're going to be more scalable when we host a FIFA World Cup or when a game out of nowhere in New York City and you have a huge amount of line of people who are asking exactly the same questions. And we know our call centers cannot handle that.
So the question is AI is ready for you, but how ready is your data for that AI itself? And this is going to happen through a lot of different scenarios. It includes having agents and fraud detection capabilities and document actions.
But primarily it is a paradigm shift which is happening from data-based data to knowledge-based agentic era, which is coming in front of us. And this is going to need access to the data which may be structured or unstructured somewhere in applications and data warehouses. And it's also going to come from your knowledge base, which is going to be about your data semantics, your connected data metrics, and interoperability across all the different applications which drive your trusted context.
And majority of the leaders believe that they're experiencing inaccurate or misleading AI outputs. And I guess we know the answer because you gave the answer yourself that the agent efficacy, AI efficacy is highly driven by the quality of the data itself. So let's talk about the quality of the data.
We call it trusted context. What does it mean? It means do we have an understanding of where the data is coming from?
Do we have an understanding of the lineage of the data, freshness of the data, quality of the data? And do we have the authorization to even look at the data? Because we have spent a huge amount of time and effort in building role-based security, attribute-based security.
How do we pass on that to agents? And how do we make sure that nobody's going to engineer a prompt specifically to get the data they shouldn't be getting? All of those answers are in data foundation.
So a successful agent government is going to rely on a very concrete data foundation. And this data foundation is highly connected. It can connect to a mainframe.
It can connect to PeopleSoft. It can connect to Infor Asset Management. It can connect to Salesforce and ServiceNow.
And it can connect to your web portal and everything else. And as it pulls the data out, it can create a single source of truth. It can make the data very, very trustworthy where you can defend the data all day.
And it's going to catalog all of that across your enterprise metadata. And it can now activate that data back to your analytics, your AI agents, your applications like CRM and case management and everything else. And this means that you will have a very scalable, very strategic, and very AI-ready data foundation.
So 40 minutes into this, let me present to you what is data foundation portfolio. It is a combination of capabilities. And for you, what it really means is that Salesforce data foundation can solve multiple problems for you.
It can allow your applications to talk to each other. It can apply your data to talk to each other. It can allow your APIs to be converted into MCB in a couple of clicks.
So you can go ahead and make that data available to LLMs. It can help you and your agents to communicate to other agents using A2A. We can catalog your metadata. We can highly govern it to MDM.
We can unify it. And we can activate it across different layers and analyze it with W. So the framework for our data foundation is basically a three-step process.
And these three steps needs to happen more or less at the same time. You need to have an ability to unlock the data through any integration pattern like streaming, batch, real-time, zero-copy, APIs, and even driven architecture. And then we need to add a trust layer to validate and protect and unify the data through data cataloging, data lineage, data quality, and then unify that through MDM.
And once we have done that, this very highly trusted, very connected data is ready to be activated. It's ready to be activated for making decisions, to orchestrate like business processes workflows, to make it available to an agent or an existing business process, or basically building a complete constituent 360. All of that is completely available through Salesforce Data Foundation.
Now, this is a quite elaborative way of you looking at the unified data architecture. But this is reality as well, because this is what's going to happen. You already have the bottom layer today.
You already have your enterprise applications, your unstructured data, your data lakes, your multiple channels and products. And they are going to work. They are going to be available to a complete agent fabric.
And MuleSoft agent fabric is a control plane where we can discover existing agents. We can catalog them. We can go on them through MCPA2A.
We can go in our existing LLMs to LLM gateway, and we can provide a complete guardrail systems we spoke about with PII detection and making sure that we understand the LLM cost, token cost optimization, which is a really, really cool thing in itself. And how can we make sure that we are not letting any prompt injections happen, right? And this is, again, a capability through MuleSoft and Informatica and Data360 in providing these application integration and data integration along with the trusted data foundations.
And what do we do with that? Well, we then expose it. We expose it to agents built on different tech stacks, like it could be AWS Bedrock or Azure Foundry or Google Vertex AI, or it could be Gemini Enterprise, right?
And we're going to also expose them to data agents and existing applications or any system out there. So now the value over here is that we are not just building a highly qualified pipeline for our existing applications and systems for our data, understanding our computers and our assets better, but also getting ready for AI in a single move. So why does it matter?
Well, because the data foundation outcomes are important for everybody. You want to, and this is something we are already doing, right? We're already trying to consolidate our platforms.
We want to reduce our IT burden by this consolidation. We are not highly focused on a dependent place. We are trying to do workflow automation, so repetitive tasks could be automated.
If you want to delay verification later, it shouldn't take like a week to get that, right? If I want to dispute a payment because I was charged twice, it shouldn't be taking like two or three days to make it happen with uploading multiple documents. This could be automated, and we can have real-time data pipelines, and we can start getting ready for agent transformation for those autonomous AI agents by building a highly governed, highly qualified data integrations as well as MDM capabilities, so we can trust our answers coming back from our AIs.
And we have seen this a lot. We believe a good data and a bad LLM or like a cheap LLM is usually going to do better than bad data and a great LLM because a great LLM is still going to hallucinate because it doesn't have access to quality data. And this also powers your analytics and intelligence, and it absolutely helps you in tracking your program outcomes and being that AI ready.
We absolutely want to be. So now you'll start thinking, well, I'm doing this role in this agency here. Who really benefits from it?
Well, everybody, everybody in a way. The CIOs and CISOs and CDOs benefit from a chief operating officer's perspective because they now have better understanding of their data and their business and their workflows and security, making sure that the data is data which you already own. You have collected very successfully for a very long time and you have built those large data structures is now trustworthy enough to empower your AI agents.
It also helps business side of the house because your program directors and program analysts can now better understand the efficacy of those programs, your program integrity, and how each and every change in policy is impacting your constituents you're serving. And overall, who benefits? It's commuters, right?
Commuters benefit most. And how do we quantify all of those benefits? Well, in a way, everyone is going to get more accurate data.
Everyone is going to get more time out of their way. But also, we're going to provide a far better standardized data governance and application governance available natively to us. So in a nutshell, you and me and everybody over here is a benefactor because all of us are touching all of these systems in every different way.
Meet Susan, right? This is a really good example because Susan Alvarez has been a fleet maintenance manager in the Tri-County transit for like more than a decade. She knows her buses very well.
What she doesn't know and what nobody at Tri-County can actually see is which of her 214 buses are quietly failing. And why? Well, because the challenges are plentiful.
But one primary question which stands out is what is the true condition of our asset and what will they cost us for the next couple of years? And we cannot answer that today. Why can't we answer that?
Because the data is captured inside our systems. Our processes are highly reactive. And the data quality is poor enough where we can't match our model numbers or our CDL numbers.
We have 215 model numbers, but we only have 217 buses, right? So who has seen that before, right? So when that happens, what is really going on is on a Tuesday morning in a nice cold February when Susan pulls her maintenance report from her existing EAM system, pick any of them, that her bus 1147 shows that this last inspection was complete and it was all green and it's all good to go.
But what the report is not telling her or anybody is that the inspection notes in a very different spreadsheet somewhere maintained by the contractor has flags that there are brake caliper issues, which never came back to the EAM. They were never captured. They just kept living in a very different spreadsheet maintained by a different team altogether.
And Susan doesn't know about it. She's going to go ahead and she's going to clear the bus 1147 for the route 22 express run. And the high ridership corridor is going to be there with a peak hour, 58 passengers.
If you live in this area, in the tri-county area, you're going to say, let's say route 139, right? Running every five minutes. And the 12 miles into the route, the driver is going to notice that the brake is fading and he needs to pull over and the vehicle needs to be towed.
And the 58 people are standing outside in February because we did not know it's going to happen. We couldn't predict it because we didn't have the right data at the right time in front of the right person. And that is a value of building a complete data foundation.
But there's also a financial penalty over this. It's not just about the human cost and the planning cost. It's also that if we can pull the bus out right before it and fix it with an OEM like clipper or something, we're going to spend some money.
But the cost of having the tow, the emergency dispatch of a vehicle, the overtime and the contractor cost and the FDA response time and the hit, which is going to come on social media and the operational disruption is going to cost far, far more than that. And it's also going to cost Susan in her performance degree because the data existed. It just doesn't exist for her, not in her timeline.
Right. And because she was just at the wrong place. So how do we do this?
And this is exactly what we do. We have to reimagine how do we do our asset management systems. We need to pull the data from all the systems with capabilities like MuleSoft where we can connect to these systems natively with our standard connectors.
We can pull the data out in real time. We can clean it. We can make sure that it's highly governed.
And then we activate the data through Data360 and analyze the data through W. And the outcomes we're going to get is not just saving the money through predictive alerts, but also having a very clean and nice compliance for federal reports. We can automate our work orders and we can have executive dashboards, which are going to help them in understanding the organization itself.
And then a while later, the Tri-County Data Foundation is live. Every inspection record, whether it originated in the EAM or the contractor spreadsheet or a technician mobile work order, all of them completely fall under that asset 360 profile and steady to be consumed. And Susan is never going to fall in a trap because next time Susan is going to see that and it's going to create a work order.
And she's going to pull the vehicle out of the route roster and a spare is going to get dispatched. And the repair cost is not going to be $47,000. It's going to be $400.
And that's a huge outcome, right? So when we think about this, what can a data foundation unlock for us as transport and transit agencies? It can give us a complete real time access of the health of our existing assets.
It can allow us to predict. It can create outcomes, which make sure that we spend least amount of time and keep our assets for longer with the predictive maintenance. It can help us in ground support for all of our equipments.
It can automate the FFA part 139 compliance documentation, which needs to be dispatched. It can help us in making sure that every asset we have, every tollbooth we have is always running healthy and it's always well managed. And we're not spending a penny more than we really have to.
And you can grow with this data foundation to making sure that the connected systems, the systems are connected. They are highly informed and they enable us in a predictive AI assisted alert mechanism so we can automate all of those work orders. So I have the last couple of key takeaways.
The first one is about strategy. AI at an enterprise scale really requires a very robust data strategy and very strong data governance. And we have to start working on it today.
And then there is about trusted context. Do we have the context which can provide meaning to the data? You may have a number called asset failure, but without knowing what it means with the context, you're not going to make a decision.
Right. You're not going to take an action. You won't be able to drive agent journey and also the completeness.
Salesforce Data Foundation is one of the most comprehensive portfolios out there. I would say it is the most comprehensive portfolio out there which can allow you to build a trusted context, which you need to build an agent in government.
[Anthony Jimenez]
Thanks for listening. And thank you to our guests, Geo George and Amit Sehrawat. Don't forget to like, comment and subscribe to Carahcast and be sure to listen to our other discussions.
If you'd like more information on how Salesforce can assist your organization, please visit www.Carahsoft.com or email us at Salesforce@Carahsoft.com. Thanks again for listening and have a great day.