Practical AI for the Enterprise: Lessons From Salesforce Customers

Explore the business impact of AI with Clara Shih, CEO of Salesforce AI, in CXOTalk episode 823. Learn about AI transformation, future trends, and ethical AI practices.


Feb 09, 2024

In CXOTalk episode 823, Clara Shih, CEO of Salesforce AI, shares insights on the integration and impact of generative AI in Salesforce's offerings. Shih discusses the development of AI applications like Service GPT and Sales GPT, emphasizing Salesforce's commitment to ethical AI through its Einstein Trust layer. The episode covers the practical use of AI in enhancing business operations, with examples from companies like Gucci, and addresses the broader implications of AI on the workforce and ethical considerations. This conversation provides a comprehensive look at Salesforce's AI strategy and its implications for businesses.

Episode Highlights

Generative AI's Impact on Salesforce

  • Salesforce is incorporating generative AI into its products to streamline processes and enhance user experience.
  • Generative AI capabilities will be used to improve the efficiency and creativity of sales, marketing, and customer service teams.

Challenges of Enterprise AI Adoption

  • Security, privacy, and trust remain key concerns when businesses integrate AI into their operations.
  • Companies need to ensure the responsible and ethical use of AI, prioritizing transparency and protection of sensitive data.

Strategies for Successful AI Implementation

  • Companies should start with clearly defined use cases with potential for tangible results.
  • A phased approach is advisable, beginning with pilot projects and gradually scaling AI initiatives as expertise grows.

AI's Transformation of Customer Interactions

  • AI is revolutionizing customer service, with intelligent chatbots and virtual assistants providing 24/7 support.
  • AI optimizes marketing efforts with personalized recommendations and targeted campaigns.

Automating Tasks with AI

  • AI automates repetitive and time-consuming tasks, freeing employees for more strategic work.
  • This leads to increased efficiency, reduced costs, and improved productivity for businesses.

The Critical Importance of Data in AI

  • High-quality data is the backbone of successful AI projects, enabling accurate predictions and actionable insights.
  • Businesses need robust data governance strategies to manage and prepare their data for AI initiatives.

AI and the Changing Job Landscape

  • AI will displace some jobs while creating new opportunities requiring specialized skills.
  • Businesses and individuals need to invest in continuous learning and adaptability to navigate the evolving workforce.

Education and Training for the AI Era

  • Educational institutions need to revamp curricula to prepare students for a world driven by AI.
  • Businesses should upskill and reskill their workforce to leverage the full potential of AI technologies.

Collaboration for Responsible AI

  • Governments and businesses must work together to promote ethical and responsible AI development and deployment.
  • Regulations and guidelines are needed to safeguard against misuse of AI and potential biases.

The Future of AI

  • AI has the ability to transform various industries, from healthcare to finance and beyond.
  • The future success of AI hinges on its ethical use and society's ability to adapt and thrive with this technological revolution.

Key Takeaways

AI Adoption Requires a Strategic Approach: While AI shows immense potential, companies cannot approach it haphazardly. Successful implementation requires starting with well-defined use cases, ensuring data quality, prioritizing security and privacy, and approaching it with a mindset of responsible use.

AI Transforms Customer Experience: One of the most significant and immediate applications of AI is in revolutionizing customer service, marketing, and sales. Businesses can expect AI-powered tools to deliver personalized experiences, intelligent support, and targeted campaigns with greater efficiency.

The Future of AI Depends on Collaboration and Ethics: AI's long-term success and positive impact depend heavily on collaboration between governments and businesses. Regulations, guidelines, and continuous education around the ethical and responsible use of AI are crucial to ensuring that AI is a force for good that benefits society as a whole.

Episode Participants

Clara Shih is CEO of Salesforce AI. In this role, she oversees artificial intelligence efforts across the company, including product, go-to-market, growth, adoption, and ecosystem for Einstein GPT, the world's #1 AI for CRM. Einstein GPT delivers over 1 trillion predictions and generative automations every week across Sales Cloud, Service Cloud, Marketing Cloud, Commerce Cloud, Industry Clouds, Mulesoft, Tableau, and Slack. Previously, Clara led Salesforce Service Cloud, the world’s #1 customer service, contact center, digital service, bots, and field service solution. A digital pioneer, Clara has been named one of Fortune’s “40 under 40” and “Most Powerful Women Entrepreneurs,” Fast Company’s “Most Influential People in Technology,” and a “Young Global Leader” by the World Economic Forum. Clara is a member of the Starbucks board of directors and serves as Executive Chair of Hearsay Systems, a privately held digital software firm she founded in 2009. She graduated #1 in computer science at Stanford University, where she also received an M.S. in computer science. She also holds an M.S. in internet studies from Oxford University, where she studied as a U.S. Marshall Scholar.

Michael Krigsman is an industry analyst and publisher of CXOTalk. For three decades, he has advised enterprise technology companies on market messaging and positioning strategy. He has written over 1,000 blogs on leadership and digital transformation and created almost 1,000 video interviews with the world’s top business leaders on these topics. His work has been referenced in the media over 1,000 times and in over 50 books. He has presented and moderated panels at numerous industry events around the world.


Michael Krigsman: Welcome to CXOTalk episode number 823. We're speaking with Clara Shih, who is the CEO of Salesforce AI.

Clara Shih: I lead our AI efforts across the company. Previously, I was the GM and CEO of Service Cloud, which is our portfolio of customer support technology. And a couple of years ago, we started prototyping generative AI applications. And this is, you know, well before ChatGPT was released, we worked on some very early prototypes and, you know, all the events from the last 18 months have really, you know, interest across all of our customers in these products.

And so last year, I took on a newly created role of a leading AI across the company. And our first order of business was to, to productize and to ship the initial set of generative AI applications, including Service GPT and Sales GPT and our Einstein trust layer. And then the second order of business, which we are just a few weeks away from releasing, is building our AI platform, a set of shared services that all the Salesforce clouds internally can consume.

But also our customers and our ISV partners can also use too. And so specifically we're going to be releasing our Prompt Builder, our Copilot and our Copilot Studio on February 29th.

Michael Krigsman: Clara, you work with so many customers of Salesforce. What is the state of adoption of AI in the enterprise today?

Clara Shih: It's varied, and Salesforce works with companies of all sizes from the fortune 100 all the way down to one and two person proprietorships. And so now our business has always been it's a third, a third, a third to be a medium sized businesses and a very large enterprise. And so, you know, across the board we're seeing very keen interest in generative AI.

There's been a lot of customers that have been trialing our GPT products that are in-market and a very a smaller number of them that have rolled out their first use case. And then, as you can imagine, an even smaller number that have rolled out multiple use cases into production.

Michael Krigsman: Do you see patterns by industry or type of company? Are there are there concentrations where AI is being adopted more rapidly or where there's resistance, regulated industries?

Clara Shih: There's just more considerations in terms of proceeding. but I'd say across the board, whether it's been a regulated industries or an unregulated industry, is every company that that we meet with their number one question is around trust. And both on the data privacy data security side as well as on the ethical guardrails side. And so this is really why, you know, one of the very first products that we've built and put out in the market is our Einstein Trust layer.

And that encompasses everything from data masking of proprietary or personal identifiable information, data grounding with our data, cloud citations, keeping an audit trail toxicity filter in may be the most important is the zero retention prompts, so that nothing that's sent to a large language model ever gets retained, into the training of that model. And so, I think the companies that have gotten comfortable with AI, it's really because they feel confident about the security and the data privacy guardrails that we put in place.

And they've been able to scale out from there. And I've seen we have we have companies, you know, from Heathrow airport to triple insurance to, you know, Gucci that are, you know, really across industries that are adopting these technologies and really starting to see an ROI.

Michael Krigsman: You know, it's interesting that you say that the trust, the security and the privacy are the number one concerns. I would have thought that it's the capabilities and the opportunities of AI. So it's a little bit counterintuitive to me.

Clara Shih: It's surprising. I think maybe this is if we can trust it to the metaverse or blockchain. Right. People are like, well, what's the use case? And then, is it? I think in this case, everyone gets it right. Everyone understands because they played around with ChatGPT or with Bard. They get that generative AI is really powerful. And so, I think they kind of skip ahead to say, okay, this is really powerful.

A lot of knowledge work is going to be automated or at least drastically accelerated. How do I make sure I deploy it in a secure and safe and trusted way?

Michael Krigsman: So the issue is not so much is this a beneficial thing to us? The issue is we are going to use this. And so the question then becomes, what's the best way to use it? And to satisfy the security and trust issues that you're describing?

Clara Shih: I think so. And I mean, of course, right after we answer the trust question and then there's the roadmap question of where do we start? And certainly, that is predicated on, on the business case and understanding how am I driving greater sales productivity? How can I take my average seller, who today is and spend 70% of their time on admin non-customer facing tasks like preparing a quote, sending an order, sending follow up emails?

How do I reduce that even by a few percentage points and get my seller more productive? or how do I streamline my customer service so that the customers are no longer waiting? No 20 minutes an hour to talk to somebody or to get an email response, but they can get it much more instantaneously.

Michael Krigsman: Clara, you were just describing the roadmap question how do companies go about thinking of their investments as they look at AI? And they're thinking, well, how do we invest in this? What do we invest in? Because it is so new.

Clara Shih: What I'm seeing customers do is they're saying, I just want I want a few use cases that I know will work. And what we're seeing and this, this has been well covered. From McKinsey reports to Goldman Sachs reports BCG, Bain you know all the consulting firms it we start typically with customer service with marketing and also with sales and software development.

These are very well-known domains that are tried and true for large language models. And so, I'll just talk about customer service. And that's a business I used to run here at Salesforce. I mean, if you think about the time and motion of a typical as to where support reps today, the customer question comes in through email, through phone, through chat, whatever channel.

And that the lag time usually involves this customer service rep having to to hunt down the knowledge to answer the question or resolve the issue. And this will typically involve, you know, one or realistically dozens of different knowledge articles, product documentation that they're consulting and kind of reading through and trying to match up the section of that knowledge, article or documentation with the exact customer configuration.

Then there's no answer. So, then they're on slack messaging. Their colleague and try to figure out the right person, get the answer or they it's just beyond their knowledge and they have to, put the customer on hold and transfer the customer to someone who has that knowledge. So, this is like a huge bottleneck. It's a huge inefficiency.

It's frustrating for the customer support team. It's infuriating for the customer. And this is where we've started. And so many of our customer starters like a a simple use case like that that can really unlock this bottleneck. And so that's why I think we've seen so much uptake in in the service reply recommendations. That was one of the first features being rolled out in service GPC is, you know, regardless of channel, the question comes in and before the human wrap is even working it, we can already have I instantly looking up dozens, thousands of articles and slack transcripts about the similar issue and drafting the answer and so that it's there waiting. 

And so, I think that's step one is like, what are the quick wins that can really automate these tasks, these owners tasks and bottlenecks? I think phase two that some companies are already thinking about is okay if I do this in mass, right? How do I actually change the job description of everybody on my customer support team?

How do I actually start to reshape this department to maybe do more? And this is exactly what we saw at Gucci. Gucci was an early prototype customer for us, and that the prototype we built together resulted in in service GPT, but it also resulted in Gucci is client service reps being able to answer customer questions much more quickly.

But the fascinating thing was that after they resolved the customer issue, they didn't just hang up the phone or end the interaction, they were able to play a bigger role. They were able to become brand storytellers and salespeople and be able to look at the marketing engagement data of that customer and be able to talk to the customer.

Okay, we saw the issue with, you know, the broken handle on your suitcase. I see that you have our latest season handbag in your e-commerce shopping cart. Let me tell you about it. Let me tell you the story. The heritage of this bag, how it's made and crafted in Milan, Italy. And it's been amazing just to see this transformation in this client service center really into a concierge service for their clients.

Michael Krigsman: So, the initial use case then involves a very clearly defined result that you're trying to achieve, where you can very clearly see the ROI. Is that a correct way of saying it?

Clara Shih: Yeah, it's a quick win with immediate ROI, using metrics that we track today, like average handle time and first call resolution and CSat. And then the secondary effect is even bigger. Right. And it's and it was unexpected. We didn't go into that. They experiment with this with Gucci thinking that we would reinvent the job description of these individuals. But that's exactly what's happening.

Michael Krigsman: We have a couple of questions that have come in on this issue of the secondary larger effect. So let's jump first to Twitter. And this is from Arsalan Khan. He's a regular listener and he asks very thought-provoking questions. And here is his question. He says, how do you make sure that AI is not just another digital transformation under IT. 

If IT disrupts the business model and the culture, which departments should take the lead?

Clara Shih: It's not just it's job, it's not just the AI and data science team's job to utilize generative AI. It really becomes everyone's job starting with the CEO, his or her direct reports, the board all the way down to every individual. And they're this is a top down and bottom-up change management exercise, just like with the internet. 

If you're if you were a CEO in the 90s, you had one job which was to provide secure internet access to everybody in your company. It wasn't a block the internet. It wasn't the telling your employees that they couldn't use the internet to do their job. It was to provide that. And then as an employee, regardless of job function, in the 90s, it was your job to learn how to use Google search and to learn how to use email and online calendar.

But maybe you could go to a workplace training for that. But really it is incumbent on you to do that. We're going through that exact same 1998 moment right now with AI company leaders have to provide a secure environment like Einstein courseware for their employees to be able to use generative AI. Some of the use cases will be prescribed by the company and by it, like service for Pi recommendations.

But there also has to be room for employees to experiment and learn, and they're going to discover it. Maybe even more powerful use cases that that leadership top down didn't anticipate, because the employees closest to the work are best positioned to do so.

Michael Krigsman: What I find particularly interesting here is, of course, this shift into our AI period is driven by technology. But as you just described, the impacts are not necessarily, well, technology impacts their business impacts that have a broad footprint. And so, there is this very strong, cultural dimension as well.

Clara Shih: Always. Right. The hardest part about technology transformation is always the people, the culture, the change management.

Michael Krigsman: Please subscribe to the CXOTalk newsletter and subscribe to our YouTube channel. 

And we have another question on this topic from Greg Walters on LinkedIn. And he asks a few questions together. He says, when creating your GPT to human psychology and relationship experts contribute to the development. And then he goes on to say, do you see AI changing the way prospects make purchasing decisions?

And then the downstream from that is the AI replacing selling professionals? So again, he's asking about the broad implications here.

Clara Shih: The first question do we involve humans, psychology and relationship experts in the development of our products. We absolutely do. Right. Because the human acceptance and viewing of AI as a copilot partner is essential in making this work. And so when I look at what we did at Gucci, we involved our office for Ethical and Humane Use of AI.

So, they were a big part of the product and prototype design. And they're actually they're part of all of our product development at Salesforce, not just for AI, but all the technologies that we develop. And then the other critical, you know, unlock for us was we didn't we don't build our technology, especially these brand new product categories from the Ivory Tower.

We went to the customer site. We went to my client, we went to their client service center on, on the East Coast. And we sat down with the team, and we talked about what they would find most helpful. And they actually, we camped out there. We had a team camped out there for two weeks and every day at the end of the day, we get feedback.

And we would immediately update the prototype based on the feedback of what was effective. And so I think that co-creation process first built trust with the team on the ground, and then secondly, resulted in a product that really is useful to the people who were using it. So that's absolutely critical in the approach, especially when it comes to AI.

Michael Krigsman: The next question is how do you see the way I may change the selling process and the implications for selling for sales professionals.

Clara Shih: The typical sales professional today spends 70% of their time not selling. And again, that's just not that it's productive work, but it's not rewarding work. And it's work that frankly can be dramatically accelerated with AI. So if you think about every step of the sales cycle, right, starting with prospecting and figuring out who to prospect to, and then once you've got your lead list, how to reach them, each of those steps can be assisted with an Einstein copilot to really help make sure that that list is really good in the first place, and that the outreach to each individual lead is very personalized and has its high yield.

And then you think about preparing for that first meeting and doing research on the prospects company. If it's a B2B company or personal situation based on any, you know, a lot of time, like upfront surveys or questionnaires that are filled out, or lead forms, and then it's the meeting itself, how I can help coach the sales rep in real time.

It can transcribe and take away insights, interest, as well as action items from that sales call. And actually, every sales meeting that happens. And then afterwards, you know, if you if there's a bidder to head off, right, there's a a handoff there that the AI can help with is providing all that context as if that aid were in that initial meeting and so on and so forth, all the way down to helping prepare quotes and order the forms and, automating a lot of the, again, traditionally manual owner steps of the day in a life of a sales person.

And so this is going to be transformative already seeing that even with our own sales team. And then I believe the final question is what this means from the customers perspective. Well, the customer, especially sophisticated B2B procurement teams and customers, they're going to be using AI also to sift through RFP responses and to, and to help make the buying process more transparent and accelerate it.

Michael Krigsman: Clara, as you're speaking, the message that I'm getting here is that when thinking about AI applying to applying AI to a business process or a job function, the key is really to is to understand deeply what that function is, how that process works, to be able to then map the AI capabilities onto that function or that process, is that is that a correct statement?

Clara Shih: That's exactly right. I mean, that's certainly the first step. And it goes back to Clay Christiansen's, framework about jobs to be done. What are the jobs to be done today? And then the second stage is we're seeing what Gucci is like, what are the jobs that could be done if you free up and automate a lot of the current menial tasks?

Michael Krigsman: We have a question on Twitter from Jake Kozlowski, who says for companies looking to adopt AI, how can they do so quickly while also instilling trust in the technology both for internal employees and externally with customers?

Clara Shih: That is exactly the reason why customers work with Salesforce for our Einstein AI. So let me let me walk you through a typical journey that we that our customers are going on. So, you know, every company in the last 12 months has been told by their board that they need to not only have a generative AI strategy, but a plan that they're operationalizing and at least experimental way.

And so what a lot of companies have done is they went out and they purchased an enterprise version, of ChatGPT or, or Bard or, you know, pick your favorite large language model. And they're like, okay, well then what? Right. This is, what am I going to do with this? How do I prop up my data?

How do I get this into the hands of my employees? Are they going to have like a checkbook window open like that? There's like adoption is very spotty and the results aren't very good. And so, what they've come to us to do and what's so exciting is, is that okay? I want this quick win. Going back to our earlier conversation in Service Cloud, I want all of my customer support team members to be using generative AI with reply recommendations and case summaries.

Those are the two biggest bottlenecks in the day in the life of. I know that I'm going to get more efficiency and higher employee satisfaction and higher customer satisfaction rolling out these use cases, and they've been able to roll them out literally in ten minutes. Are you go to Service Cloud. You click a button, and you have your knowledge articles in Salesforce and boom you start getting knowledge article grounded answers drafted for you from service GPT.

And that's just that's it. That's this quick win. And then once once customers feel confident they they test it out with, they try to stump the AI with all kinds of questions like actually works really well. Then they say, okay, well, I want to I want to customize this product. I mean, these are this it's great that we had these turnkey prompts that have been developed by product managers in Service Cloud and in Sales Cloud and in slack.

And now I'll stop it. I want to customize it in my brand voice. I wanted these replies to be rounded, not just in my knowledge articles, but I want them to be grounded in specific estimate field and custom objects in my Salesforce instance. So, then they can open up or they'll be able to, on February 29th, open up Product Builder in the Salesforce platform and then Prompts Builder.

It'll show them the prompt template that's that was created for each of these use cases. So let's use reply recommendation or case some reasons for example. And then they can start tweaking and customizing the prop template. They can again put in their brand voice. They can reference any flow, any words field, any custom data custom object in their Salesforce instance like save.

And it instantly works. So that's this customization. So then step three is they say okay well now I feel inspired and I have a few more other use cases that my consulting partner or my SI has suggested for me that go beyond what Salesforce offers out of the box, right? Salesforce is a great library, with prompt templates, but I have my own ideas.

So then using that same prompt builder, they can create, test, validate, AB test and deploy any prompts grounded in any of their Salesforce data. So then the next step we see as customers say, well this is really great. This is this is just addressing a huge number of, of experiments and use cases that I want to deploy into production.

Now, I want to empower even more generative AI use cases. I want grounding not just in my Salesforce data, but I want the grounding to be on all of my enterprise data. I want it to be from my ERP system, from my phone tracking system, from my, you know, whatever web engagement system. And so that's when they use our data cloud.

And data cloud is a generator of data lakes, data warehouses, data silos that customers have in their organization across both structured and unstructured data. Data cloud brings all that together, unifies it, harmonizes it, hydrates it, and makes it ready for prime time for use in AI, either in training of models, fine tuning of models, and or in grounding of models.

So, then their Einstein copilot and their prompts have knowledge across their entire organization. And then last but not least, they'll say, well, I want to do I want to trigger flows and take action, not just in my Salesforce universe, but I want to trigger flows across my enterprise. And so then they sign up for MuleSoft and MuleSoft connects.

And now the Einstein Copilot not only has knowledge of the entire organization's data, but it can also take action across all the organizations, APIs and applications. So that's kind of the maturity model that we're seeing our most sophisticated customers, and I expect many more will be following in the coming year.

Michael Krigsman: So, you have a very well thought, well thought out maturity model and life cycle. How much of this is designed to drive efficiency or productivity improvements versus looking or enabling a real change in their business model so that it's not just a step change of efficiency, but it really is having an impact on how they run their business.

Clara Shih: It's really both, because you first have to enable the use cases in a secure, trusted and ROI, driven way. And then once you do that, you start to create more slack in the system, no pun intended. And all of a sudden your customer service reps who use to be spending 40 minutes per case and now are only spending ten minutes per case, they have more time to do more, to provide greater levels of service to their clients, to reinvent themselves as sellers and brand storytellers, to help out with customer success, to help out with cross-sell and upsell.

And then you really start to change the profile and shape of each department and also the margin profile of the company.

Michael Krigsman: And so do you see that actually taking place, that evolution taking place among your customers?

Clara Shih: It's very early days, but we're starting to see it take place.

Michael Krigsman: We have another question, again from Arsalan Khan on Twitter. Who wants to talk about the ethical boundaries and the responsible AI that you were speaking of earlier? And he says it looks like Salesforce is creating its own ethical boundaries of what is and what isn't acceptable in the AI, that you then use internally and with your clients.

And he's asking, so therefore, who decides what is ethical when it comes to collecting data and what not to use? Even if you have that data? I think it's this, this broad question he's asking about.

Clara Shih: There is a level between Salesforce, our customers, and our customers’ customers. Data is not our product. Our customer's data belongs to our customers. Then there is a secondary question of our customers' customer data right there. The terms of service with their customers. And so the way we've addressed this is really you think of it in in four layers.

At the very bottom is our technology and what the trust and the guardrails that are engineered into the technology itself. And this is what I was describing earlier as our Einstein trust layer. You can't turn off the zero retention prompts. You can't turn off the audit trail like that. Is it just part of, that that's how it would when there's a sensitive data field, we mask it.

So that is engineered into the technology itself. The next layer above the technology is our product policy, our acceptable use policies. And, there's they're out there. We publish them. But just to give you a few examples here, one of our public, one of our acceptable use policies is that when an end customer and I'm referring to our customers customer, when they're interacting with an AI, it could be a voice agent.

It could be a chat bot. One of our policies in the product is that it'll I self-identify as an AI, and so it's not a technical limitation. It's just something that from a product standpoint, we believe very strongly in. And so that's part of what it is. And so that's the second layer. And there's, there's a whole bunch more like that.

The third layer is around policies. And this is where, you know, every company has their own customer terms of service and their own policies like privacy policies and legal policies. We certainly do. What we've done here is we've open-sourced and we can share the link out on Twitter and LinkedIn later. But if we call it our ethical AI principles and they've been very thoughtfully crafted by our, again, a partnership between our office for Ethical and Humane use, our legal team, our employee success department, and our product team.

And this is something that we've had a number of customers and partners and be able to take advantage of, especially companies who might not have the resources to create their own. And then last but not least, we do have that office for Ethical and Humane Use of AI. And especially for, you know, our, our larger customer, as we've we've been able to consult, they've been able to consult with our office and to help them.

Michael Krigsman: So, you're providing guidance as well to your customers on the ethical and humane use of AI, in addition to developing the policies and then building those policies, building the product to reflect those policies.

Clara Shih: That's right.

Michael Krigsman: What are your customers experiencing with the shift, the mindset shift of requiring so much data in order to make effective use of AI?

Clara Shih: That is certainly the name of the game. And, you know, an AI that a company deploys is only as good as the data that goes into fine tuning it and grounding, to really again, you you mentioned hallucinations earlier. That's the way that you, you combat hallucinations is to provide context to the LM so that the LM doesn't feel like it has to make it up, pick it up on its own, and so it's this is a huge part of any AI strategy.

And again, why it's just so incredible to have our Data Cloud. I mean, the last decade has really been about how do you bring together structured data and unify it and harmonize it. And that's what CDP has tried to do. Customer data platforms. In the marketing world, Data Cloud was really groundbreaking because it provides that ability for not just marketing data and use cases, but across all customer facing functions, sales, service, marketing, products, engagement, and more.

And now we're seeing the equal need to provide this not just for structured data, but also for unstructured data. And you think about all of the fundamental transcripts from sales calls, from contact center interactions. All the chat transcripts are the emails back and forth with customers, knowledge articles, product documentation, slack conversations. There's a you know, most of the data in content within enterprises is actually unstructured, not structured.

And so, we're now building a way for companies to bring that together again, to unify it, harmonize it, be able to have it all in, in one place so that you can provide it to the LMS and as part of the grounding.

Michael Krigsman: So, you're developing the tools and the platforms that customers need in order to aggregate their data. Is there a mindset shift inside your customers with respect to understanding the data challenge that they actually face now because of AI and that demand for data?

Clara Shih: I think. So, I think any customer that tries to use, you know, generative AI out of the box without connecting it to a sound data strategy, quickly experiences that the answers are not very good and that there's a very high hallucination rate. And so I think that that is that everyone is kind of unknown, but a lot of companies are now developing that awareness.

And then when we show them the difference, when you connect in data cloud and what the output is, it just kind of speaks for itself.

Michael Krigsman: So, customers do then understand this linkage between if we have lots of high quality data, there's going to be, a clear impact on the results that we get from the LLM.

Clara Shih: That's right. It goes back to the phasing. I think customers know that they need to embark on you know, on multi months. It's not overnight. It's a multi-month data harmonization strategy. And then they need to kick that off as soon as possible. In parallel. They don't want to wait until the end of that project to be able to start deploying generative AI and getting business value from it.

And so going back to the most common example of all of what's being used in our across our our Salesforce GPC products today it's service GPT reply recommendations and case summaries. I mean, you've got customers how their knowledge articles in Salesforce and we can we just ground it. So, when there's a there's a customer question about you know my my wireless router isn't working.

And these are the steps that I've tried. And here's the weird lights that are flashing and that can you get instantly grounded in knowledge articles describing the same situation and be able to answer the question now, if for a lot of customers have more complex configurations and maybe their knowledge articles aren't complete or fully up to date.

And so that's where Data Cloud can really help.

Michael Krigsman: What about from an investment standpoint? You know, with traditional technology you buy the software, or you implement the software and you have a very defined result that takes place. Once you've accomplished those tasks with AI, you have tentacles that reach in more directions, and you have this whole data issue that we were just describing. So, given the more open ended nature of AI, how do the investment decisions figure in?

Clara Shih: It's very similar to the internet. You know, in 1998 your company and like I invest in internet access for all my employees. I invest in it in a PC or a laptop for all my employees, like, oh, that's really expensive. what's my first use case? Okay, I guess email makes sense. I can save on a bunch of, you know, written memos and faxes.

What's the ROI of that? So, there's that way of looking at it versus saying, okay, this is this. We need to do this. Some of the areas of ROI are immediately apparent to us, but the reality is we don't know what we don't know, but we know that this is going to be big, and we need to invest in the infrastructure to making this available.

And as additional applications become available or we build additional applications, we will continue to unlock more and more ROI.

Michael Krigsman: So again, you're starting with a defined use case, but keeping an open mind because there will be implications downstream that we may not see today.

Clara Shih: That's right.

Michael Krigsman: Arsalan Khan comes back again questioning this. The organizational issues. He says, okay, so we don't have an internet department going back to, when companies were implementing internet. So, the question is, we don't have an internet department in our organization today. Does that mean that we will not have a department that does AI in the future?

And does that mean that I become so, so spread through the organization that it becomes a commodity? Ultimately?

Clara Shih: Probably, yeah. Every application is going to have AI. And so the owners of those applications will have to become AI experts and users. But there's in the interim, right. It's all about how do you get to that end state. And depending on the organization, it can be helpful to have a center of excellence or an internal training function or, you know, any, any combination of efforts to kind of kickstart it.

And that's why I think we saw in The New York Times, there's an article last week about how one of the hottest new jobs is being a company is Chief AI officer responsible for identifying use cases, change management and deploying those use cases internally across the company.

Michael Krigsman: What about talent? There is such a shortage of talent right now. So how does Salesforce manage that? And also, what are you seeing with your customers?

Clara Shih: That's the real challenge is, there's not very many generative AI experts because that's so new. And so, you know, I think sales are really fortunate. We have a very strong, employer brand, and it's a place that people want to work, not just because of the business opportunity and career growth, but just because of our values. so we really just have world class talent.

And we've had a number of, of excellent AI hires, even in the last 12 months since I've been in the role. of course, you know, it's an ongoing challenge, and we have to always be vigilant and also proactive in scouting the very best people to join our teams. I think, but in parallel. Right.

Both at a company level as well as an industry level, there's not enough existing AI talent out there today for all of us to hire the people that we want. And so a big part of this has to be upskilling and retraining and seeing who has an extreme growth mindset. I had somebody join my team recently who had never done anything in AI, and now he knows as much as anybody, and the resources are out there.

Anyone can start using, even ChatGPT on a personal level. They can start calling APIs for, you know, Google's models and Amazon's models and, and open AI models. And, and there's a tremendous amount of resources out there for anyone who desires to learn.

Michael Krigsman: And what are your customers doing? Same kinds of activities as Salesforce, I would assume.

Clara Shih: That's right. Yeah. Everyone is very focused, is the hottest area of talent and probably the most important right now t0o.

Michael Krigsman: Arsalan comes back again. He's on a roll today and he's asking about this data question when collecting data, do you make recommendations on what would be the most efficient or the standard way of data collection that will be helpful for the AI, the AI? So, are you providing that kind of consulting for your customers?

Clara Shih: We do. I mean, we've got our professional services team, and that's led by my colleague Mark Wakelin. And then we also work with all of the systems integrators and consulting firms. And so this is again a big change management exercise that involves people. It involves data, change management. It involves, you know, testing use cases and then deploying them into production.

And so, it takes a village to make it work. And in terms of the data, how do the best practices? I think most companies today have multiple data lakes and data warehouses that are deployed to different departments. And so the name of the game is how do you bring that together again in a safe and secure and trusted way.

And so data cloud, for example, is not meant to replace Snowflake. It's not meant to replace Google BigQuery. We partner with all of these companies we federate in. Now, if you're a company that is very large, and I've seen some of these where you've got a BigQuery instance, you've got a Redshift instance, you've got a Databricks, and since you've got a snowflake instance, great Salesforce Data Cloud partners with all done, we have zero-ETL integration.

Michael Krigsman: So again, you're creating the foundational platform that your customers can use integrated with whatever their landscape is right now. Exactly I is changing so rapidly. So how do you see your customers managing the rapid? The rapid shift in the technologies and in the capabilities that those technologies and therefore the implications for their processes and for their organizations?

Clara Shih: It's very difficult, and I mean, even for ourselves. I mean, just to think about I mean, I'll just share what we've done at Salesforce. I talk about our customers. I think there's some interesting parallels. So, we saw this opportunity, you know, last December, I mean, we saw this report. We were experimenting with our own large language.

Salesforce research has been building large language models and publishing papers on them since 2018. So shortly after that work that Google did and so we've had our, our own models, and then we started playing around with the other models that are out there and we said, look, we need to get something to market, but that can't be the end game.

It's got to be this continuous innovation. And so we divide it up into horizon one, horizon two and horizon three. And horizon one is what I said we shipped last summer. It's these turnkey use cases for service GPUs. L-shaped, multi-commerce, TPC stack, GPT, etc. and we knew that customers would want to use something right away. And we did that all on top of our Einstein trust layer.

Horizon two we defined was how do we build out the shared services in a platform that, again, all the internal applications of our partners at Salesforce for each of these app clouds can use, but also all of our customers and ISV and SI partners. And that's what we're releasing in a few weeks with Prompt Builder Copilot in Copilot studio, and also Einstein Studio.

Einstein Studio allows customers to bring their own models and to fine tune our own models within Salesforce environment. And then horizon three is an ongoing set of experiments, right? Just like just like, the work that we did that resulted in ServiceGPT that we call that are super cheap. 

We've got, you know, a dozen or more of those types of experiments going on at any given time. And, and those show promise with our early development customers. We then graduate them into our existing products. So that's what we do. Customer. And I think I to do the same thing. Right. They they've got to start with a quick wins.

Then they have to think about a broader strategy of reinventing every function at the company. I mean that's, that's what my, my colleague one who's our CIO, that's what he's doing. He's partnering with every functional head at Salesforce, from our chief marketing officer to our chief revenue officer, to our chief customer success and customer support officer to reimagine each department function using AI.

Just as, again, people in those roles did in the 90s around the internet.

Michael Krigsman: I think what you're describing makes perfect sense. You start with something that is achievable. Demonstrate value, make it manageable but relevant, and having value and get some experience and some success and then expand radiate out from there, being aware of the possibilities and the implications that may exist downstream in a positive way, and being prepared for also unexpected issues that that may arise as you broaden out.

Clara Shih: That's right. It's a very agile approach that combines the best of top down and bottom up. And what we've seen is like when you when you see those initial use cases to people, especially if you co-create with them like we have with our teams and like we did with the Gucci client service team, then people like come up with their own idea.

They get really excited, and you create forums for them, like hackathons to partner with your developers, and then they get to test out their use cases, and those become the next set of use cases in that department. And then the next session, you do that every quarter. And before long, you know, you reinvented a lot of the jobs to be done in that in that function.

Michael Krigsman: We have a question from Wes Andrews who says that I think he's making two points. Number one is efficiently the efficient implementation of AI projects wherever it might be, because historically, software implementations have been really difficult for the customer. That's number one. And then separately, where are we in terms of being able to deliver really customized LMS solutions?

Clara Shih: For the first question, I think AI itself is going to be a very powerful tool and improving the quality success rate and time to deployment, time to the value of traditional software projects and that spans both non AI software projects and also AI software projects. I mean the classic the classic way that implementations fail or go off the rails is that usually it's, you know, a, a partner giving bad advice or someone giving bad advice and not understanding the out of the box capabilities of Salesforce, and thus over customizing and then having to maintain those customizations versus using something that just works, natively out of Salesforce.

Often, it's no-code. and then that being maintained by Salesforce and I mean, that happened to me when I was leading my startup, hearsay was we hired and we were trying to cost economize. And we hired a consulting partner that was an expert in Salesforce Sales Cloud, and we just misconfigured it. And we really actually just had to start over and take care of that.

I think I will be able to catch that much earlier and provide a lot more implementation guidance. I mean, Salesforce is also come a long way in providing better defaults in in the solution. I think one of the great powers of Salesforce is, is how customizing Splunk configurable it is. Think if you don't know what you're doing, it can also become a liability instead of having those preset defaults have helped a lot there.

And I think AI is going to help and we're already seeing that our professional services team is using AI itself to help with those implementation plans to help with the change management. And again, as we talked about earlier, so much of this is not just the technology and how it's configured. It's bringing everybody along. And so, being able to use AI to to generate better documentation and to solicit more input and feedback, going back to the importance of getting buy in from your users and human psychology, I think I can help with all of that.

Michael Krigsman: We have another question. an important question from Arianna Gaspar, who says, how can someone who is a beginner in AI start learning new skills to help stay ahead of the evolving workforce? Were there any upskilling trends that you saw with the Gucci team.

Clara Shih: In the workplace? It's really about having a growth mindset and, you know, using these technologies and and really thinking about, okay, how is this going to change what my goals are and how I accomplish those goals and how I spend my time, I think more broadly than than a workplace. But then being a participant in a workplace deployment, you know, I think it's just it's using these technologies on a personal level, and I, I use Bard, I'm on the PTA of my son's school, and I have to some from time to time, have to, write emails or create fliers or have a help with community events and party plan.

And Bard has been an amazing tool to help me do that. and so, I mean, just like with the internet, right. Like a lot of, you know, the consumerization of IT in the early two, thousands came from people using, Yahoo and Google and Gmail and Hotmail and expecting their, their work systems to be as easy to use.

And so I think a lot of it is just, is just being curious and starting to use these on a personal level and then thinking about how that what that means for your job.

Michael Krigsman: I think that's really good advice. You know, when I am writing things or doing research or analyzing things these days, I will use multiple AI tools and ask similar or the same questions each in order to cross-reference and correlate the results.

Clara Shih: Oh, I love that. It's like, yeah, consulting multiple experts.

Michael Krigsman: Arsalan Khan comes back, asks an important question, but I'll ask you to answer relatively quickly because we're simply running out of time. He wants to know about the the impact on jobs and job displacement. And he says, what about people purposely sabotaging or slowing down an AI project because they feel their job may be displaced?

Clara Shih: Well, the latter sounds like exactly what the Luddites did destroying factory machines. I don't think it's a good medium or long term proposition. in terms of job displacement, there's going to be job displacement of some level, but there's also going to be a lot of new jobs created. And this is exactly what the internet did, right?

The internet transformed entire industries, starting with media and retail. And there were a lot of jobs that were lost in the process, a lot of store closures. and there were a lot even more, a greater number of jobs that were created, for online commerce and, with influencers and, digital marketing managers. And so I think there is a role and responsibility at every stakeholder level.

Whereas I spent the week in DC talking to members of Congress, you know, at the government level, we have to reform K-12 education. And it's just that I know it's not easy to do because if we have to go state by state. But I look at what, you know, the kids are learning, and my kids are learning in school today.

And I don't know if that's it's preparing them to be successful in, in AI era. And so that is an imperative is we need to reform education. We do need to modernize the curriculum, learning how to code, learning how to use AI, understanding the shortcomings of AI have become as important as reading, writing and math. And so we need to relax our curriculum to do that.

So there's a role of government. And then for those already in the workforce being able to support reskilling and training for, for our citizens, very important. And at the company level, you know, there's things that companies need to be doing, right. Again, providing a safe, secure environment for employees to be able to use these technologies versus banning it altogether.

That's probably not smart. Is that a wise move for the business to not allow AI that's going to hurt their competitiveness? And then there's a role that managers can play, right, is to think about what does this mean for my function, for my employees. And then it's incumbent upon each individual employee. Right. In the 90s, there's no one who's going to.

This is going to make you learn how to use the internet or make you learn how to type. You've got how to learn and figure it out yourself. Right. So, we each have individual responsibility in using these technologies too.

Michael Krigsman: A great summary of the complexities and what we should be doing ourselves, which is developing our own sense of curiosity and the mindset that we want to go out and learn. Clara, any final thoughts very quickly before we finish up?

Clara Shih: This is a pivotal time not just in technology but in society. And I think it's, you know, we're having these dialogs, and this is this is going to change the world. And we're just at the very, very early days. So I'm looking forward to it and learning, looking forward to continuing to learn from all of you and partnering together to build the future.

Michael Krigsman: Well, this has been a very action-packed hour. I want to say a huge thank you to Claire Shy, the CEO of Salesforce AI. Clara. Thank you so much for being with us. I'm so grateful to you.

Clara Shih: It's been a pleasure.

Michael Krigsman: And thank you to everybody who watched, especially you folks who ask such great questions. You guys are amazing. Now, before you go, please subscribe to the CXOTalk newsletter and subscribe to our YouTube channel. We have great shows coming up. Check out and we will see you again soon. Take care everybody.


Published Date: Feb 09, 2024

Author: Michael Krigsman

Episode ID: 823