Why Most AI Projects Never Deliver ROI - A Salesforce CTO's Fix

AI Transformation or AI Theater? What CEOs Get Wrong About AI ROI

Most companies are investing heavily in AI.

But how many are actually seeing meaningful business results?

In this episode of the AI Visionary Podcast, we sit down with Gavin Barfield, CTO & VP Solutions for Salesforce ASEAN, to explore the difference between AI experimentation and real AI transformation.

From Agentic AI and customer experience to leadership, trust, adoption, and ROI, Gavin shares practical insights from working with some of the region's most innovative organizations.

If you're a CEO, board member, business leader, AI practitioner, or technology executive trying to understand how to unlock value from AI investments, this conversation is for you.

In This Episode

✅ Why many AI initiatives fail to deliver ROI

✅ The difference between AI transformation and AI theater

✅ Why LLMs alone are not enough

✅ The critical role of context in AI success

✅ How Agentic AI is changing customer experience

✅ What leaders get wrong about AI adoption

✅ The future of AI-powered customer interactions

✅ Why trust is the foundation of successful AI

✅ Real-world examples of AI delivering business value

✅ How leading organizations are preparing for an AI-first future

Timestamps

00:00 – Introduction and Agentforce World Tour Singapore

00:41 – How Gavin stays ahead in the age of AI

01:37 – AI Transformation vs AI Theater

03:49 – Why most AI projects struggle with ROI

05:36 – How Agentic AI is changing customer experience

07:53 – The rise of AI agents as customers

08:39 – How AI agents will interact with software differently

10:35 – The biggest misconception about AI agents

12:26 – Why context matters more than intelligence

13:39 – Leadership lessons for AI transformation

15:02 – Strategy vs use cases: where companies go wrong

15:50 – A real AI ROI success story

17:53 – How to identify high-impact AI opportunities

18:45 – Rapid Fire Round

19:01 – Gemini, Claude, or ChatGPT?

19:45 – The most underrated AI trend

20:48 – The most overrated AI trend

21:53 – Gavin's favorite AI tool

22:45 – Final thoughts

Key Takeaways

The companies generating the greatest ROI from AI are not simply deploying more models.

They're redesigning how work gets done.

The future belongs to organizations that combine trusted data, business context, and Agentic AI to create entirely new ways of serving customers and operating at scale.

🎙 Subscribe to the AI Visionary Podcast for conversations with world-leading AI executives, founders, researchers, and innovators shaping the future of business.

#AI #ArtificialIntelligence #AgenticAI #Salesforce #Agentforce #DigitalTransformation #BusinessStrategy #CustomerExperience #GenerativeAI #Leadership #Innovation #AIROI #FutureOfWork #EnterpriseAI #TechnologyLeadership

The AI Visionary Podcast is a platform where AI leaders share bold, unique and tangible insights about pathways to realizing meaningful ROI from AI, pitfalls to avoid and their success stories.

These conversations go beyond the hype to uncover what it takes for organizations to be successful in their AI transformation journeys from a leadership, talent, strategy and operating model perspective. 

Co-hosted by Joel Azariah and Milind, new episodes release every month via You Tube, Spotify, Apple Podcasts and various other Podcasts Channels

Links

LinkedIn Podcast Page
linkedin.com/company/the-ai-visionary-podcast
Joel Azariah LinkedIn Page
linkedin.com/in/joelazariah
Milind Linkedi...

Transcript

SPEAKER_00 0:00

Welcome everyone to this unique episode of the AI Visionary Podcast. We are on site today at the Agent Force World Tour at MBS in Singapore. And we have with us today Gavin Barfield, who's the CTO and Vice President Solutions for Salesforce ASEAN. Welcome, Gavin. Thank you.

SPEAKER_02 0:23

Great to see you here at World Tour, Agent Force World Tour Singapore.

SPEAKER_00 0:26

Excellent. No, we're really happy to be here, both Melan and I. And so let's kick off with something uh slightly uh more personal. Um, outside of work, what's the one activity you do that helps you perform better in the workplace?

SPEAKER_02 0:40

Well, I spend a lot of time reading and researching. I mean, the speed at which AI has gone. I in fact, I found a YouTube video that I did in in 2015 or so, which said the speed of technology innovation is like nothing we've ever seen. I was citing examples of Netflix reaching 100 million consumers in X amount of time versus the telephone. I look back at that, you know, 11 years later and think, what has happened? I mean, these things have had these things change in in weeks and months now. Um, so I spent a lot of time really researching to try to desperately keep up with what's going on, both internally in Salesforce, because we move very fast and in the market. Uh and I'm starting to build my own agent, starting to build and I'm playing around the moment with clawed code. So once the kids have gone to bed, uh I get that, I get there and I've got my command line interface, which I haven't used for a while. And then we're getting back to typing like the things when I first started out 20 years ago uh in Unix programming. But uh back to back to starting to build these things, but amazing to see what technology can do now. And and if we talk again in six months' time, I'm sure it's gonna be very, very different as well.

SPEAKER_00 1:37

No, absolutely. And I think just to build on that, right, guys. I mean, I mean, we see everyday you know companies calling themselves like AI first companies, you know, as a CTO on the ground and you know, here in ASEAN, how much of that is do you see as real AI transformation versus theater?

SPEAKER_02 1:56

I think I think that this the landscape has changed. If you look at the second generation of AI, the generative stuff that came in a few years ago, uh, a lot of that was technology looking for a business problem. And I think what happened is many CEOs and chairmen and business leaders saw products like Chat GPT, their their children, nephews, grandchildren, whatever came and showed them this technology. The companies that they were competing against were announcing we're doing this and this. So they gathered the leaders and said, right, how many Gen AI projects are you doing? Double it by next week. So you had this technology, and we worked around to try and find a problem. Hey, we'll Gen AI that will automatically generate that. And when you do that, you forget about the ROI and the business case and the real value. And then of course you have adoption issues. And part of it was also that this wasn't built on trust. You couldn't really trust the results of these of these uh generative AI stuff. You didn't know where the data was coming from, and they weren't in the flow of work. I like to use the analogy you may have heard before about a sat nav. You know, sat navs are great, but if the map is wrong, the underlying data is wrong. Or if that sat nav exists in the trunk and every time you get to a junction you have to check it, it's not really in the flow of work and it's not really doing the job. And also these agents weren't really doing, or these uh generative AI functions weren't really doing the work. They were giving you a suggestion, a draft, uh, and it was a bit cumbersome. What we've seen now with agentic AI is the ability, and when this is where Salesforce is working on that trusted data layer, that trusted uh agentic AI. Secondly, in embedded inside the flow of work. And it's not about bolt on, it's about embedding it in, and then also then having that ability to take action and to do things. So companies are now looking at these things as massive productivity gains and efficiency gains. And I think they're they're keen to do these, uh, you can get into this area, but with a very strict focus on ROI and a very strict focus on the business value. Because the last thing we want to do is leave more projects in pilot and POC land. They've had to explain these away for the past three or four months. So they need to make sure that they're delivering some value.

SPEAKER_00 3:49

So you make one very important point, Gavin, about ROI. And I'm sure you're having these conversations with C-level leaders across your clients. What's the one thing they get wrong when it comes to generating meaningful ROI from their AI investments?

SPEAKER_02 4:04

So, one area I think is that there's an understanding that LLMs are the answer. Um, and uh, you know, talking to um a company the other day who was sort of saying, right, Salesforce, what's your what's your token costs? I said, well, you're you're sort of missing the point, right? If we're really looking at ROI in terms of who's got the cheapest token, uh, I think that is that's not necessarily it's like trying to work out the cost of a cake by working out the cost of the sugar. It's just one element within it. Um, and and then I so I think if you if you're looking at ROI from those perspectives, you're not necessarily looking at it from the right from the right angle. We see LLMs as providing the raw intelligence and there's amazing tools. I mean, we're all you know blown away by what they can do, but the ability to then layer on top of that the business context, the data, to be able to do those in a secure environment, to be able to then work out how you embed them in the flow of work. And where we're seeing great ROI is when companies not just sit there and think about how agentic can affect the bottom line. It's quite easy to say, right, if I've got 10 people in my call center, 50% of those calls can be handled by an agent. I've only got five people, the CFO can do the maths and work out the ROI on that. But what's really exciting is when you look at how this new technology can actually change the complete way you work. Start off with a whiteboard and say if I had unlimited labor, if I were if I could scale infinitely, what could that customer experience be different to what I do at the moment? And how can that create the competitive edge? And those conversations around how agenda I can actually start generating value, start generating revenue, generating customer experiences are very exciting in addition to the cost-saving ones that we we hit time and time again.

SPEAKER_01 5:36

Bouncing off from that point, yeah, since you mentioned customer experience, what do you think would be the expectation of customers from their uh from the companies in terms of customer experience within the next six to uh six months to one year?

SPEAKER_02 5:52

I think the level of customer experience is uh that the expectation of customers is increasing. Uh, and I think a recent survey showed that well over 40% of customers are expecting much more from companies. And we've got used to processes and procedures, we've got used to doing things in certain ways because we've understood that companies are resource constraint. We've got used to the idea that when I call, I probably won't speak to a person. I will press 15 different buttons to try to get through to the person. And then I've got used to the fact I'll probably get a bit of hold music and listen to some pan pipes for 20 minutes. That customer experiences, we've got used to doing that. When I when I check in on my airline, I I know I'm gonna have to go to the NA navigate these things around. But I think now people are using agentic AI in the consumer world and they're starting to say, no, this is this is not the way anymore. Um, companies step up. You know, I shouldn't be if I want to change my address uh or my telephone number on my credit card, I don't need I shouldn't need to wait on the phone for 15 minutes until I speak to somebody. Give me the ability now to do this because I because I know that agentic technology is there. So it's raising the bar. Um, another interesting point, if you look at ASEAN as a whole, many companies, countries in ASEAN have huge populations and the ability for companies to actually service the number, you know, Philippines 100 and something million population. Some some of our companies that we work with have 50, 60, 70 million customers. How can you service that number of customers with humans alone? You get a scalability issue. And as a result, you segregate service. If you're a premiere member or gold card or you know, privilege, we'll give you a phone number where you speak to a person. If you are not, then you press one, press two, press three, and etc. And the exciting thing is the gen tech technology customers are saying everybody should have that access to that experience. And you've now got the ability to scale it.

SPEAKER_00 7:37

Can't wait for lowering the time span that you have to be on hold when you call customer service.

SPEAKER_02 7:42

Correct. I think you know, we talk about uh affecting jobs. One one job I hope it affects is those who create the the whole music, you know. But that industry, I hope we can get we can we can help that one.

SPEAKER_01 7:53

Actually, on that one, yeah, since you mentioned that you have created a uh EA and you absolutely enjoy playing around with it, yes, it's your assistant, yes. Uh what is very interesting, I came across uh some number from Gartner study that by 2028 we might have as many as like 15 billion agent customers. And if you have an agent, yeah, which is uh your executive assistant, chances are that this executive assistant will also do a lot of the things uh with the yeah uh that you want outside of just the work environment. Um how do you suggest companies uh or how do you observe what you are seeing? How are companies um adapting to this change in the customer behavior?

SPEAKER_02 8:39

I think we're gonna see you know, agents interact with software um in in different ways than humans. And traditionally, we've only interacted as humans. We only humans have interacted with software traditionally. And that's why we have UIs. And those UIs they've got better. I mean, when we all started, it was green screens and tabs and you know, commands slash and this uh that we're getting back to yet now with with uh what I'm doing with Claude Code. Uh but we you know we had this interface, and then it moved to you use a mouse and the UI gets prettier and easier to use. But ultimately it's humans using a UI. The way we interact with software is gonna change dramatically because you're gonna have agents directly interacting. So, and that's why we launched Salesforce Headless 360, because we recognize that there's two different groups of people who are gonna interact with Salesforce: humans, and they need uh a UI, although that UI will move more to being a conversational UI, voice and uh and a sort of two-way conversation than actually clicking buttons, and agents, and agents don't want to interact, or autonomous agents don't want to interact with um with AI using a UI. They want to be able to use things like CLI and uh and um MCP servers. Um, and interestingly, I think agents will do a better job of interacting with software than humans because you, you know, humans, however good your implementation Salesforce is, and we even in Salesforce, we like to think we're obviously one of the best in using our own product. Do you fill in every single field? Do you really go back at the end of the meeting and write detailed notes and descriptions of what happened in that meeting? Do you put every single piece of information in? Now, if you do that, products like Salesforce work very, very well because they have all the information they can make their decisions. When you skimp on that information, you know, then obviously the ability to utilize the features of software is not there. Agents will have no problem in doing this. Agents will summarize a meeting perfectly, they'll fill in every field, they'll make sure every box is checked and everything is all done. And I think that's going to allow agents to utilize software like Salesforce more effectively than what humans have been able to do.

SPEAKER_00 10:35

So do you feel, Gavin, you know, as the use of agents becomes more mainstream, is there something that companies or folks are getting fundamentally wrong when it comes to agents or deployment of agents?

SPEAKER_02 10:48

I think the to my previous point, an LLM is not enough. And uh, you know, raw intelligence comes from an LLM, but you need a bigger, a bigger sort of ecosystem around it. And one of the critical things you need is context. And context is to me what gives the intelligence in artificial intelligence. And you know, to give an example, if I have uh two employees, so one just starts today and I put them in a room and I give them a document on product A. And I take my best employer who's been with the company for 15 years, and I put them in another room and give them the same document on product A. And then I take those two employees and put them in front of a customer to do a pitch. Both of them have had the same input data.

SPEAKER_00 11:27

Yeah.

SPEAKER_02 11:28

One has got 20 years of context. They know the other products, they know how it works around here, what they should say, what they can't say. They've seen the presentations, they've done the, you know, all of that. The other one just has the context of that piece of data that they've seen. And obviously, we'd expect the one with the 15 years of context to be better in terms of you know, provide more intelligence. And I think that is true with AI. Connecting it into data is easy. You can point the AI to a data source. What's much more challenging is giving the AI broad context. And that's what Salesforce has offered from you know 26 plus years of developing workflows, industry standards, the data, the structured data, the unstructured data, stuff coming from Slack, stuff coming from other productivity tools. When you bring all that together, you create both structured and unstructured data, you create context. That's what agents need for the intelligence. So I think you know, one of the key things that, you know, misconceptions is just plug it into an LLM and plug the LLM into data. You will get a result, but that result might not be optimal.

SPEAKER_00 12:26

And also for, if I may add, I mean, the people who are using the agents, right? Like you said, the context matters a lot. So does like the change and transformation mechanisms that enable them to use these technologies well.

SPEAKER_02 12:39

And I think you know, when we when we first launched Agent Force and when we first uh other companies have launched it, it's it started off as a bolt-on, right? You have your product and then you bolt your AI onto the side of it. But we're now, you know, reorganizing, we we've renamed all of our products Agent Force for Sales instead of sales, cloud, et cetera, because the agentic has been built right in. We've gone right to the bare bones of it and redesigned the way we're doing these things with agentic technology and AI embedded into each of these different process steps, not you do this and then there's an AI thing at the end that might generate uh an email for you. How do you embed that within it? And I think this is where companies have a great opportunity, not just to think about, you know, okay, I need to stick a little bit of AI onto this, but start with a whiteboard and say, all right, with an unlimited workforce, with the intelligence being coming in from these AI, with the data I've got, the context I've got, how do I re-completely reimagine the process, completely reimagine the way I do things? And those things are exciting use cases to deal with.

SPEAKER_01 13:38

On that point, yeah, since you mentioned uh this uh this has a lot of impact uh on how uh leadership should be thinking about their organization, their processes. Uh what is it that you're observing in ASEAN or uh in your experience? Uh, how is leadership adapting to these changes? And where do you think a little more work uh is required?

SPEAKER_02 14:03

We we see tremendous excess of organizations that have a very top-down uh pressure on transforming. Uh and you know, if you if you see some of the organizations that that have that we've been working with, um companies like Singapore Airlines, you know, the the the CEO of Singapore Airlines has made a very clear statement that you know they don't want to just be the best airline, they want to be, you know, the the best AI energetic fueled airline in you know in the world. And he's created that as a great top-down direction. And you see that filtering down the organization and you see the the passion and the innovative nature and the ability to go into these areas. And so I think having that leadership drive is is is very important. And leaders have to sort of talk the talk, you know, walk walk the walk is not just talk the talk, right? They've got to be able to do things. Um that's why I'm trying to learn core code and stuff myself. I'm not trying to get back into uh you know developing, but I think you know, leaders have to be able to look and appreciate the technology so that they can understand the potential that it has.

SPEAKER_01 15:02

So, one of the things that you correctly pointed out, like you are uh one of those who's actually getting your hand dirty and saying this is different enough for me to try and uh learn myself. But uh once, even if uh the top leadership says that okay, we have to become the top AI-driven company, the next step, very often that uh uh I have observed is collect use cases from across the organization. Now there are two possibilities. Yeah. One is I collect all of these use cases. The second is I have a grand strategy. I set the vision that this is what my organization should look like as three to six years. What is your opinion? Uh, how much of the tool that you are seeing, and what uh what are the differences and outcomes that you see from this?

SPEAKER_02 15:50

I think you you have to set an overall vision and a strategy on where you're going. But I think many companies uh get tied up in that, uh, and the experimenting and the implementation and the speed to market, you know, becomes an issue. And that's what we're trying to offer in Salesforce, that ability for that speed to market. I'll give you a great example of a company I like to talk about, which is MaxiCare. MaxiCare is a company in the Philippines, uh, they're a healthcare provider insurance company. Um, so they were going on their agentic journey uh with us, and uh, one of the pain points they noticed was letters of authorization. So when you go for a dental appointment, um, you can get your check up, and then if anything needs to be done in a filling or whatever else, you had to get a letter of authorization because they've got to check against your entitlement, how much you've got in you know, your balance, and if you had that particular item covered by your insurance. That used to take about 40 minutes to do.

SPEAKER_00 16:39

Okay.

SPEAKER_02 16:40

So you can imagine you're lying in the dentist's chair, right? And suddenly the dentist says you need a filling, uh, and you've even got to stand it, sit there with your mouth open for 40 minutes, or you go outside to the waiting room, you hang around for a while, they're on the phone, and then you've got to get back in. Now we cut that process down to less than 20, 30 seconds because there is an agent that's simply the the person in the room can can log on, can jump this agent, can give details of the person, their healthcare number, uh, what the procedure is to be done. It can go through, it can check the entitlement, it can instantly offer that letter of authorization and send back. So is it life-changing? Well, no, but it's a it's uh it's important if you're lying in the dentist chair. And if you think about they've got that one, that one success, and now they're thinking the next one, the next one, the next one. And as you're proving the ROI and prove the agentics stuff, having those quick wins and having those ability to show, actually, this has reduced 40 minutes to 30 seconds, and it has a tangible effect on customer satisfaction and it's helping people out. And once you do two of them, three of them, four of them, ten of them, twenty of them around your organization, it's these small little improvements where you're embedding AI that you'll start to see a big change.

SPEAKER_00 17:46

So now the only pain they experience is because of the dentist, nothing else.

SPEAKER_02 17:49

Not because not because of waiting for the uh waiting for that one.

SPEAKER_00 17:53

So in that example, right, Gavin, what do you feel was the key element that helped to realize this benefit? Was it because of the management they realized, okay, if we focus on this and we we make a change, it'll have a significant impact?

SPEAKER_02 18:06

Yes, I think it's because they found a use case that is a real use case, uh and they found a pain point, no pun intended, within the within the process, all right? That that they said this is this is and and they found a uh a tangible benefit, both in terms of the dentist himself being able to speed up the way that he's able to treat patients after bringing people back and forth, but also to the customer, and then worked out that this was that that this is a tangible use case, and then they got going on it. They got implementing, they took a simple enough use case, they built it, they they tried it out, they got results, and they moved and they're moving on. So I think that was the big difference, the ability to to actually just get out there and and try some of these use cases.

SPEAKER_01 18:45

So this brings us to the last part of our conversation, yes. And this is something that we have not shared with you already. Yeah, we call this the rapid fire section.

SPEAKER_00 18:54

Okay. So, first question Gemini, Claude, or Chat GPT?

SPEAKER_02 19:01

Uh I use a lot of Gemini uh and uh you know for for a lot of the the research that I'm doing. I also use a lot of Slackbot. Uh and Slackbot has been uh transformational in the way that I've done my done my work. So Slackbot is uh you know embedded inside Salesforce because it's connected to all that trust trusted data. So I use Gemini a lot uh and I use uh Slackbot every single day for uh for. In fact, this morning I was showing a customer I was taking a bit of a risk by by by sort of saying I'm preparing for a meeting with this customer, tell me all everything I want to know. It even pointed out that him and his son enjoy Lego. So it gave me such, and he was like, How did you know that information? So it's been a fantastic tool for us to be able to uh prepare for meetings, understand what's going on in the business. Excellent.

SPEAKER_01 19:46

Most underrated AI trend.

SPEAKER_02 19:49

I think we're gonna see, I mean, we're gonna see an evolution of the personal, the personal AI, uh, but in a trusted way. So I've um I've decided, well, you can't get Mac minis for love nor money around here anymore. So I decided I wasn't gonna go and buy one and build my open claw or whatever I wanted to do to build my agent, because for me, the trust is important. And unless I'm gonna be able to link these agents in to trusted data, and I'm gonna trust what these agents do, I don't want to give this access to my bank account, to my to my to my personal stuff. So I think we'll see the rise of those personal agents, but they've got to be done in a way in a different way where trust and and and and then the sort of security falls behind it. I'm much more comfortable in using these AI tools like Slack and Agent Force inside Salesforce because it's connected to trusted data. I've got the security, I've got the whole surround, you know, ecosystem surrounding it. Um but interesting how I uh how these things which I have in my work life are now gonna start coming into my personal life eventually. Next one, Gavin. Most overrated AI trend. Over-reliance. So maybe people are over-relying on AI and taking out the human touch. I think you know, with that where there's still gonna be things in my life that I want humans to do. I think there's gonna be a balance between what I'm happy for AI to do and what I'm happy for human to do. And I'll take a banking situation. Am I happy for an AI agent to change my um change my billing address or change my telephone number? Sure. Quick transaction, I want to be done. Do I really want to, if I'm taking out a mortgage for a large amount of money, do I really want an AI agent to handle all of that? Maybe bits of it. Maybe I'm happy for the AI agent to handle the document processing, uploading of all the things. I want a human involved as well. I'm making a big decision. I want a relationship between a real human who's gonna talk me through these things. I want to see someone face to face. So I think the idea that, you know, the whole world is gonna be run by agents and we're gonna have we're gonna put agents in charge of everything, I don't think is the case. We're gonna find particular things where agents are useful, but humans are still super useful and and we still need that that human touch certain for certain transactions we want to do as well.

SPEAKER_01 21:54

Okay. Your one uh favorite AI tool.

SPEAKER_02 21:57

Well, of course, Slackbot, I would say. I mean, in all in all seriousness, I mean, I'm using this 24-7. Uh, it has totally changed the way that I'm I'm working. I'm I'm it's I go into this thing, I ask it to prepare meeting notes, I ask it to give me all this information. Uh, it has an amazing ability to go across structured data inside Salesforce unstructured data. That one I said has single-handedly changed my life. Um, outside of Salesforce, I'm I'm playing a lot with um video generation. I do a lot of presentations and various other stuff, and I'm finding that some of these ability to create these um these AI avatar videos uh is quite useful and these technologies moved so you can get very, very realistic stuff, and it's uh that's a bit of a habit to play with. But I I think Slackbot is is has been a real a real life changer for me in my work environment.

SPEAKER_00 22:46

Excellent. Great. We're we're at the end of the episode, Kevin. Thank you so much for this wonderful chatting with you. It's been great. And thanks to everyone who's uh following our channel. Look out for our next episode on our YouTube, Spotify, and Apple Podcast channels. Thanks again for joining this episode of the AI Visionary Podcast. Thank you so much. Thank you. Thank you, thanks, Kevin.