Why 90% of AI Agents Are Stuck in a Pilot

This is Executives Unplugged. I'm Keith Cowing, your host and executive coach, and with me today is Akshit Kandi. Akshit is the CEO of SkySync, a New York City-based company that implements, deploys, and manages AI agents for companies of various sizes. Akshit has a terrific background mentoring 100-plus startups and working at Salesforce on the Agentforce platform.

And today, he helps companies implement these agents and manage them just like they manage employees. And on the show, he will walk us through four key moves that you can leverage at your company to deploy and manage AI agents and to handle the organizational change required to make it successful.

Let's jump in.

Keith Cowing: You say right on your site that most AI projects never pay off. Why is that, and what's the gap that people are coming up against?

Akshit Kandi: So most AI projects, From our experience right now almost ninety percent of them are kind of stuck in a pilot, And the core problem in the landscape today is you know, there's always, kind of human silos in the picture and even maybe a perception of what people think an AI agent should do and how, an ROI should come out of it, I would say that's kind of like the core, like two things that people really are stuck in at the moment, In order for your AI to work, right, obviously you need your data, your systems, and, you know, processes to be, in harmony, So that's really like the first step, right? And in order to achieve that, your different departments within a company need to be in harmony as well.

And most of the time, the departments are not in a harmonized manner at the moment, right? So what we come in and do is really focusing on breaking those like human silos, right? And, you know, keeping the CIO and the CEO, and maybe even marketing and any other stakeholders that you are thinking about deploying an AI agent for, in the same room and actually align on the systems, processes, data.

The second step after they figure out all of this, they think about AI agent as a one-time thing that they would just deploy and start getting ROI on, right? But what we found out is AI agents are almost like employees?

Think about your best employee in your company and what you're doing to them, right? You're probably doing, weekly one-on-ones or, aligning on the job description, maybe even like goals if they're, you know, really like execs right now. Every agent in your company needs to have a proper, jobs to be done set up. And ultimately, you need to focus on the outcomes of each of these agent. And as if you would manage your employees, you need to make sure every day you are actually like checking if this person or this AI agent in this case is actually getting you that level of outcome that you initially set in?

If it's not doing that, you have to actually do the same thing what you would have to do with a human, right? But a little bit more nurturing probably is required initially? To make sure you actually come in and, resetting the goals or giving additional context or help, or in this world of agents, maybe it's additional like data context?

And that's really how you would go from just thinking about an AI agent, which is a pilot phase, which is ninety or maybe even ninety-nine percent in my experience is pilots at the moment to really getting that ROI. We actually been working with a, you know, a green energy customer. Y- you know, they have a lot of amazing like use cases and customers, We found out that they're actually you know, not even, you know, going about expanding their online like lead gen, right? So for instance, what we actually did is we actually took care of their entire marketing?

For instance, we're running their marketing with our AI agents, and that's a business process that was, underperforming in our opinion by comparing to all our other customers.

So that's kind of the first step, right? Really doing that discovery work and spending a lot of time before anything is built out? But after that, we really start unlocking value immediately, right? So we obviously like put out a roadmap of when, you know, certain things will happen and, you know, when your AI agents can come into place.

But most importantly, I would say is like the iteration and making sure the team is also comfortable with this kind of move. A lot of the times what's happening in the market today and the teams, right, like we turn on an AI agent and they're just not ready for it, right? So it's really making sure we go in and educate them on, you know, how to, you know, work alongside with an AI agent, right?

And not think about an AI agent you know, sometimes as, y- you know, kind of even a replacement to a person. You know, that's definitely like one of the things that are happening, in the market, and there is definitely places for it, right? But there are places where an agent could be very powerful assistive agents as well.

So- it really making sure they're comfortable with the processes and the education around it I think is very crucial, right? Because we launched this marketing agent and you know, we put this exact plan together for the company, and, we got amazing outcomes. They went from almost like no business through online social media marketing through going over like maybe like three hundred percent increase in business because of, our AI agents, right?

It's extremely powerful, and it's also, right, not only getting the top of the funnel leads. What we also did was like reducing like the time that they were taking to reply to customers, right? AI agents, when done well, are extremely powerful. In this case customers were just getting replied within two minutes, You know, they dabbled around with running marketing, but maybe there's a performance or, you know, feedback loop which is, which is missing, right? and iteration is missing as well. So we came in and took care of that entire like pipeline as a process, right?

But in our case, we actually achieved more than like twenty, twenty-five percent of conversion rate. Again, like tying back to our ROI story, right? Not only we expanded their market for this company, right? But we also helped them really achieve that ROI, which ultimately what I am focused on in my business, right?

I want my customers to get that ROI and that's really how we did this and advise the customers.

Keith Cowing: So two things I'd love to dig in on there. One is the marketing piece, and one is when you talked about making the team feel comfortable. So on the marketing side, how did you discover that you needed to tie marketing into this, and why was it important that you take responsibility for that as well?

Akshit Kandi: So that's really how we're able to obviously like identify the business problem, right? And if you think about the training of the employees and making them comfortable, right? Really mapping out who are the stakeholders, right? Within this operation, right? But beyond that thinking about how do they map to the ultimate like goal of, you know, maybe increasing revenue.

And we actually come in and when we're doing the agentic, transformations, we are thinking about how a person's role is maybe going to change a little bit, right? Because change is hard, right? But if you are doing an AI transformation, change is crucial, right? This is very important, and I think many people do not take this, as a fundamental approach of going about, an AI transformation, right?

And it's not talked about a lot, to be honest, right? So we actually come in and…

Keith Cowing: Change management is one of the hardest things here. It's,…

Akshit Kandi: It is very hard.

Keith Cowing: Are harder than the tech systems…

Akshit Kandi: Yes. It is extremely hard. And one of the things we do is put a plan together for every individual person and tell them and coach them, "Hey, this is how your day-to-day might change because of this new AI agent," right? And…

Keith Cowing: Think this point is really important where you're talking about making them feel comfortable, and there's a lot of fear and uncertainty in the air. And fear and anger drive the worst business decisions of all time, and fear is the one that tends to be around a lot right now, where people are uncomfortable about the changes, the world is chaotic, there's a lot of things happening. And so you have to have teams with a confidence and a swagger in order to take the risks that are necessary to make the change.

The riskiest thing is taking no risk right now, so you have to try this. But it's uncomfortable, and it's weird, and it's different, and it's very reasonable that people are not really sure. And like getting them to that level, that's about sitting down one-on-one and walking them through what the changes could look like and seeing what their discomfort is and coaching them through that.

What does that look like?

Akshit Kandi: It really starts off at the highest level, right? Change always needs to be accepted by the leaders and it's almost as a mindset, right? If we're going in and doing a transformation the first thing we do is making sure the execs are really comfortable with, this type of transformation.

Because it is sometimes gonna be very uncomfortable, right? if you don't have this conversation, right? And, you know, by experience, we are getting alignment from execs. But also I would say it's very important that the company is, installing in that mindset of, "Hey, we're actually going to be part of this, like, next era of companies which are going to be doing the AI transformation, and we're actually going to get amazing output from it," right?

And it's a growth strategy, right? Versus just like a cost optimization situation a lot of the times. So that mindset has to be, aligned before we dive into the day-to-day operators and how they would use it. But really…

Keith Cowing: On that, on the mindset, what's the gap that you see between the companies that are really effective at installing that mindset throughout the org and the ones that aren't quite there yet that are struggling?

Akshit Kandi: You know, if I'm to define that successful company what it feels like is the leadership is really eager, right? They're really eager to learn about AI. They're not looking at it as a risk. They're not looking at it as a threat. And they should have that almost, growth mindset in place.

But there's the other type of customers where they're failing in our opinion, is, they are these rigid firms which have rigid processes, right?

But the ones which are not successful have that notion of, hey it's maybe going to be a threat to my company at some point. So they're not really looking at it in a growth mindset. And I think that's really the key, right? And it's mostly, I would say, in the hands of the execs at the moment, right?

And it's a decision that they really need to make. ultimately it boils down to that. But, everything below it, right? Because you are setting that company vision and that goal, and you are communicating to your team things get much more simpler and easier than when you're thinking about it that way…

Keith Cowing: Yeah, that resonates a lot with what I see as well, where it's offense versus defense and the execs that are saying, "Let's play offense. This is the opportunity of a lifetime. It may mean our company needs to look different. It may mean we need to change the way we play the game 'cause the rules of the game have changed, and that's hard, but let's figure it out 'cause wow, there's never been a time like now," versus the ones that are scared.

And it's that creativity and it's that ambition and the hunger and the curiosity that tend to lead towards, the companies that find a way and the rapid change man, there's big unlocks that are happening. So it's an exciting time. It is scary. We can't pretend it's not, but it's really exciting if you have an offensive mindset.

Let's talk about technology a little bit. So when you come in and you talk about working across the different functions, one thing that's really interesting is you can dabble, like you said, on a solo basis just as a single player use case or on a single function use case. This is how product is using it, this is how marketing is using it, et cetera.

But some of the biggest unlocks in my experience are going across these different functions and require technology decisions that are broader across the company, be it, you're using Salesforce or Agentforce, or you're putting all your data in Snowflake or Databricks and you're using that as a repository.

These sort of cross-cutting things where the context layer within the company can be shared across the agents and the orchestration can be shared so that you can accomplish these, let's call it multiplayer use cases or at least multifunction use cases that are very powerful. So when you work with companies today, you spent a lot of time at Salesforce, you worked on Agentforce, now you help people implement and use Agentforce.

You also work with OpenAI. I'm sure you see a lot of data lakes and things like that. What are some of the key tech decisions that need to be made centrally in order to unlock the biggest opportunity here?

Akshit Kandi: The way we are looking at it again, right, it's, very similar to you just said, right? We are finding the biggest opportunities where it's not just only focusing on a specific individual task versus it's focusing on an end-to-end outcome, and you are given almost like sub-agents to kind of work towards it, right?

So I wanna talk a little bit about this, right? There's a super agent which say you are focused on marketing, right? And then you have your sub-agents, But where a productivity is really gained is like going into that and tapping into other teams as well, right? If you go beyond marketing a lot of the data science is just sitting in Snowflake, Databricks.

It's a soup of data that is available for people to just go in and tap into and unlock like tremendous amount of value. You have your system of records like Salesforce, where all your customer data is sitting. There's obviously like an inherent, still an inherent data unification problem that most companies have not really achieved.

One of the most powerful internal AI agent that we build for our customers, is ability for a specific business user, say like a sales exec, to just get a three-sixty degree view with a one single prompt.

What's happening in my business today across the board?

Keith Cowing: That's the Holy Grail, right? People have been trying for decades with BI tools, eh, decade after decade have been trying to achieve that. And it may be possible that we're now almost there…

Akshit Kandi: It does come with a little bit of thought before you actually deploy this, though. Governance is a huge factor by the way, Keith, because we're talking to customers across the landscape, right? The bigger customers especially have to be very careful with MCP servers and pilots. What we're seeing why they're also failing is wrong information is being sent out at the wrong place, right?

And it shouldn't be happening. And we actually take care of the governance process, really making sure the agents are assigned to the right persona within your enterprise, and especially if it's external agents, the right data is allowed to go to the customers, right? And it's really crucial to take this a little bit with a thought and again, right, this has to be exercised, workshopped, and who gets what type of access to what type of data.

And but done right, we're seeing tremendous, like, decrees in swivel chair, right? Where personas of people, like salespeople have in the past had to rely, you know, reaching out to, their data science team or rev ops or maybe some other team to just get a piece of information. And that's right now decreased from like almost sometimes like a week's time to just like seconds, right?

Keith Cowing: And if a team is trying to set up the right governance, what are some examples of questions that you ask that help create, "Okay, here's your governance playbook"? What questions should people be asking internally?

Akshit Kandi: Yeah, the first question I ask them is if they already have a governance, external like governance management like system. Large enterprises do have this, right? It's a little bit different than an MDM or a data management platform as well. There are specific ones but in our experience, they'most companies do not have this system.

And what we almost have to do is really talk to their, CISOs, right? And understand what systems are in place, what are the policies around like data sharing and visibility throughout their company, right? And, a lot of the times there are even holes there, right? There's always holes in every customer we talk to.

And when we're thinking about the agentic transformation and what, which persona should have access to what data it almost goes down to try to maybe even you know, simulate what they already have or a lot of the times just going in and advising them, "Hey, this type of persona in our experience has this type of data."

For instance, individual sales teams are not required to understand like sales data across the globe because it's a GDPR like compliance issue, right? And data residency is also an interesting one. So we actually try to understand which all countries do they have a presence in, right? And a…

Keith Cowing: Lot of, say data residency, you mean the data from individual customers staying in their country or region?

Akshit Kandi: Exactly. So GDPR is a huge example. Even India, where we have customers, there is a requirement, certain type of data cannot leave the country, right? It has to be stored in there. But there are countries where, for instance like the US largely speaking in the US, there could be like data, be some part of the data could be sitting in some other country as well.

And my job, the most important job here is you know, really not try to disturb something that is working for them, right? It's very crucial, especially if you're at that large scale, say if you have more than five thousand employees, right? To not really disturb something or try to fix something that is not broken, right?

But what we do really do is try to understand like how these processes are already working and come up with a solution a generic solution, right? And break it down by personas, right? There's role level access or security policies. There's you know, object level security policies that needs to be set out for each of these persona.

And it looks somewhat different when you have unstructured data, right? Because unstructured data you know, traditionally, right, it was sent over an email or a Slack message, or it's stored in a Google Drive. And, each of these policies are kind of different for each of the different systems, right?

So a lot of our time and our team spends to analyze these different data systems and understand and try to map what type of security policies we need to think about, right? And say the internal employee agent use case, right? we are making sure and testing it on all those security policies before we roll it out because it could mean a breach if we don't do that.

Keith Cowing: There's three themes that are coming through to me very clearly from what you're saying. One is mindset, starting at the exec level of the offense of, "Hey, a wave is coming, let's ride it," as opposed to, "Let's try to hun- hide underneath it and try not to drown." The second one is the data unification of doing work internally so you can put your data in one place, the agents can really run wild on top of it.

And then the third one is governance, to make sure that you have the right controls in place so you can actually do this successfully for, meaningful business and working with CISOs, et cetera. And then one piece we haven't talked about is the actual business workflows themselves. And a lot of times codifying what somebody does so that you can automate it is actually the hard part of just figuring out like how do we take what you do every day and actually turn it into something you can put on paper and make it repeatable, either so that we can take parts of it and automate it, or so we can supercharge you so you can do a lot of other things.

Talk to me about the codification of the business workflows and how much of that is what they do today, how much of it is saying how do we just rethink this completely of what this role could look like?" How do you get those business workflows down so you can build successful agents?

Akshit Kandi: So there's two things, right? It's, what is already possible to achieve with a high confidence in the market already, right? So that plays a huge role. And for example, in this situation, we're already looking at systems and processes, right? Say for instance, like marketing, right? If you have like cracked down all these like different jobs that your company's already doing, and if it could be with a high confidence be replaced by an AI agent, we would highly suggest you to go in that direction.

And there are the second things where human still needs to be in the loop, right? For instance, right, AI is really good at, you know, summarization, writing great articles with enough context given to it. so in almost this like today, right, like SEO is almost could be replaced by an AI agent, and we are already doing that, right?

It's completely replaced. But if you think about where you still need a human in loop, AI is great at generating images and content, but you need a human to come in and give some content, and you need approval. So we are building these human-in-loop processes, right? And when we do our exercise, we are mapping out what processes could be completely automated and could be autonomous, right?

And where is a requirement for a human to come in the loop and really give their feedback and approve it. So But, what I'm really, I think, seeing where the ultimate effect is in all of this here is we are having a super AI agent on top of all of this, right? So super AI agent for marketing, kind of covers and goes in and gets input from all the different AI agents, like running ads SEO, AEO, like, content creation as well.

And it is also getting feedback from the human-in-loop, right? So when the AI agent is executing, marketing AI agent is executing and try to continuously like send feedback every morning. I'll, I'll walk you through what happens today. Every single morning my internal team gets a brief from our AI agent, which overlooks marketing for all our customers, right?

Across the globe, and we get a brief of what we should be looking at, and it gives you tips of how we can improve it. so it's very important to think about, these two things, right? Defining what tasks can be completely automated by AI and thinking about the task that needs a human-in-loop and where escalations are happening.

Keith Cowing: And where do you recommend people get started? You've said marketing a few times. That's one example. If they have the mindset, they're starting to work on unification, not perfect, but they're at least aware that they need to do it, a little bit of governance in place, and you need a starting point. A super agent so that you can tie it all together and see end to end how it works.

You can start to build the muscle. How would you advise somebody to figure out the best place for them to start?

Akshit Kandi: My suggestion at the moment, the best place to start is internal agents, right? Especially if you're just starting off in the AI journey pretty new as a company. The most productive thing is to harmonize your existing data and putting an AI agent on top of it. I think that's the number one thing that I would suggest, which is gonna get you highest ROI and save you a whole bunch of time.

Keith Cowing: What are some examples That agent could do just so people can really understand tactically?

Akshit Kandi: Yes the internal AI agent can be simplified as simple as, hey, I have ten different data sources which my team goes in and spends some time in pulling in some data. Could be your Snowflake, could be your Salesforce, could be your Asanas or Jiras or any of this information, right, which is all over the place, and your team is going in and maybe updating the system to keep continuously keeping a record, for instance.

One of the biggest use case, right, with Salesforce is and the sales team is, I have an agent that can go update all my notes. It's connected to my Google Calendar where notes are being taken, and you simply just say like, "Update my notes," and it just goes and does your notes for the entire week.

You don't have to spend like roughly like ten hours or maybe even like five hours, like because every single day you're going in and like updating certain things in the CRM. So it's very simple in that situation, right? And a lot of the times this data also has to be reported somewhere to a certain dashboard, and it needs to go be pushed over there.

So while your employees are doing all of this there could be even on the back end things going and automating things and workflows happening on behalf of, people like triggering an agent as well. So internal agent and that type of use case, right, wherever there's data unification required and data retrieval or data entry required you know, that is the most powerful AI agent I would say at the moment.

Keith Cowing: And one thing that I think you're pointing at is you already have systems in place, and if you can connect to that, then just put everything there. It'You don't have to reinvent the complete wheel. If you have a Salesforce, if you're really heavy on Atlassian and you have Confluence and you have Jira, if you have data in a place already today, you can re-architect what perfection looks like, and that's great.

Maybe you try to get there. But also, if you just have a thing and you just put more in the thing and you just keep it all there, it, that's very powerful because, Salesforce has great APIs and so do these, all of these data lakes, and so does Jira, so you can pull all your context out of there.

And then another thing tied to that is I see a lot of value in the meeting transcripts when people are transcribing meetings and in presentation materials that they bring to meetings, 'cause there's so much rich summarized content in those discussions that happen in rooms virtually or in the real world that can be fed to these agents if you're having some kind of unified, set of guidelines.

What do we record? What do we not record? Where does it go? What are you seeing in terms of content from meetings, be it presentations and the transcripts, in terms of how people are successfully leveraging that?

Akshit Kandi: I'll give you an example. One of my customers, right, they have a different accounts receivable system, and they have Salesforce for their record and then they have Snowflake which kind of gets in maybe information from their product, right? And, you know, meeting notes happen as well for them through, say, like Google Meets, right?

And it's very interesting in this situation. You know, one of the most popular kind of automations, right, we have created is almost like have an agent which goes in and reminds like, say, employees to go nudge customers on accounts receivable and, vice versa, like accounts payable also, right?

You know, there's use cases there. But thinking about the email how they would have generated this particular email in the past, they would have to go in manually look up in their CPQ system and figure out like, "Hey, a customer is actually not paid." They have to go maybe, you know, draft that email manually, and then they also have to go update the Salesforce CRM.

And usually, they're also, like, getting on a call to talk with this customer to do this, right? Because now agents are, like, synced up to all of this you can automatically have you know, that personalized email generated, right? Especially for high-touch customers. There's automations and everything, but sometimes, our customers, you know, they still have to nudge their customers to kind of get money back.

And this has been very, like, time-saving because they can generate, like, very personalized emails, which is getting in data from everywhere and also updating, like, records and systems at the same time. And, it's up to date of what it should you know, kind of type in because it's also connected to your last customer call that you had with the customer.

Keith Cowing: That's awesome. And given that, I think I might re-articulate part of the key messages that I'm hearing to say, number one is mindset, and make sure that you're just excited about this, you're leaning into it, and you're getting everybody jazzed about it. The second one is the data unification. The third would be the governance.

And the final one is pick an effective starting place where you can have a super agent and then build around it all the sub-agents. And if in doubt, just do an internal one, which can be really powerful, and then you can start going external. And I think you gave some really interesting examples and tactics that people can use.

Is there anything I didn't ask that I should have?

Akshit Kandi: The most important thing is, again, the ROI, right? Having a dashboard where you can see all your spend for your AI agents and the sub-processes, right? It's also very important. The credits that you burned, if it's ads, it's like the ad spend. If it's you know, procurement, like the money that was spent, everything you can think of has to be put on a dashboard, and ultimately, you need to look at it as an ROI game, right?

Keith Cowing: This is token measuring, not just token maxing…

Akshit Kandi: It's very important. Yes, exactly.

Keith Cowing: Okay. And final question for you. You could build this company anywhere. Why are you building in NYC?

Akshit Kandi: We met in a room where a couple of weeks ago where we had the best people in New York. The energy, in that room is why I'm still in New York. I've been in New York for more than a decade. I mean, even the Knicks have won this over the weekend and, everyone probably saw like the energy and, you know, most of the people I talk to they just dream about feeling this energy.

I was talking to my friends who are y- you know, other countries or even other cities, SF, right? They're just calling me: "Hey, like, what are you doing this weekend?", and it's not just the Knicks game, right? It's just one example where, you know, people just realize, hey, like the energy is just wild.

But, going back to our time in that room, that energy still existed, right? I would say we have obviously like the best operators in the world, right? We are, you know, not only thinking about innovation, but we are also thinking about, I guess, like ROI, right? Because inherently New York is, you know, very financial heavy, but we also have a lot of amazing, like tech talent.

I was just checking the numbers, right? From the New York Tech Week. I think we had more than three thousand events that one week. And apparently, like we had like sixty thousand attendees, right? That's like triple the size of what, SF Tech Week is. So whenever someone says like, "Hey, you know, maybe you should be in SF for building a tech company."

Maybe it's true for very s- you know, few cases, but you know, the energy, the ROI, the finance, at least like what we're in the business for at the moment, getting customers ROI from the AI investments, I think New York is the best place to build for that.

Keith Cowing: There's a lot of energy and I love it. Akshit, thank you so much for joining me. Perfect note to finish on.

Akshit Kandi: Thank you, Keith.