AI Assistants and Preparedness Among Enterprises - with Chris Caldwell of Concentrix

Welcome everyone to the The AI in Business Podcast. Today's guest is Chris Caldwell, President and CEO at Concentrix Corporation. Chris joins us to break down where agentic A I actually breaks inside the enterprise, not at the customer interface, but in approval gates, order trails, and compliance checks built for human throughput. He walks through why digital delegates with bounded authority are safer than full digital twins, how poorly tuned agents can burn more cash than the humans they replace, and what most leaders are getting wrong about readiness.

In this episode, we cover the operational realities of deploying agentic A I inside enterprise environments. To go deeper on this topic and learn how to evaluate AI vendors by assessing leadership expertise and why funding benchmarks can signal product maturity and stability, download our free PDF report, five ways to select the right AI vendor at emerj.com/aiv2. That's E M E R J dot com slash A I V two to download your copy.

Now the conversation with Chris. Chris, it's lovely to have you in our Emergent AI research studio today.

Great, very nice to be here.

So we've been hearing a lot about A I moving from something that used to just help us write and summarize to systems that actually act, act in making decisions and sometimes even transacting on our behalf. So there's a lot for us to dive into today and to dissect, especially with the hype around agentic A I and digital delegates running well, basically running into end workflows at this stage. But I think the best starting point would be for me to ask you, where do you see enterprises feel the first points of friction when assistants start acting inside of workflows?

Well, it's a great question. The reality is is that most uh enterprises haven't necessarily thought through all the parameters they need to deal with. The first is, you know, even when you're thinking about, you talked about digital delegates. Is it a digital delegate? Is there a digital twin? Do you have all my authority and are doing everything that I can do? Or am I giving you a set of authority that you can then act upon? And some companies even get tripped up on that.

That's the first thing. The second thing is a lot of the workflows and systems that enterprises have have been really designed for humans. I mean, they're used to a human speed, they're used to a human number of transactions going in. They sometimes have uh shut down windows on the weekends at nights to kind of do the batch processing on the back end on some of the big, large enterprises. And then all of a sudden, with uh some of the agentic work that you can do, you can 10 x that, 100 x that, 1,000 x that, and then bottlenecks start creeping up within the workflow that starts to cause a lot of problems.

And so you somewhat have to almost stress test the system to make sure that it can handle the amount of volume that you're going to kind of anticipate and drive. And then really sort of the last one is really around kind of thinking through the security measures that happen. We we see oftentimes people try and rush into it, they start to kind of implement stuff and then they start to realize, my goodness, uh fraud is picked up because we didn't realize that a human was actually needing to touch this.

Or they start to realize that humans, because of the volume, are just now pressing click click, approve of all the agentic work versus actually kind of looking at it and digesting it, which has historically has happened when it's more of a human pace. So lots of sort of challenges and need to drive all the friction points that you have to kind of get in front of before you can really get some benefits out of the agentics.

So so two points that I want to get onto that. The the first thing, if we talk about speed and and I think the term that is being thrown around is is machine speed. What is machine speed actually break inside the process that is designed for humans?

Well, it's it's a couple things. Uh, some of the legacy enterprises that we deal with, where machine speed breaks down, is that they're not used to handling the number of transactions. Like just in terms of the the end of day clearing that they need to do, the system downtime windows that they need to do, are just not big enough to process the amount of transactions that have now been processed through through agentics. That that's one. The second thing is that there's some physical sort of signoffs that sometimes happen within workflow processes where a manager approves all the work that the people do.

Now that manager normally approves maybe a hundred, 200, 300 claims a day, and suddenly he's now needing to look at 5,000. Do you really think he's spending the time looking at it, or do you think he's just going approve it? I'm sure the machine did it properly. Exactly. I'm sure the machine is better than than me, so I'm just going to approve it. So those are the things that we we talk about, uh, machine speed, that somewhat break some of the processes that have happened, where systems have historically been built off of human labor that are going into them.

So is the solution to that employing more humans?

Actually, no. The the the solution to that is actually, first of all, looking at the journey map of what you're trying to automate. Two is making sure you're putting in the control points that are either, you know, frankly, other bots doing the work. So we talk about compliance agentic bots that actually are separate bots that check work of the other bots, believe it or not, as a as a check and balance to make sure that they're doing the right thing.

It's also making sure that you're putting in sort of the the right amount of horsepower in your systems, for lack of a better term, to process the amount of transactions that you can now do. No one is looking at this as adding more humans into the loop. What it is is really about making sure you're putting humans in to safeguard what you need to do, and enough, you know, agentic horsepower into the system to make sure that you're able to process what you need to do safely and securely.

So essentially, we'll have bots micromanaging agents. Is that the the picture?

Uh, uh, yeah, absolutely, we we have that now, you you know, we have so sort of collection agentic bots that are out there that kind of go and do things. And then we have other bots that check that they're compliant to make sure that they're doing things in line with the financial institution. As crazy as it sounds, just think of the the human work environment. You're the same thing. You have internal audit, you have security teams, you have all these, you know, you have your your compliance teams.

All these teams already function. Now they will have agentic bots to effectively help and and do their job at the same machine speed as the processing that's happening.

That makes that makes complete sense and that the picture is actually very clear. The second thing that I wanted to just also pick your brain on in terms of friction is we've now seen or or you've explained where we see those friction points. But where does the friction show up? First, is it normally where we are customer facing, is it in employee workflows or back office operations? What is that first point of friction that we normally identify?

Uh, actually, that's a great question. You know, normally it's not necessarily if you are putting in agentics in a very conscious and thoughtful way, it's normally not customers who have friction first. If they are able to complete their task and they're able to get through the journey from start to finish, they're pretty happy. It's normally on the back end. And it goes back to those points around making sure that, you know, you've got the compliance in place, you've got the security in place, that's able to get to the data that it needs to.

Your data is clean. When you talk about sort of big enterprise level friction points, it's more about setting it up. Can you get to the data, right, to to make sure that it's done what it's done. Do you have the right security, uh, approvals in place? A lot of times, people think agentic, uh, bots are kind of out of the box and you you you put them in place and they run. My best analogy is like hiring someone right out of university.

You have a really smart person, but they have no idea what you're doing, they have no idea about your company, they have no idea about your processes, and you've got to spend the time. And a lot of uh organizations kind of forget that. And so the friction is actually getting those bots up and running in the first place and training them with the right skills to do what they need to do and exactly what they need to do and not more than that they need to uh they they need to do.

And then lastly, I'd say we see friction primarily around the audit trail. If you don't have the complete end to end audit trail in place to understand what can go wrong, to understand what's actually happening with the transaction or or process that you're doing. Oftentimes, we'll see companies, security and compliance teams go, whoa whoa, what are you doing? Where is this coming from? Because they've suddenly seen the volume, because it's not sort of this holistic view of what's happening and why you're putting this technology and what the outcomes you want and that they're the right outcomes and they're confirmed outcomes.

So that's kind of the things that we see within the enterprises are working and not working.

Well, it sounds like you've had a lot of experience with this already, uh, being able to identify those first friction points. And uh, with the term friction, uh, I think some leaders would push back and say friction is overblown, that the term is is being overblown. Because most enterprises aren't running agentic A I at full scale yet, so the whole narrative of operational risk, uh, is kind of a little bit premature. How do you respond to that?

Uh, I would say there are many enterprises who have put some things into production that should not, and they have made the front page of the paper or been on the TV, or been been very well, you know, characterized in in social media and and sometimes reported on the news on television. So, th those things, mistakes happen. Is agentic full blown, full blown as what we expect it to be really almost by the end of 2026?

We think this will ramp up very quickly. No, it it's not, but there's enough people playing with it. And, you know, uh, um, I think OpenClaw was really a turning point at the beginning of the year. Even people who weren't really into AI or agentics generally heard about OpenClaw, saw what you could do, started playing around with it. And we're seeing that creeping into the enterprises as as people coming into their office saying, hey, we should we should do this.

So it it is there. Is it as as massive scale as you would expect? It is not, but it will get there pretty, pretty quickly.

And and I think at this moment, this it's it's almost as if that turning point is happening faster and faster. Just I thought about it the other day that four years ago, my boss would have said to me, why are you using A I on the job? Whereas now the the instruction or the question is, why are you not? Why are you not using A I to do this? So there's a big, that mind shift together with understanding the capabilities has definitely increased the speed of what enterprises will integrate A I into their workflows.

Do you agree on that?

Yeah, I do agree on that. And and let's be clear, like A I in some workflows, not agentic A I, but other types of A I are already in the workflows. Like it's already in risk models in banks. It's already, you know, obviously doing customer service and collection work and a whole host of other different things. When we talk about agentic A I, that really started coming up more at the beginning of this year. But we expect it to absolutely, uh, accelerate faster through the back half of the year and really drive a lot more value than what you've traditionally seen from.

I'll call it frontline A I, which is, you know, the the the basic question answer in your in your staff or help me write an email or put a white paper together report, which now is fairly, you know, common in everyday use within the within the enterprises.

Next term that we also hear so many times and and almost in every enterprise level agentic AI or a I in general conversation is is readiness. So readiness is one of those words that we see being thrown around everywhere, uh, every vendor, every analyst report, uh, every keynote, um, it's everywhere. And with MIT's gen AI divide report saying that pilot failure is around 95%, and then we have the EU A AI Act kicking in for high risk systems later in this year.

There's a real appetite for clarity on what it actually means to be ready. So readiness everywhere, but what does it mean, or what does ready for assistant driven work actually require?

Y- Yeah, you know, that's very interesting because the vast majority of pilots were failing, probably, uh, uh, you know, two years ago, certainly a lot of pilots were failing last year, and even pilots, uh, kind of misthought of this year failing. And that's really driven by, uh, organizations kind of trying to be completely A I ready on the enterprise, versus saying, okay, I just want to put A I in this task. I just want to put A I in this end to end journey or this process.

When it's very defined what you're trying to do, where there's very deliberate outcomes that you're trying to achieve, those actually, they don't even go to pilots anymore. They generally go right into full production and drive, uh, drive drive value for the for the organization. So I think it's first of all, making sure that you understand what you're getting into and define it very, very clearly. Big enterprise ready comes down to security, comes down to data governance, comes down to now with agentics, we believe that there's going to be a a fairly large growth of sort of these new human positions around orchestrating work between digital workers and and human workers, uh, that need to happen.

And so making sure that you understand how to to to do that. Managing the cost structure. If you think about token consumption, um, people kind of take it for granted, but we've been playing around a lot internally and we've found that badly tuned agentic agents, uh, are way more expensive than humans. They can burn through a lot more money. Like it's cooler, but it's a lot more expensive. And so it's it's understanding, like, okay, how to monitor that, how to measure that, how to optimize it are all part of this sort of the readiness scheme that need to to happen.

And again, trying to do it across the the the large enterprises, very very few enterprises are 100% A I ready. I I just don't believe it. If they tell you they are, I fundamentally don't believe it, cause we see the inner workings of it. What we see is where it's really targeted, you can get some very good, strong outcomes, both from a financial basis and, uh, you know, growing your revenue, cutting costs, et cetera, et cetera, whatever your your achievable focus goals are trying to be.

Right, so so you've given us kind of a few foundations here, saying that, you know, to to before we think about being ready, understand what the business problem is, um, understand the security and governance requirements, understand the managing the costs of it. So of these foundations, uh, which one are most enterprises the furthest behind on right now?

Well, most enterprises are not looking at the the monetary aspects of A I at all. They're simply saying, hey, we'll we'll deal with it and costs will will come down. I think the next thing that uh, most enterprises are behind on is really understanding the security risks of this. And I think that's what we've talked about where, you know, bad things have happened and shown up in the paper as people didn't really understand what's happening. It's funny, like A I is like shining a flashlight on the most hidden garden secrets in your company.

It it finds everything. So if you have stuff that you you don't want it to find, you need to make sure that that is is securely. And and that is is really not as well understood as we'd think. And the last thing is really around the the the data governance and the training of what you need to do. I think we still deal with a lot of enterprises who have very disparate data. They have, you know, fi everything all over the place and and and kind of getting that into a method to really power the A I effectively.

It's coming up the curve, don't get me wrong, uh, it's a lot better than a year ago, but it's it's nowhere close to where I I think it needs to be for enterprises really to get the best benefits out of A I and and see the benefits to their cost cases when they're when they're doing these big deployments.

Makes sense. And and I think, uh, I like the flashlight thing, cause I've always thought about it as, you know, when people say, oh, since we've integrated this A I, we now have this problem. And then many times it's not the problem has been there all along. Um, A I has just brought it to the surface. So that's that's actually what happened.

We we see that all the time. I mean, people put something in A I and they I I'll give you a simple example. We had a company that was showing us this knowledge base that they built and they were very, very, very proud of it. They were saying that they did it all themselves. And we said, well, you know, what did you put into it? Oh, we've hooked into everything. And literally, you could ask this knowledge base, what is this company's top ten customers?

What's their revenue? How much do we, uh, how much do we make off of it? And it spit out the data and the the the senior executive almost fainted because it was highly confidential. Anyone in the organization, anyone in the organization could could query it. Th- this is not a lone case. This happens more than you would think when people start to to kind of put this in un unguarded into the enterprises. And and so, yes, absolutely shines a flashlight.

You need to make sure that your your house is clean before you turn on the lights. I guess is probably the best analogy I can think of.

That that sounds about right. Uh, you've also mentioned digital twins and digital delegates. So how should leaders think about setting boundaries for what a bot is actually authorized to do? What should their thinking process be there?

So it's a personal opinion. I'm a big believer in digital delegates. I'm a big believer in that if I am hiring either a human or, you know, a so-called bot, that I want to set the parameters of what this individual can do. I want to train them appropriately. I want to make sure that they have the right kind of processes and then I want to make sure that they have the right safeguards. And I want to have the right controls on it.

I think digital twins are incredibly dangerous in the enterprise, while they're more productive and easier because it's basically running on your system, saying, hey, have access to everything, do everything, use my passwords and credentials. Uh, I just think horrendous, horrendous things can can happen with that. And so I'm strongly in the camp of of digital delegates. And I think that mindset helps business leaders think about A I. They think, okay, I'm gonna hire, you know, hundred new graduates this this quarter.

What would I do for a human? Okay, I need to do the same thing. I need to train them on the right material. I need to make sure that they have access to the right systems. I need to make sure that their work product is checked by a supervisor for a period of time to make sure that that's gonna work the right way. I need to make sure they're compliant and secure. And then let's uh let's start them off.

And as they kind of build their qualifications, for lack of a better term, you can give them more and more complex work and they continue to to go on. I think that's a far safer bet than than saying, hey, I'm a a a C E O or senior executive of a company. I'm now giving this A I, uh, agentic bot on my system and access to everything that I can do, that'd be horrendous. I I would not suggest that in any way.

I'm sure someone will disagree, but I would absolutely not do it.

So, but but uh, it sounds as the reasoning behind it is because you you've seen things happen, so it's not just a random thought. So, uh, uh, someone recently said to me that all types of A I when when we integrate them, and probably not all type, but but the let's take the delegates for as an example. So, um, they should be treated as lifelong interns. That's that's how you should see them. And I like that idea because it's like, okay, I'm not going to just get these delegates and send them into the field.

And then, like you said, they're university students that just graduated. So, lifelong interns, meaning that they will have to learn and keep on learning and they might create liability at some point. So I need to make sure that they're monitored. So it it completely makes sense.

Yeah, w- I will say what we've sorry, I I would say what we found actually. The lifelong intern is is uh a good analogy, but I do think that some of those interns graduate. And and we've seen that ourselves where once you get to a certain level of of knowledge and and and experience and processes, some of those interns can actually tell you how to do your job a lot better. Some of them process it a lot better.

They can almost be almost a peer or mentor to you for for some of the stuff that happens. And I do think that can happen as well. So, maybe lifelong interns in some parts of the organization, but absolutely, I I think they do graduate and get promoted in in other parts of the organization.

S So is that consistent? If they've graduated and this delegate can do my job better and and advise me on doing my job better, is there no room for them going back into intern mode? So is is that consistent? Can I be guaranteed that they will continuously be this graduated?

There are no guarantees with A I. None. Anyone who tells you that there's guarantees with A I, I have some some farmland in Florida to sell you. It it just it's not, it's not the case. I I think you have to be cautious still at this point, at least from what I see. And I think, um, you continue to need to have some safeguards, you need to have some structure. I'll give you a very simple analogy.

W- we are rolling out E A S that are uh bots within our organization. And when we first started, uh, playing with them and start putting them in place, they would do basic scheduling. But they wouldn't have any external calls, they wouldn't talk to anyone internally. You work with them and then they would try and arrange your schedule. They would do research on where you had to have a client meeting, all that type of stuff. We're now sort of graduating to saying, okay, now when there's a conflict, please call the customer and move this meeting with with with voice, uh uh, or through WhatsApp or or through, uh, you know, other types of communications, Microsoft teams.

So we've graduated it into a skill. As that's been building, we've now said, okay, now we want you to do project management. Okay, that's starting to work, where it's communicating with other people in the organization. But what we haven't said is you can start sending emails externally, you can only start sending emails internally. At one point, next point, we'll say, okay, now you can, uh uh, serve emails externally. What's coming out of this is a lot of good data about, you know, who's doing their work projects on time, how are things working in the organization on time, what information that you might not necessarily see is is coming to light with it.

And so they're actually becoming more sort of helpful, both from a tactical perspective as as and and then more as a sort of a senior perspective. That's the type of graduation I mean. Would I ever have it take over my job? I'm who knows, maybe, but but probably not this year. Let's put it that way.

Hopefully not, they're not ready to retire just yet. So we've we've been hearing a lot about the post hype reality check setting in. And I think with it already being predicted that over 40% of agentic A I projects will be cancelled by 2027, um, most leaders might feel like they only have twelve to eighteen months of a window to get this right. So to close us out today, um, I'd love to ask you what decisions will determine whether assistance will improve R O I or just create operational drag?

We strongly believe it's coming back to having a very clear case of what you're trying to use the agents to do. We still think enterprises over expand. They they s read some article, listen to something and think it's so easy to kind of deploy these and why can't they deploy? And it's it's missed expectations. And then what happens is they spend a lot of money, a lot of effort, a lot of time and go, wow, this is not giving the R O I that we need.

W- w- you know, kill it or cut it down pretty dramatically. The organizations that are really doing well with this stuff are starting off with, uh, like what we talked about. We're we're helping executive assistants by by giving them, uh, these bots. Simple process, easy to do, very defined R O I, uh, and you can build on that. And then project manage, et cetera, et cetera, et cetera, and you can kind of go through. That's our example.

Many other companies have d have have different examples of what they do. It. When it's clearly defined and and and done, it's the first step. Second thing is understanding that these things don't come out of the box and work, uh, you know, perfectly. It's like Christmas morning when you have to assemble the toys for your kids. Like, it's a great surprise when it's all built, but it's miserable as a parent trying to put these things together before before your kids wake up.

These bots are exactly the same way. You need to make sure that you're able to to to build them out and give them the skills and that they take uh care and feeding as you go, and it's not one stop shop. So setting those expectations right and getting the training done and then having the safeguards so that they don't go off and do something silly and and you wanna cancel your project because you've had some additional losses.

My my best analogy for A I that that, uh, I would love to take credit for, but I heard from someone else is, you know, the the early bird catches the worm, but the second mouse eats the cheese. And I think in A I, a lot of the times, you wanna be the mouse to get the cheese, not the early bird.

Yes, someone else go and make the mistakes and we'll learn from it. That's what we want. I've asked this question so many times, so I'm gonna flip it a bit. We normally ask what is something leaders should start doing at this very moment? But uh, we get different answers depending on industries and perspectives. So I'm gonna flip this and ask you, what is one thing leaders or senior leaders, uh, actually need to stop doing in terms of their A I journey?

I generally are suggesting that senior leaders ground themselves in what actually works and what is available now. I think too many enterprises are trying to to like, look, there there's a new A I announcement every week, every single week. And it's bigger, it's better, it's faster, it's flashier, it it does more things, it comes with steak knives, right? Like, it's it's ridiculous how much the hype is. And there's a lot of good stuff, but take it at 50%.

And I think if you wanna get your A I journey started is to have a clear direction of, okay, this is what's practical, this is what we can do, this is what we need to do for for the next six months, eight months and and set on that. Otherwise, you kind of weave back and forth. Just realize that a lot of your investments that you're going to be doing over six months, you are going to need to kind of refresh.

Don't don't expect your A I journey to start now and last for three years, which typically is how a lot of enterprise technology is thought about, is, you know, implement for a year or two, then get up and running for three or four years, it might be ten years you're on something or twenty years you're on something. With A I, we're finding that we're having to replace things every six months to a year and invest accordingly to get payback from those types of, uh, uh, investment cycles.

But great great idea. Uh, I want you to leave our audience with, uh, a visual. Uh, w- to you, what does leadership look like in an era where bots and humans share a workflow?

Uh, leadership looks like embracing it. Uh, leadership looks like using it yourself. Leadership looks like, you know, discussing and talking about it and being curious about it. Leadership does not look like telling your I T department to go and deploy A I and hope it works. And being, you know, cavalier about it. I think if you're really serious about this, y- you have to be one of the early adopters. You have to to walk the talk.

And, uh, you you have to kind of drive that passion within your organization. That's that's personally what we talk about all the time in our organization. And we see the organizations that embrace A I and get the value, having similar types of leadership characteristics within the business.

And I think that's a great visual to leave our audience with. So, uh, Chris, my my favorite parts from our discussion today was first talking about the bots, micromanaging the agents. I like that idea. I like that you gave us good foundations on what solutions would look like. Um, I love the fact that you said we need to make sure the house is clean before we turn on the light. That is because everything will come to the surface.

And to for leaders to really believe in the A I that they are integrating into workflows, uh, for their staff. So it's it's been great having this discussion with you. Thank you for sharing your insights, uh, with us. I cannot wait to hear what the audience has to say.

Great, thank you very much for having me.

Wrapping up today's episode, I think there are three key takeaways from our conversation with Chris. First, agentic A I exposes every weak point in legacy systems the moment it operates at machine speed, which means readiness has to start where existing processes break, not where A I looks impressive. Second, bounded digital delegates with trained authority and visible order trails outperform unbounded digital twins across every dimension that matters in the enterprise. Security, compliance, cost, and trust. And finally, leaders should anchor investment in narrowly defined use cases that can ship to production and plan to refresh the stack every six to twelve months rather than chasing each new model announcement.

To go deeper on this topic and learn how to evaluate A I vendors by assessing leadership expertise and why funding benchmarks can signal product maturity and stability, download our free P D F report, five ways to select the right A I vendor at emerj.com/aiv2. That's E M E R J dot com slash A I V two to download your copy. On behalf of the team at Emerge, we'll see you on the next episode.