Sprinklr Spring 26: The Governance Frontier For Autonomous AI

Autonomous AI is moving from “interesting” to operational, but CX leaders are still balancing excitement with real anxiety about risk, compliance, and quality.

In this interview, Karthik Suri, Chief Product Officer at Sprinklr, breaks down what it takes to make AI agents precise, predictable, and trustworthy, and why governance is the new battleground. He shares how Sprinklr’s Spring 26 release focuses on calibration, longitudinal context, and large-scale testing to improve resolution outcomes while keeping leaders firmly in control.

For more Customer Experience tech news visit https://www.cxtoday.com

Transcript

SPEAKER_01 0:07

Hello and welcome. I'm Rob Wilkinson and today we're taking a closer look at Sprinkler's new Spring 26 product announcement and we're looking at what it tells us about where AI customer experience is heading. If you're challenged with proving AI outcomes in the contact centre or maybe you're exploring how to govern automation safely across CX and marketing, then stay with us because you're going to get a practical takeaway you can use in your business this quarter. Because I'm joined by Karthik Suri, Chief Product Officer at Sprinkler. He's very well qualified to help us understand what these innovations could mean, especially for CX leaders, the IT teams, and obviously importantly, those frontline teams. So welcome, Karthik. Thanks very much for joining me.

SPEAKER_00 0:54

Thank you so much, Rao. I'm stoked to be here and uh hey everyone.

SPEAKER_01 0:58

So let's um set the scene a little bit to start with and just kind of let's break it down into plain English for people. What are CX enterprise leaders most worried about right now when you think around AI moving from like assisted to actually autonomous? What's the biggest fear?

SPEAKER_00 1:16

Yeah, uh I think it's both a fear and opportunity, it's excitement and anxiety all uh culminating at the same time right now, because this is uh one of the most momentous uh periods in the technological history, at least uh since I've been in the workforce for the better part of uh three decades right now. Uh look, CX leaders uh clearly see what AI agents can do beyond just assist, right? Like so they, you know, the autonomous nature of it, the understanding of the context, the depth of uh use cases that it can solve for are all crystal clear to them. And it is something that they appreciate and they want it. But not all customer interactions carry the same weightage or type of risk, right? The getting an FAQ wrong uh, you know, is usually, you know, it has a different risk profile than having something that is, you know, from a compliance perspective that backfires. So CX leaders have to understand the context of this automation, context and power of the AI engines there, and be able to precisely tune it to what we call the governance frontier. The governance frontier takes everything up, you know, across from like you know, FAQs through to resolving basic issues, like billing disputes, retention and cancellation, and then compliance-centric responses. And if you look at all of this, the governance frontier says what portion is autonomous, what portion can like you know, have human in the middle, and what portion should be frankly human-led so that they can balance the risk reward on each of these in a thoughtful, deeply contextual way, Rob. Um, and that is top of mind, and getting that balance is top of mind for uh for a lot of the CX leaders. Um, previously, life used to be a false choice, right? Do I want automation or do I want compliance and accuracy? With AI, actually, you can get both. The power of and is so incredible here that you can get both. You just need to be in a position to have and pass the precise context at the right time for the right solution there.

SPEAKER_01 3:19

I love that. And I think um I'm gonna pinch your phrase there, the yeah, the governance frontier. That's uh that helps really kind of think about where it where it sits in the middle of everything. So uh thanks for that. Um where's the pressure landing right now? So when you think about is it cost, is it the speed, quality? We've touched on compliance, that's obviously plays into that. Or is it something that I've not even mentioned there?

SPEAKER_00 3:48

Uh so you typically, if you look at the business cases that we talk about, right? Like so when we we track uh outcomes fairly rigorously uh in partnership with our customers, in partnership with our large large-scale partners as well. Um, so your business case typically comes from cost reduction, increase in customer satisfaction, speed of resolution, et cetera, right? So that's where the majority of like you know, what portions can be auto-resolved, what portions can be deeply assistive in nature, or how much of the after uh after-call work can be automated, et cetera, around that. And then the constraints typically uh are viewed as like you know, compliance, right? A single compliance failure could be catastrophic for certain industries, particularly highly regulated industries, and quality and consistency are considered as constraints. So your your opportunity set is speed and cost reduction and case resolution, satisfaction, and your constraints are typically quality, consistency, and compliance, etc. With AI, you're in a position to actually offer the best of both worlds, right? You can offer a business case that unleashes the potential of the organization, does a far higher automated resolution to that with a human-centered design in the middle of all of this, speed of resolution, we have seen average handling time reduced by more than 50, 60, 70%, if not instantaneously, which is greater than 90 to 100%, etc. What is important is to ensure that through the course of all of this you're amplifying and increasing quality and consistency. Generally, right, most of the time doesn't cut it, right? In the world of CX. You have to be precise, you have to be predictable, and you have to be trustworthy. So, towards that, consistently and constantly calibrating your AI agents and dynamically adjusting them to understand the risk reward of those resolutions and where you need human in the middle becomes a huge part of your evaluation uh criteria there. So, how do you drive that quality and consistency that is at or better than what you're serving your customers today? How do you ensure that deterministic view of the compliance world where you cannot guess your answer and get away with it, etc.? And how do you make sure that your AI agents are beautifully calibrated with the deepest of the context in the customer journey is the name of the game. And this is why what we are really excited about this release, this what we call release uh 26.4 colloquially internally, is the fact that we are actually providing those tools. Like, you know, typically uh in a in a human worker, a human agent, you have a lot of quality control and automated quality management capabilities. How do we have that or better governance towards autonomously calibrating the AI agents such that you get the benefits of productivity, benefits of customer satisfaction, whilst you're also tightening your compliance and tightening your predictability and consistency as well. And that is the cost speed business case, um, and actually with enhanced constraints that as well uh that we are able to deliver there. I hope that made sense, Rah.

SPEAKER_01 7:01

It it does, and actually you've you've kind of summed up several things there in one because you talk around uh we well, we were talking about how amazing the technology and what a time to be in technology it is, and what you've just said there really paints the picture of the art of the possible, I guess. Um you've also covered that you know um there's there's some really uh strong results being achieved, you know, really significant uh reduction uh in you know in resor improvement in resolution. Um just before we go on to kind of uh the kind of word automation can can trip up sometimes. What about that that that stat you just mentioned there around the kind of really strong resolution performance? How what was the main thing that that drove those improvements? Because that's a really interesting stat.

SPEAKER_00 7:50

Yeah, I would say there are there are three three huge uh criteria that that drive outcomes for customers in a very controlled uh and quality sensitive, consistent sensitive way. First and foremost is having uh uh a robust uh set of data, right? Your your structured data, your unstructured data, what you know about your customers already, what you know, it's uh that forms the basis of everything that you're doing. That that system of record that powers all of this, that understands the customer deeply, um uh is thing number one. Thing number two is rich, deep and specific customer context in a longitudinal way. Context is not a point-in-time thing, it actually dynamically changes, right? Like what happens before and after, like momentarily could actually and dynamically uh uh change the outcomes for your customers uh based on what you know already. Uh so deep customer context in a temporal or a longitudinal way and that is continuous and that persists throughout is the second most important factor for predictably and uh and efficiently delivering these capabilities. And third is an ability to convert these insights into actions, going from actionable insights to actually predictive foresights whereby you can detect and resolve issues or anomalies before even the customer feels it. Uh, what is the Arthur C. Clark term that I love is technology advanced enough is indistinguishable from magic. And for us, that magic comes from data and it comes from context and the ability to translate these insights into action, which is ultimately the goal, right?

SPEAKER_01 9:32

We've spent decades recording conversations for training and reasons, never done anything with that data, and now we've kind of got the ability to you know map it all out, look at it all, turn it into stuff that we can actually then go forward and make changes off the back of. So it it it really is exciting, uh, and thanks for clarifying the kind of the key the key ingredients there to those successes that that you're seeing people have. Um to flip it and to get back kind of back on track a little bit. So where we're not where we're looking at where things sometimes go wrong, so because it you know it is possible for things to go wrong, it it we are in a a great place, but you know, sometimes these things don't work. So when that happens, when when automation in uh particularly doesn't work out or goes wrong at any point, who's the people that feel that first? Is it is it the customers, is it the agents, or is it is it the teams leading those guys?

SPEAKER_00 10:24

I I fundamentally believe that when automation goes wrong, it is the customers that feel it first, right? Uh and and we can have all kinds of statistics that we want. We can even say Six Sigma and it's 3.4 defects per million opportunity. But for that customer, that one person, that human being behind that particular problem, it is a hundred percent failure for them. You know what I mean? So the the strength and numbers may be good in aggregate, but for that every single individual customer that our brands want to serve, a wrong resolution, a dead-end loop, a tone that feels off or it feels tone-deaf, etc. Each of these is both immediate and it is personal. Because behind every brand is that human voice that wants to be listened to, that wants to be validated, etc. Right? From that particular perspective, uh any AI failure is an impact directly on the customer, first and foremost. Second is the agents, right? They are the ones who are quietly absorbing all of these volumes of failure, doing damage control on them, doing cleanup, and they're gonna be frustrated as well. And they are like again, they are uh agents are employees of brands, etc., that represent that particular brand. And therefore, we cannot ignore that sentiment because that is gonna you know transmit to the customer as well. And the third is leadership, right? Typically, they see the metrics, and for most of the leaderships in traditional non-AI powered world, all of this is a rear view mirror, right? They they're they're reacting to metrics that have happened 30 days back. This is like the light that has left the star a while back that is reaching Earth now, right? So for them, this is this is this is all about being reactive. The proactive pattern, you can also flip this on its head and make this actually the best gift that we have given the whole CX experience end-to-end. First and foremost, anomalies can be detected proactively through synthetic testing. That is exactly what we want to do, and we want to calibrate our machines with millions of data points, with a variety of use cases, corner cases, edge cases, to see how the engine responds to it. So at that point in time, your customers have minimal or zero impact because the model is trained, it understands the depth of the context, it understands their sentiment, it understands their status, etc., and it persists it through the particular experience. Your agents are now guided and they are not blindsided. They can intervene early, even if the model fails, or if the customer context does not persist properly. So they have all of the tools, which is why I believe automatic, automated, agentic AI that operates autonomously with a human agent co-pilot can work symbiotically together. And thirdly, leadership now have alerts and scrolls that are proactively detected, like you know, with the right telemetry on the agent itself, so that they can intervene, they can understand, they can trend the signals on a periodic basis, and be more proactive about how uh agents operate on an ongoing basis, and even performance manage AI agents as we are supposed to in this dual working paradigm uh that the world is moving towards. So, customers, human agents, leadership in that particular order of the impact, but a lot more parallel, a lot more cohesive, and a lot more uh real-time in an AI-centric, AI native world.

SPEAKER_01 13:53

And it's it's that real-time proactivity that's different, isn't it? That's the shift that we have to recognize we can do and and embrace, I guess, because we're gonna it's gonna take a lot of change in terms of how people are used to working, because there's uh there's gonna be no hiding place as well for some people. So uh that's exciting again, though. Um so um the announcement that that you've just made that that really does lean on trust quite a lot. Um just in practical terms for the audience, what what does trust mean when you're talking it from an AI agent or a customer service perspective?

SPEAKER_00 14:30

Yeah. Um trust in this context, and everything is contextual, Rob. I know you I hope that point has come clearer whether it's AI or whether this conversation context is um uh all important. Trust in this particular context is about consistency, it's about control, and it is about the end-to-end experience. So it's just not enough for AI agents to be right most of the time, as I mentioned, right? It fundamentally needs to know how it behaves within a set of guardrails every single time, especially in high-stakes situations. So trust comes from being able to visualize and see what the agent is doing, audit it both automatically and manually, and understand why it made a decision and create a feedback loop such that trust gets reinforced and augmented continuously. So this is about predictable outcomes at scale, not occasional successes. It is also about visibility, transparency, and control so that teams can actually monitor, audit, and intervene with confidence and in real time. So, what this means ultimately is predictability and consistency with an ability to intervene and monitor real time and dynamically adjust these such that we are increasing it every single decision that we are making.

SPEAKER_01 15:49

So that's so exciting to think of the the you know the continuous improvement that's running on autopilot, constantly learning, constantly getting better. That's that's kind of the holy grail. Um, so so that sounds amazing. I love that. Um talk me through the the kind of autonomous evaluation part of it. What is it actually evaluating and how does it keep teams safe and honest?

SPEAKER_00 16:16

Yep, yeah, that's it goes back to trust, doesn't it? So ensuring that you're getting all of these benefits, but you're you're ensuring that um uh that the results and outcomes are trustworthy. So autonomous uh uh evaluation actually adds to this equation tremendously by stimulating or simulating how uh essentially your AI agent biv uh works before and after it's deployed and through the courts off its journey as well. So you define key scenarios and the system generates uh thousands and thousands of simulated customer interactions and evaluates the agent against each of them. And each interaction is now evaluated using an LLM that is you know uh fine-tuned and grounded as a judge to determine whether the agent's decision has passed or failed based on the expected outcome. So this is not about just a checkbox on the answer, it is fundamentally stress testing an agent through a wide range of scenarios to see where and whether it breaks. And as the output uh becomes more actionable, it highlights those gaps, the spikes and valleys, and how the agent has been contextualized, taught, trained, and defined, and what are the missing uh links and knowledge gaps and fine-tuning of guardrails that are required that the agents can then continuously improve on the decision that they are making. And when there is human intervention, feeding back the thought process behind that human intervention, that the, you know, if you think about this in the allegory of an iceberg, a vast majority happens in an unsaid manner through organizational contextual context that may not be codified into rule books, right? Signifying and understanding those and and augmenting the rule books and then testing your agents based on uh on those is the autonomous evaluation package um all in one.

SPEAKER_01 18:17

It's amazing to see um where we are today in such a short period of time that this technology has been around because you know there was it's not that long ago that that people were was worried about hallucinations and and didn't trust the outcomes of AIs and and and now we're literally having this autonomous approach to solving customer problems. It's um it who knows where it's gonna go.

SPEAKER_00 18:42

Provided the depth of context is there, right? Right. In the absence of a depth of context, the uh you know what is the the the usual uh uh adage that if Einstein walks into a room in the midst of a conversation without knowing what they're talking about, even a genius like him is not going to be in a position to join the conversation. Um and and you know, uh geniuses walking in blind into a conversation, are going to spew off stuff. I have an uncle like that who does it, right? For the depth of the context, which is why I believe data, context, and access to the right systems of record and the journey is important. But what is progressively separating winners from losers is the access to tacit context, right? Things that are unsaid, things that happen conversationally that actually impli uh that that have implications on a decision. How do you feed that into your system and into your knowledge base for that to be more and more predictable and fine-tuned? And that's where that's the new battlefield right now.

SPEAKER_01 19:45

So let's um we're running out of time, say, but but I'm but I do I do I just do need to talk around this this new AI studio that you've that you that you've talked recently about, this this release. It sounds it it if I have to picture it in my mind, it sounds like a control room where it enables everything to happen. Is that it's just that just my imagination getting carried away. Where does it make scaling AI agents easier? Do teams still find it need to be careful?

SPEAKER_00 20:12

Or I love that imagery, honestly. Um it's it's uh reminds me of uh Star Wars and Star Trek, etc. Right, like so it is actually the control layer where all your Gen AI capabilities come together, right? It gives teams a uh you know, single pane of glass, uh, you know, one place to configure your agents, adjust your prompts and your guardrails, monitor performance, and see what's actually visualize what's happening in production, right? Because it's all bits and bytes, and sometimes not having that imagery could be dangerous because you need you need that visibility and control, because that is what drives consistency and that consistency drives scale. And that scale drives managing AI's performance with your toolkit that you have here. So it makes scaling easier, it's centralizing your configuration, your monitoring, your tuning so that teams can move faster with full visibility, and at the same time, it also minimizes and or contains the risk of human error. So teams still have uh and need a strong governance in how they configure and how they manage these systems, and that's precisely Rob what the AI Plus Studio helps us do. So now the next step is to making this control smarter where humans remain in charge, but AI actively helps guide decisions so that the teams can make the right decision with precision, consistency, and confidence.

SPEAKER_01 21:40

That is it's really exciting times. Um I I my in back in my contact center career, I this this is the kind of the the stuff that I I I wished I'd had. Um we are there, yeah. I didn't get I had to put out put up without it, but the good news is for everyone at home who's watching this.

SPEAKER_00 22:00

this is this is your opportunity to to to uh kind of take advantage of this so um just before we wrap things up um i i i think um it'd be really really good for you to share with our our audience where you know what where's the best place for them to go next if they want to kind of find out this uh about this in more detail um maybe to get in touch with um you know with what you guys are doing at sprinkler no absolutely uh you know get get down to sprinkler.com uh our website is uh all-encompassing our recent press release uh for example goes deep into each of these capabilities the what the so what and the now what for CX leaders uh to build that trust into AI agents uh look at some of the testimonials on outcomes that our customers are enjoying with the power of either assisted autonomous or even connected AI uh uh in the world of real-time intelligence etc um and of course uh follow us on any of the social media uh including LinkedIn so you should be in a position to gather all of this information here amazing amazing uh also don't forget that you can find uh a wealth of related resources stories and other videos just like this one at uh cxtoday.com um but um for that does wrap everything up uh I'm Rob Wilkinson from CX Today uh thanks for joining us