Value Driven Data Science · 2026-05-20 · 12 min
Key moments - from our scoring
Substance score
65 / 100
Five dimensions, 20 points each
Enterprises frequently approach ProArch believing AI is the obvious solution to their problems, but Santosh Kaveti argues this assumption is often wrong. He provides concrete examples: power plant operators struggling with multi-vendor OT network collaboration where AI cannot yet replicate the empathy and judgment needed; HR and wellbeing applications where AI lacks genuine human care and empathy, potentially causing psychological harm when systems change behavior; and cases like water heater pump monitoring where existing systems already contain the necessary capability but lack proper training or workflow optimization. Kaveti's core insight is that 90-95% of customer conversations that start with AI actually point to underlying data, people, process, or integration complexity issues that must be resolved first. He advises against "bolting on AI" as a patch, advocating instead for organizations to redesign workflows as AI-native from the ground up. For data professionals wanting to position themselves as strategic advisors rather than fashionable technologists, the key is working closely with business operators to understand real problems before recommending solutions.
Organizations should avoid AI when they have fundamental issues with their processes, workflows, tools, or data that haven't been addressed first, or when the problem requires human empathy and judgment - such as HR and employee wellbeing applications where AI cannot genuinely care or build human-level relationships.
Approximately 90-95% of customer conversations that begin as AI problems are actually rooted in data, people, processes, or integration complexity issues that must be resolved before AI can be effective.
Evaluate whether your current tools and systems have the capability to deliver the business outcome - often they do, but lack proper training, documentation, or workflow optimization to unlock that capability.
Bolting AI onto broken processes adds another layer of technical debt that becomes even harder to address later and doesn't solve the underlying workflow, people, or process issues.
Work closely with business operators to understand their real constraints, gather evidence about underlying issues, tell a compelling story about why other solutions are needed first, and position yourself as a partner seeking the right solution rather than the fashionable one.
Our reviewer’s read on each dimension, with quotes from the episode.
The episode delivers several concrete, non-obvious insights - particularly the distinction between AI as a lever vs. a fundamental solution, the 90-95% statistic about underlying problems being data/people/process rather than AI, and the water heater pump example showing when existing systems suffice. However, significant padding exists: the host's transitions are verbose, some claims (e.g., 'AI shouldn't replace humans') lack depth, and the HR/empathy discussion, while interesting, remains somewhat abstract without specific enterprise examples. The core thesis is solid but unevenly distributed across filler.
in 90 to 95% of the cases we go down, it's either data. Or people or processes are just complexity of their entire integration. That's the issue, and we end up having to deal with that first before they even talk about ai.
don't patch your systems by bolting on AI to solve your problems that are fundamental to your processes, people tools. Because all you're doing is adding another layer of technical debt
The core contrarian position - that AI is often not the answer - is genuinely counter to current hype and refreshing in the B2B data podcast landscape. The water pump case study and the framework of checking whether existing systems already solve the problem are practical and relatively fresh. However, the execution retreats into familiar territory: the 'human-centric' positioning and the emphasis on 'understanding the business first' are well-worn consultant platitudes. The insights are original in direction but not fully in execution.
By default, I would say no, aI is not where you start at all.
AI is an answer when you are willing to redesign your entire workflow and reimagine it to be ai, native
Santosh Kaveti is a founder/CEO with hands-on operational credibility - he works with power plants, industrial IoT, and enterprise clients, and speaks from direct execution experience rather than theory. His examples are grounded in real customer interactions and technical problems. However, the episode doesn't establish his track record deeply (no revenue scale, no specific customer outcomes disclosed), and he's a repeat guest on a show where he may be in friendly territory, slightly reducing the independent credibility signal.
We work with many power plants one of their main concerns especially in their operational OT side of their network. They work with a number of vendors.
In this case, it was a technical problem. It was easy to fix. Sometimes it's not technical. It involves people, processes
The episode includes concrete examples (power plants with 20 vendors, water heater pump case, HR/AI empathy concerns) and one quantified insight (90-95% of problems are data/people/process, not AI). However, specificity is thin: no company names, no revenue figures, no timelines, no before/after metrics, and no concrete data on the power plant or pump cases beyond narrative description. The HR example about suicides from model behavior changes is alluded to but not substantiated with specifics. Claims would benefit from named references and hard numbers.
We work with many power plants one of their main concerns especially in their operational OT side of their network. They work with a number of vendors. Sometimes there are 20 vendors
in 90 to 95% of the cases we go down, it's either data. Or people or processes
Hayes asks sensible setup questions ('how often does this happen?', 'how would you recognize when AI is the wrong tool?', 'do you get pushback?'), and she does attempt a mild challenge about the 'AI-first' branding trend. However, follow-ups are generally soft and accepting - she rarely pushes back on vague claims, doesn't ask for naming of specific customer outcomes or measurable results, and tends to confirm Kaveti's points rather than probe them. The conversation feels collaborative but lacks the tension and rigor that would signal genuine critical inquiry.
Sounds like AI's good in that it brings people to you so that you can have these conversations with them to make them aware of the areas where they have problems that need to be solved.
So it's about good communication, it's , not about unwillingness or inability. It's about helping them to get the right solution.
Computed from the transcript - who did the talking, and the words that came up most.
These days, every organisation wants to describe themselves as "AI-first". But in the rush to find opportunities to use AI, it can be easy to forget that AI isn't always the right answer. In this Value Boost episode, Santosh Kaveti joins Dr Genevieve Hayes to explore the situations where AI isn't the answer, how to recognise them, and how to have the conversation with stakeholders who are convinced it is. In this episode, you'll discover: The types of problems where AI consistently falls short [01:36] How to recognise when AI is the wrong tool for the job [04:46] Why most AI conversations eventually lead back to data, people and processes [06:25] How to push back on an AI solution without losing stakeholder confidence [09:43] Guest Bio Santosh Kaveti is the CEO and Founder of ProArch, a technology consultancy that helps enterprises operationalise AI securely and at scale. His expertise spans critical infrastructure industries, including power generation, manufacturing and healthcare, where he has seen firsthand how AI can drive business transformation in complex regulatory environments. Links
Transcribed and scored by The B2B Podcast Index.
Dr Genevieve Hayes: Hello and welcome back to Value Driven Data Science, where data professionals become strategic experts. I'm Dr. Genevieve Hayes, and I'm here again with Santosh Kaveti, CEO, and founder of ProArch. Last week, Santosh and I discussed how to build a compelling business case for AI initiatives and how to operationalize those initiatives once the business case is approved.
Today in this value boost episode, we're exploring the situations where AI isn't the answer. How to recognize those situations and the strategic skill of knowing where to push back. Welcome back, Santosh. Santosh Kaveti: Thank you.
It's good to be back again. Dr Genevieve Hayes: Last week we discussed how to turn AI ideas into reality by building a winning business case, but the episode was predicated on the assumption that AI is in fact the right solution to an organization's problems. In the rush to find opportunities to use ai, it can sometimes be easy to forget that AI isn't the only way to solve problems, and it's not always the best. In your experience, how often do enterprises come to you with a problem where they believe AI to be the obvious solution, but it actually isn't?
Santosh Kaveti: It happens. It happens a lot. And I personally believe that AI shouldn't replace humans, and I don't think it will. It'll enable us in many ways.
And it's a force multiplier, to our own human capabilities, that's for sure. I truly believe that. But. I'll give you several examples.
We work with many power plants one of their main concerns especially in their operational OT side of their network. They work with a number of vendors. Sometimes there are 20 vendors supplying different instrumentation, SCADA instrumentation, ICS controls, PLC devices, modern, sensors, IOT devices and edge devices and so on. A lot of times for them just having the visibility of what's on their OT network, understanding the data flows, but more importantly, here's where the kicker is when there is an issue.
Going and really being able to articulate that to the vendor or set of vendors who are partners who really need to resolve that. That's a huge issue for them. Because their job as a power plant operators is to focus on producing and maximizing the throughput. They're not technical enough to be able to say, Hey, this is where the problem is and this vendor needs to solve it.
So when they collaborate with all the vendors to figure out what the issue is, they struggle. So when it comes to that collaboration. Our teams jump in and help. I don't think that AI is quite there to help in that scenario.
The second scenario is when you're applying AI to people-centric processes like hr, you have to understand that AI tool in whatever shape or fashion it is deployed into your enterprise. It doesn't have empathy and it doesn't genuinely care. But it could give you or an employee false impression that it cares because it's echoing an acting as an echo chamber or even a listening partner. And that could be a good thing temporarily.
As models evolve, and let's say it's behavior changes, it could have a serious impact on one's personality. In the world we're seeing now where people are unfortunately committing suicides because models change their behavior. So you cannot build. Human level relationship with ai, and we need to make sure that doesn't happen in enterprises.
At the end of the day, as HR and everybody's deploying these AI systems even for wellbeing, I think one needs to be extremely conscious of the fact that AI tools genuinely are not humans and genuinely will not care with all of our biases. All of our faults would still. Have the judgment to speak the truth when the truth needs to be said, even if the truth is not comfortable, and that's how we've evolved. So that's another example I would say where no I wouldn't recommend AI as a solution Dr Genevieve Hayes: so how would you recognize when AI is the wrong tool for the job?
Santosh Kaveti: Early on when we started this journey, AI was the answer to almost every problem, and very quickly we realized that it is just being used as a leverage in most cases. AI is not. It's your process. It's your workflow, it's your current tool set, and that really needs to be looked at very hard before you say AI is an answer for me here.
I advise to our customers. Constantly and all the time is don't patch your systems by bolting on AI to solve your problems that are fundamental to your processes, people tools. Because all you're doing is adding another layer of technical debt at some point that's gonna come back and bite you, and it'll be even worse to deal with that debt at that point. There are so many examples like that where AI is not the answer, AI is an answer when you are willing to redesign your entire workflow and reimagine it to be ai, native AI is an answer to solve, really good problems.
But it's not an answer to fix your current issues with processes, workflows, tools, or even data. In fact, in most cases, when we get into conversations with our customers, it starts with AI conversation, but 90 to 95% of the cases we go down, it's either data. Or people or processes are just complexity of their entire integration. That's the issue, and we end up having to deal with that first before they even talk about ai.
Dr Genevieve Hayes: Sounds like AI's good in that it brings people to you so that you can have these conversations with them to make them aware of the areas where they have problems that need to be solved. But AI isn't automatically the solution to those problems. Santosh Kaveti: By default, I would say no, aI is not where you start at all. Dr Genevieve Hayes: Is there a framework you use or any specific questions you ask to evaluate whether AI is the answer or not?
Santosh Kaveti: Yes. But it also depends on the context. I'll give you one example. So when we were looking at applying or building an agent we're beginning to build agent factories now, but before we even started this journey.
We were dealing with some water heater pumps. There was already a system monitoring the health of the pump. And the customer in this case wanted to accomplish a certain business outcome and thought that they could simply leverage AI to. Make a decision , in this particular case, for that particular pump, there was no need to go to that extent of leveraging ai, whether it's ML or generative.
Their current system was capable of doing it. They just didn't know. How to work with the system, didn't have training, didn't understand how they can improve the functionality of that system to be able to compare the real time behavior or, look at the manufacturing curve. I look to see how, the pump is behaving over a period of time to be able to make those decisions.
So they already had everything that they needed to be able to do that and they don't need AI system. Now, where AI system comes into play is. You bring in all of the other equipment, all the pumps, all the IOT devices, and you truly want to create an ontology of the entire plant and then start building some agents that are capable of not just looking at one pump, but having that context of the entire plant, not just ot, but also their IT systems, be able to say, Hey, I see this behavior.
Tells me that we need to do maintenance, but you know what? I'm gonna go to it, check to see what the warranty is. And see if we have a part in the inventory. If not, I'm gonna go search.
See, these are the automated tasks where the agents can become then really good at. I'm then start storing that additional context and knowledge for future. So that's one example where. A lot of times you have what you need.
You just need to fix it first. In this case, it was a technical problem. It was easy to fix. Sometimes it's not technical.
It involves people, processes, and that's where it also gets super complicated. Dr Genevieve Hayes: I know a lot of organizations wanna be known as AI first these days. That seems to be the latest buzz word. It's replaced data driven.
When you tell your. Prospective clients know AI isn't the solution. This is what I think you should do instead, which doesn't necessarily involve AI straight away. Do you find you ever get pushback on that?
Santosh Kaveti: Our approach is. Look, we would like to build a human centric, AI native organization. Every organization should aspire to move towards that. Yeah.
When we go in there and when we start challenging their. Preconceived thesis. Sometimes we get pushed back because, no one likes to be challenged. But then.
Our goal is not to offend anybody as much as give them the data and the evidence. I, the story and the director to say, look, this is why, we wanna be great partners for you, but this is the real problem. This is what you need to solve first. And once they get it, we typically don't get any pushback.
They actually understand why, and once they buy into the Y, then they're aligned. Dr Genevieve Hayes: So it's about good communication, it's , not about unwillingness or inability. It's about helping them to get the right solution. Santosh Kaveti: In most cases it is about telling a story that they can really believe in and evidence that.
Dr Genevieve Hayes: So what's one final piece of advice you'd give to data scientists who want to become known as someone who recommends the right solution rather than just the fashionable? Santosh Kaveti: My advice would be work with the operators, get closer to the business, really understand the business, and then also understand how you need to work with other experts to help you build the right solution that will not, come back and bite you. Whether from security perspective, risk perspective, but first step is get with the operators, get with the business, and that's where the real knowledge and the power is.
Dr Genevieve Hayes: So that's it for today's conversation with Santosh. If you haven't already listened to our previous episode where Santosh and I discussed how to build a winning business case for AI initiatives, you'll find it at value-driven data science.com or on your favorite podcast platform. Thanks for joining me again, Santosh.
Santosh Kaveti: Thanks for having me again and. Dr Genevieve Hayes: And for those in the audience, thanks for listening. I'm Dr. Genevieve Hayes, and this has been Value-Driven Data Science.
Other episodes covering the same guests and topics, from across The B2B Podcast Index.
Eat your security vegetablesAdventures in DevOps · on Technical debt88 / 100
Salesforce Team Risk: The Leadership Gap That Breaks OrganizationsThe Hiring Edge · on Technical debt81 / 100
S1E5 - Curating Innovation for FinTech Success - Alessandro HatamiTales from the #FinTech Crypt · on Technical debt77 / 100
Chris Coyier: The Long Game of Maintaining CodePenMaintainable · on Technical debt75 / 100
78. We need to talk about The Vibe ChasmThe Difference Engine · on AI-native workflows75 / 100
Make the move successful: Replatforming content opsContent Operations · on Technical debt75 / 100Episode 112: [Value Boost] Lies, Damned Lies and Stakeholders61 / 100Episode 111: Building Your Defences Against AI Misinformation83 / 100Episode 110: [Value Boost] Why You Need Less Data Than You Think93 / 100Episode 109: How to Measure Anything and Make Better Decisions98 / 100Episode 108: [Value Boost] How to Use AI Without Losing Your Edge75 / 100