The “Yesterday” Problem: What I’ve learned from customers who challenged the traditional procurement timeline  

By Jag Gattu, CEO, UptimeAI  

This blog was originally published on LinkedIn.

I have this question I ask almost every operations leader I meet: when do you actually need this running? I ask it half out of curiosity and half because I already know the answer. Nine times out of ten, it’s some version of “yesterday.” Somebody’s losing sleep over a unit that keeps tripping, or a maintenance backlog that never shrinks, or a retirement that’s about to walk two decades of institutional memory out the door. There’s real urgency in the room.  

And then, almost every time, that urgency hits a wall when it comes to the procurement process. First, they’re waiting on legal. Next, it’s IT. Someone has to find out who actually owns all the systems that need to be connected together. And by the time all of that has happened in sequence, the six months that everyone swore they didn’t have at the start of the project have quietly gone by.  

I want to be careful here, because I don’t think the lesson is “move faster at all costs.” I’ve watched companies get burned by exactly that instinct — skipping a scoping session or waving off a data question can add more delays later. Whatever we build, it has to earn its way into a plant’s operations, and that takes real diligence. What I’ve learned instead, from watching a lot of these cycles up close, is that diligence doesn’t have to walk a straight line. Many procurement tasks can occur simultaneously; they just usually don’t because it’s not the way things have historically been done.   

Here are a few of my favorite techniques that our customers have used to accelerate the value generation of their AI project by choosing not to wait on traditional timelines.   

1. Be an advocate for imperfect data  

I remember a call with an operations director at a cement plant who was clearly ready to move, but his team kept getting stuck arguing over whether they had “enough” and “good enough” data. Every week someone identified a new handful or tags, or a new sensor that would be “nice to have when trying to predict XYZ.” It could have gone on indefinitely.  

At some point he just made a call: they’d agree on the handful of failure modes and the data around the assets that actually mattered for proving the case, then treat everything else as things that could be done in parallel once the solution was live. That one decision point—to accept that some things would be ready now, others would improve over time, and even more opportunities would be uncovered by their use of the technology—avoided weeks of delays in getting the software operational.  

2. Know your scope, your systems, and your stakeholders  

We worked with a project manager for deployment at a large refining client whose team showed up on day one with a clearly defined pilot scope, the KPIs they’d measure success on, the data they’d need to deliver successful results, and the people with the right access to the relevant systems to make it all happen.   

This might sound trivial, but I saw another engagement get delayed by over a month because a new data source was added to the scope which lived at a different level of the enterprise architecture and had a different owner than our other refinery data sources. That owner wasn’t up to speed on our project, and we were fighting for their time with other priority projects that they’d known about for much longer.   

3. Identify, and challenge, dependencies 

One of our early power generation customers taught me something I hadn’t fully appreciated: a lot of the delay in these deals comes from waiting on others out of habit rather than requirement. 

Their security and architecture review ran at the same time as the final commercial conversation, not after it, because somebody on their team asked “why would these two things need to happen in order?” and nobody had a good answer. Procurement and legal moved on a similar clock. Nothing about the actual approvals changed — same reviewers, same rigor. What changed was that they finished around the same time the contract did. That business was ready to execute the week they signed, instead of six or eight weeks later, which is close to how long that kind of review usually takes when everyone assumes it must come last.  

What I want you to take away from these stories 

None of these customers moved faster by cutting anything out. They moved faster because they stopped assuming that due diligence must happen in series. Confirming the use case, mapping the data, starting the security and legal review, building the onboarding plan — almost none of it actually depends on each other.   

If your team is telling you they needed this yesterday, I’d take that seriously — not as a reason to skip anything, but as a reason to ask, honestly, which parts of the next few months are actually sequential out of requirement, and which ones are just habit and can be accelerated to help get them the solution they need, yesterday.  

I’m grateful to the customers who have proved to me that it’s possible to go from demo to go-live in less than a quarter. We’re taking those learnings, applying them to every customer sale, and helping our new customers push to the frontier of adoption of AI reasoning agents for industrial operations.   

Our Commitment to Customer Value: Why We Asked Someone Else to Grade Our ROI

By Jag Gattu, CEO at UptimeAI 

This article originally appeared on LinkedIn.

Every industrial software vendor promises transformation: faster diagnoses, fewer surprises, a healthier bottom line.  

Today’s market places too much responsibility on the customer when it comes to validating these promises. For executives deciding where to put scarce digital investment dollars, the ambiguity gets expensive, and it can be the reason that a promising pilot never reaches enterprise scale. It’s not because technology isn’t valuable, but you need to be able to build a business case that finance can trust. 

This is why, in a market swamped with self-reported ‘proven results’ we invested in helping your teams build a confident business case. Conducting economic studies like the Verdantix Verified Value Delivery (VVD) methodology is not inexpensive and it’s not trivial. It involves hours of interviews with UptimeAI customers to parse out exactly how they are realizing value from the technology, then building that into a calculation framework that can apply across company sizes, industry verticals, and products deployed.  

We think it’s important that our customers understand what value they should expect from the earliest days of engagement – in the form of proof, not promise.  

How do you measure the impact of something that never happened? 

One of the biggest challenges for predictive technologies, particularly in the reliability space, is the definition of success. When our software works exactly as intended, the outcome is often nothing happens. The cost savings of avoiding unplanned failures and unit shutdowns are typically only quantified when the shutdown actually occurs. So how do you quantify the value of something that never happened? 

Since UptimeAI started, our customer success team has worked with every customer to track every alert and diagnosis that prevented a failure and aligned with the customer on what that catch was worth. They comb through past failure events, the lost production, the maintenance expenses, then the system learns the value in warning of that failure mode, so that next time value is automatically assigned. We know that our champions are constantly being asked to justify spend on operational technologies, and we view it as our job to make that task as easy as possible for them.  

Customer interviews confirmed UptimeAI’s commitment to defensible value 

Verdantix conducted interviews of 5 global customers, using their responses to build a model of expected returns for various company demographics. Specifically, for a model $200M-revenue manufacturing site, 3-year ROI of 197% grows to 250% as they expand enterprise-wide, and the software has fully paid for itself in 11 months. In some cases, it’s much faster.  

For one major global cement company – “We had 4 critical alerts in first 6 months of the pilot, when we were very conscious of the value to justify further investment in the solution, and this recouped the cost of the investment.” We continue to compress this payback period by delivering more value, sooner, through more agent products and workflows. 

An exciting future ahead 

197% ROI is exciting, but that’s just getting started. In just 6 months since the study was completed, there are already so many additional value streams that our customers are seeing leveraging new agents. I can’t wait to see what the next year will bring for our current and future customers. 

For the full details, read the full Verdantix VVD study.

Build vs. Buy, Part II: The Very Real Barriers to BIY in Highly Regulated Industries

By Jag Gattu, CEO, UptimeAI 

This post originally appeared on LinkedIn.

My first post on this topic highlighted the shift we’re seeing in how many tools we build in house vs. buy, and where we draw the line. At UptimeAI, our line sits at the transition point from retrieval to reasoning.  

I took a step back to see where others in the market are drawing the line. While I found general excitement about the potential of BIY (Claude Code and the likes), I found that leaders are being appropriately calculated in their consideration of these projects. There isn’t just one line that they’re drawing for when to build versus buy a solution, but many, and the lines link back to the same challenges that have plagued digital projects in our industry for the past decade.  

Here are the three biggest challenges with BIY agentic AI projects in the process industries.  

1. Governance

In a recent blog post, Verdantix highlighted that AI sprawl is becoming a C-suite problem. Sprawl is a second-order cost to unlimited BIY development. Every vibe-coded agent is also one more tool requiring governance — and a new source of shadow AI, data leakage, and regulatory exposure if it’s not accounted for. Even Amazon has acknowledged internally that “the AI boom is driving duplication and fragmentation, not less of it.” Building a tool and then governing it responsibly, indefinitely, in a live plant are two different jobs, and the governance doesn’t get easier just because the build got faster. Industrial software companies have been operating in this area for decades and have the data governance and cybersecurity infrastructure already in place to deploy agentic AI projects fast without risk. 

2. Scale

According to McKinsey’s 2025 State of AI report, less than 10% of organizations currently experimenting with building AI agents are scaling them. BIY pilot development is easier than production deployment. You can have a demo working reliably on clean, well-behaved offline data in a matter of hours. Standing up something that holds up to the challenges of real operations data — fragmented, inconsistently modeled ERP, MES, historian, and SCADA data, spanning reliability, process, and operations teams — is another. Agents built on top of fragmented data don’t fix fragmented decision-making, they automate and amplify it. That’s true whether you built the agent or bought it. The difference is whether you’re building the curation, integration, and verification layer from scratch, or standing on one that’s already been proven across other plants like yours. 

3. Impact

In a recent LNS Research blog post, Research Analyst Vivek Murugesan emphasizes the importance of “leading with business problems and not technology itself.” Are you looking for a personal productivity tool or a way to make margin-impacting operations decisions? A tool that surfaces the right document or summarizes a work order can be easily built, and it is useful, but it isn’t the same as a system that provides operators with confident decisions to high-stakes calls. Margin is lifted through improvements to availability, quality, and reliability. Personal productivity can support improvements to these areas, but it won’t move the needle on margin the way the components of operational excellence can. An enterprise deployment of a Root Cause Agent, for example, requires automated expert reasoning, robust data handling, and continuous learning and improvement loop. That’s not retrieval, it’s judgment, and it’s proven to provide rapid technology ROI through improvements to diagnosis time, accuracy, and avoided downtime. 

Draw YOUR right lines 

The challenges raised above aren’t reasons to avoid BIY altogether. There are problems where in-house solutions are sufficient, like our sales enablement tool I mentioned in my last post. Retrieval has become easy to build and scale in-house. But when making build v. buy decisions around reasoning and judgement—the agentic AI capabilities required to make impactful decisions in a regulated industrial environment—trusted partners can effectively overcome the governance, scale, and impact hurdles. 

The Stories Behind the ROI: What 197% Looks Like on the Plant Floor

By Shruti Kela, Engagement Manager at UptimeAI 

This blog post originated on LinkedIn.

I have spent my career on one question from both ends: what is the real business impact of the work that keeps a plant running. At SLB, I worked in the field, close enough to the equipment to know what a failure costs when it lands. At McKinsey, I built the business cases that decided whether a plant spent millions to prevent that or lived with the risk. Knowing the price of a breakdown is what makes a prevented one so valuable, and also what makes it so hard to prove. The biggest wins in this business are the failures that never happen, and you cannot show a board a breakdown that was quietly avoided.

Unless you catch it in the act. UptimeAI flags a failure early, names the likely cause, and recommends the fix, so engineers act before the loss lands, and every catch is logged as it happens. The breakdown stays invisible, but the savings do not, and those records are exactly what Verdantix set out to audit when UptimeAI brought them in to conduct a Verified Value Delivery study.

That kind of proof is rare. In their 2026 Global Corporate Survey, Verdantix found that 85% of industrial firms say measuring the ROI of their AI tools is a real barrier, which is how promising pilots quietly die. Working from the records of five UptimeAI customers, their independent study landed on a 197% three-year return for a typical site, growing toward 250% at scale.

A high-level metric like that is only as trustworthy as the data that sits under it, so instead of the top-down math, I sat down with my colleagues on UptimeAI’s Customer Success team to sum up the individual moments that compound into triple-digit ROI.

The building blocks for triple-digit ROI

A failure with no alarm

At a cement plant, the kiln is the whole business. Every ton of clinker produced runs through it. One of the sneakiest ways to lose output is coating building at the kiln inlet, and there is no alarm for it, because it never shows up as one bad number. It shows up as small shifts across many at once, pressure edging up, gas composition drifting, each too minor to notice alone. Operators cannot watch every point at once. UptimeAI read them as a system, recognized the coating pattern, and flagged it hours ahead with a clear instruction: ease off the fuel, check the chemistry, prepare to clean. The team cleared it during a short, planned stop instead of a reactive multi-day shutdown, and kept the kiln producing. The value of this catch was amplified a few weeks later when the same issue occurred on a kiln at another plant; six figures and dozens of production hours were kept rather than lost.

One flat gauge hiding many moving ones

A gas turbine at a power operation was holding steady at full load, nothing an operator would look at twice. Underneath, UptimeAI saw what no single gauge would: temperatures rising together across the bearings, the wheel space, and the exhaust, all while output stayed flat. Read one by one, nothing alarmed. Read together, they pointed to one cause. UptimeAI traced it to an inlet guide vane that had drifted open, pulling excess air through the compressor and loading the bearings, and pointed the team straight at the vane hardware. They inspected it, found heavy wear, and planned the fix on their own terms instead of losing the machine to a trip. The turbine kept generating the whole way through. Value the customer confirmed: $280,000.

Diagnosing the cause, not the symptom

When a pump at an oil and gas plant started shaking, everyone looked at the bearing. That is where the vibration showed up, so that is where a normal investigation goes. UptimeAI’s Root Cause Agent looked wider. It read dozens of signals together, pulled in the plant’s own maintenance records and years of documents, and built a causal chain in minutes. Its top answer, at 90% confidence, had nothing to do with the bearing. A seal job weeks earlier had been aligned cold, and once the pump warmed up it pulled out of true. The agent even surfaced a write-up on a sister machine with the same story. The team corrected the alignment instead of tearing into the bearing, kept the unit running, and avoided a failure worth more than $500,000. The fix was never where everyone was looking.

When the smartest answer is “it’s not broken”

Not every alert is a real problem, and the safe reaction, shutting down to go look, is exactly how you lose production you never needed to lose. At an oil and gas operator, a bearing alert fired, and the obvious move was to plan a repair and take the unit down. UptimeAI’s Root Cause Agent challenged it, worked through the data, and found the bearing was fine; the instrument reading it was faulty. It cleared several instrument faults in minutes. The team kept the unit running and skipped an outage and a repair it did not need, worth about $750,000.

Right-sizing maintenance, not just cutting it

At a large power plant, the preventative maintenance plan was set by equipment class and almost never revisited. Every pump on the same oil-change clock, every bearing swapped on the same calendar. Revisiting it meant weeks of expert time pulling work histories, so it rarely happened. UptimeAI’s Maintenance Optimization Agent analyzed the plan continuously and provided ranked recommendations to minimize spend across preventative and corrective maintenance. On a single pump, it found four moves at once: do more where failures were slipping through, less where the data proved it safe, add a missing task, and drop a calendar task that live sensors already covered. UptimeAI pushed approved changes straight into the CMMS, no re-keying. One pump surfaced tens of thousands in savings. Extended across the site, close to $300,000, without a reliability engineer touching a spreadsheet.

A capability that sits inside the customer’s team

The real test of a monitoring tool is whether the customer’s own people run it. At a major North American chemicals producer, the engineers approved and closed most alerts themselves, requiring minimal support from the UptimeAI team. When they wanted a second opinion, they asked UptimeAI’s GenAI copilot, Rooty. A pair of catches on a critical pump went onto the maintenance plan before either failed, so the unit kept running. When the team wanted to test UptimeAI’s predictions against their own historical data, it flagged real past failures months ahead of when they actually happened. With UptimeAI assisting them, the site team caught and fixed problems before they ever became problems.

The math underneath

Every one of these stories represents a single line item in the Verdantix calculation of 197% return. The full study lays out where each dollar comes from, an 11-month payback, and how the return climbs toward 250% as the technology is rolled out across plants or sites. If you have ever watched a promising pilot stall because no one could prove what it was worth, a read of the full study is worth your time.

Get the full Verdantix Verified Value Delivery study

Build vs. Buy: Advice for Operations and Digital Executives in the Age of Generative AI

by Jag Gattu, CEO at UptimeAI

This article originally appeared on LinkedIn.

If you’re a CIO, CDO, of VP of Operations and you aren’t asking which of your SaaS tools could be retired in the age of generative AI, you’re missing a big part of your job right now. 

Here’s a small example from our own experience. Last week, one of our team members used Claude Code to rebuild a sales relationship-mapping tool we’d been paying for as a SaaS subscription. A few days of work, and we no longer need to renew that contract. Genuinely great use of GenAI — a well-scoped, single-purpose application, recreated in-house faster and cheaper than re-negotiating the contract with the vendor. 

Here’s what we didn’t try to rebuild: our CRM. Nobody on our team seriously proposed it, and not because it wasn’t technically possible. It’s because scope, scale, complexity, and the sheer depth of integrations a CRM touches make it a terrible build candidate — even in a world where a single engineer with the right AI tools can do more than an entire team could five years ago. Add in how fast the economics of the big model providers are shifting, and “build it ourselves” gets shakier by the month, not sturdier. 

That contrast is the whole build-versus-buy question in miniature. GenAI has genuinely moved the line on what’s worth building in-house, but it hasn’t erased the line. 

And the data backs this up. MIT’s widely-cited 2025 study (enter “95% of AI pilots fail to make it to production”) on enterprise AI found that internally built AI deployments succeed at roughly half the rate of externally partnered ones — 33% versus 67%. Gartner has separately warned that at least half of generative AI projects will blow through budget due to poor architectural choices, and that most organizations attempting to build custom models will eventually abandon those efforts due to cost, complexity, and technical debt. That’s not a knock on any one company’s engineering team. It’s what happens structurally when you take on a mission-critical or business-critical system that has to keep working, keep learning, and keep being right, indefinitely. 

So where does something like UptimeAI fall on that line? For the majority of industrial organizations, we see a strong case for not building it yourself — but it’s worth being precise about why, because it’s no one reason between scale, complexity, risk, and upside.  

Most organizations that try to BIY this space end up building a knowledge graph, wiring up retrieval over their operational data, and putting a chatbot on top. That’s a real project, and a legitimate one — but it’s retrieval. It answers, “what does the data say.” 

Bridging from retrieval to reasoning is a big leap. Ours come with the data access, the knowledge graph, the contextual relationships between assets and failure modes, the domain skills of experienced engineers, and the orchestration across sub-agents already built in — tuned to respond to specific high-impact business problems like “what’s actually wrong, and what should you do about it” the way your best expert would. That’s not a chatbot with good retrieval. That’s domain expertise and engineering judgment, encoded, and scaled. It’s a much harder thing to stand up yourself, and it’s exactly the layer where buying beats building. 

Retire the SaaS tools GenAI can genuinely replace. Just don’t confuse that win with being able to BIY it all. Reasoning and retrieval are fundamentally different jobs — and at UptimeAI we take pride in doing the hard jobs in a way that generates big returns for your business. 

Schedule a demo to see our off-the-shelf reasoning agents for yourself.