An interview with Oli Giordimaina, Chief Product Officer, AI, Lakeside Software

Q: There’s a widespread assumption that introducing AI into the workplace automatically improves productivity and employee experience. From the endpoint data you see, how accurate is that assumption in practice?

A: Honestly? It’s wishful thinking. Indeed, most CIOs have AI on their roadmap, but actual deployments have barely moved the needle. And when you look at the endpoint data we’re tracking (literally millions of devices sending updates every 15 seconds), you can see exactly why: most companies aren’t set up for AI to work well, and bad implementations actually make things worse.

AI doesn’t magically fix stuff. It amplifies whatever you’ve already got going on. So if an IT environment is already a bit of a mess with blind spots and performance issues, AI just magnifies those problems. According to Gartner, data availability and quality remain among the top barriers to AI implementation.

If companies did the boring groundwork first, they’d know what’s happening on their endpoints, could measure everything, and use that to understand where users hit friction. AI requires solid data to work with, not just any data, but high-quality data with both history and context.

The “AI fixes everything” hype has led to a lot of wasted money. The bottom line is that AI can genuinely transform things, but only if you know exactly what’s happening on your devices first. Otherwise, you’re just adding complexity, crossing your fingers, and hoping for the best.

Q: When AI tools start to become part of everyday work, what changes tend to show up first in terms of device performance and user experience?

A: Changes in resource consumption appear almost immediately and are quite dramatic. Across millions of endpoints, AI applications consume CPU, GPU, and increasingly NPU resources in ways that could surprise IT teams. Tools like our SysTrack AI platform provide deep observability into this use, often revealing unexpected patterns.

The first noticeable effects include elevated baseline resource usage: a device previously comfortable at 40% CPU may jump to 65% due to always-on background AI processing. Memory usage rises as models remain resident, and battery life on laptops drops significantly. Users experience lag in applications that previously ran smoothly, often without realising the AI tool is the cause.

Integration friction emerges quickly too. AI assistants requiring repeated authentication across multiple systems, or failing to integrate with core applications, create new points of failure rather than saving time. Application switching increases, and real-time queries slow down previously fast actions.

The consistent pattern is that performance degradation appears before any productivity gains. Organisations deploy AI expecting instant benefits, but the initial weeks frequently show higher resource strain, application conflicts, and frustrated users. 

Q: How do most organisations currently assess the impact of AI on day-to-day work and efficiency, and why do those assessments often fail to reflect what employees are actually experiencing?

A: Most organisations assess AI impact using high-level indicators like adoption rates, licence usage, or periodic employee surveys. From what we see in endpoint data, those metrics rarely reflect what’s actually happening day to day. Usage doesn’t mean productivity, and surveys miss the friction employees quietly work around.

What’s usually missing is continuous, objective visibility into the digital experience itself – how long tasks really take, how often users switch apps, where authentication loops appear, and how AI tools change CPU, memory, and battery consumption in the background. Without a baseline and real-time telemetry, organisations end up measuring assumptions instead of outcomes.

Q: What patterns or signals indicate that AI is adding friction or complexity rather than simplifying work for employees and IT teams?

A: You can spot trouble pretty quickly when you are tracking the right data. One of the clearest signs is when lots of people are using the AI tool, but work is not getting done any faster. In some cases, tasks actually take longer, which is a strong signal the AI is not simplifying anything.

Another red flag is when AI processes are consistently driving up CPU or memory usage without any improvement in productivity. At that point, organisations are effectively taking a performance hit with no real return.

Problems also show up in day to day operations. Anomaly detection often flags unusual resource spikes, application crashes, or users getting stuck in authentication loops, sometimes affecting a small group before spreading more widely. Support data tells a similar story. If Level 1 tickets stay flat but Level 2 and 3 issues start rising, it often means the AI is creating extra complexity rather than resolving it.

You also start to see previously responsive applications slow down, with more not responding errors and increased app switching because integrations are not working smoothly. 

The most concerning signal is when people quietly stop using the AI or stop reporting problems altogether. Adoption numbers may still look good on paper, but endpoint data shows growing numbers of unresolved issues. When that happens, it usually means employees have lost confidence in both the AI and the support process.

Q: As organisations look to improve efficiency and reduce costs with AI, how can IT leaders avoid changes that disrupt core systems or degrade everyday user experience?

A: Visibility first, deployment second. Get proper baseline data on everything before you start, and I mean continuous monitoring, not samples. Track your CPU/GPU/NPU usage, how apps are performing, how long tasks take, where users hit snags, endpoint health, the works. 

Roll things out slowly. Start with a small pilot group, actually measure what happens, and watch for resource spikes, app conflicts, lag, crashes, or workflow weirdness. If you see warning signs, stop and fix them before going wider.

Use AI to watch your AI. We’ve got 220 prebuilt remediation scripts and 1,300 anomaly sensors that catch problems early, even when they’re only hitting 10% of users.

Before you deploy anything, map out how people actually work using your endpoint data. Figure out which apps are critical, where authentication is a pain, and what takes forever. Then design your AI to fit into existing workflows instead of forcing everyone to adapt.

Track metrics that actually matter. Don’t get distracted by vanity metrics like how many queries the AI gets or adoption rates. Keep tracking everything, and have a clear rollback plan if things start going downhill.

Build feedback loops using real telemetry, not surveys every quarter. The organisations getting this right use predictive analytics to stop problems before they happen, give Level 1 support the context they need, and fix issues fast.

Get your data infrastructure solid first; breadth, depth, history, quality. Monitor continuously during rollout. Make decisions based on actual data, even if it means slowing down. The companies successfully cutting costs with AI while keeping users happy are the ones that built strong visibility before doing anything else.

Lakeside Software is the first AI-driven digital employee experience company that works with enterprises to understand how people actually experience their digital workplace using endpoint data. The company launched SysTrack AI last year. SysTrack AI applies machine learning to Lakeside Software’s telemetry to surface patterns and insights that would otherwise be difficult to spot.