The Impact of Remote AI on Smart Home Technology

Artificial Intelligence (AI) is changing our world at a dizzying pace. It promises to improve lives and make us all better, faster, and stronger (or maybe unemployed!). I spend a considerable amount of time studying where AI might impact the smart home, particularly in the arena of remote monitoring. While this category is still in its infancy, it promises to affect three key areas in the coming years. We can call them “improving the signal-to-noise ratio,” “self-healing,” and “moving from artificial intelligence to intelligence assistance (IA).”

Improving Signal-to-Noise Ratio

Depending on who you ask, remote monitoring notifications from services like Domotz, BakPak, and OvrC are anywhere from one-to-10-percent effective. That’s like saying your car will only start three to 30 days out of each year. Integrators on the receiving end of that noise tend to tune it out like the boy who cried wolf. This renders remote monitoring tools worthless without doing a ton of custom configuration. If everything’s important, nothing’s important.

It’s possible to dial in these tools to deliver more meaningful results, but it requires time and expertise. We’ve seen only a small percentage of current integrators willing to invest. Companies like the video conference monitoring platform VisibilityOne saw solving signal-to-noise challenges as paramount, driving the creation of their product.

“We had a client video conference for over 3,000 people, and the call dropped,” said Von Bedikian, co-founder and COO of the company. “Their CIO called me that afternoon and was very upset. Thanks to our software, we were able to instantly identify that the issue lay with the ISP – not us.”

Solutions like VisibilityOne can keep an eye on much more than whether or not a device is on the network. Video connection quality, component temperature, and battery life are just a few of the parameters that are measured and analyzed to deliver more uptime for customers.

Even after a remote monitoring tool delivers a meaningful result, you still have to decide what to do with it. This labor-intensive work is out of reach for many integrators to do themselves. Therefore, we see the rise of remotely managed service providers (MSPs) like Parasol and OneVision Resources. (Full disclosure: I’m one of the co-founders of Parasol.)  These MSPs handle subscriber support issues from network operations centers staffed around the clock via phone, text, email, or chat.


AI promises to help with the triage process of tech support, ultimately leading to self-healing systems. After delivering meaningful results, AI can be trained to perform routine tasks. These include rebooting devices or acting as a first layer of technical support before a human is required. Because professionally installed smart home technology customers have such high expectations for their uniquely designed systems, that’s a tall order. Integrators and MSPs are reluctant to reboot or fix anything remotely without taking into consideration what else might be affected.

“We have a ‘Do No Harm’ rule similar to the Hippocratic Oath,” said Ted Bremekamp, one of my fellow Parasol co-founders and the company’s director of operations.

How do you train a machine to observe a “Do No Harm” rule? As the old adage goes, you only need three things to be successful: “practice, practice, practice.”

Scudo Labs thinks they’ve solved both issues with their Scudo Box. The Scudo Box leverages a large database of “heartbeats” (unique to each product) developed across hundreds of home technology devices.

When the Scudo Box sees devices like an Apple TV or cable box failing, it can reboot it automatically. “Devices don’t usually fail suddenly, they fail slowly over time,” said Jason Blais, Scudo Labs’ lead engineer. “We want to be able to predict product failures down the road and proactively notify our clients, reducing support calls and creating revenue opportunities for integrators.”

Artificial Intelligence (AI) to Intelligence Assistance (IA)

Thomas Friedman’s book Thank You For Being Late has some insight here. It describes a key inflection point when AI starts to help workers make informed decisions at a faster rate. This frees up opportunities to spend more time working on ways to improve their industries. In remote monitoring, this dovetails perfectly into helping integrators. Its asked them if they’d like to automate routine behaviors, or power cycle a system that may affect other parts of a home. Each time the integrator replies back to the AI, the machine learning improves, delivering better future results.

Companies from industries of all stripes are working hard to create basic AI enhancements to their businesses. Typically, they train on routine tasks first and then move on to more complex operations. It’s not uncommon to hear examples of learning AI applications in call centers recording customer service representatives. Programs listen to these takes and eventually try to solve the issue itself while being corrected by the human “trainer.”

Solutions like Zendesk’s Answer Bot are great examples of early AI at work trying to better assist support professionals. While customer support agents help customers, Answer Bot offers up suggested solutions in real-time based on chat or email sessions. By measuring the efficacy of its own answers, Answer Bot gets better and better each time using machine learning.

So What’s Next?

While AI might be just getting started, we have a long way to go. “Eighty percent of our customer issues aren’t solved with a simple reboot,” said Joseph Kolchinsky, founder and CEO of OneVision Resources. “Today’s support data aren’t clean enough to draw meaningful conclusions using AI.”

Remote monitoring solutions will be some of the earliest beneficiaries of innovation in the CI channel. Plus, it looks like a human/machine-friendly partnership is much more likely than an outright replacement of talented technology professionals. I’m excited about the possibilities and will be positioning my technology integration company to take full advantage.

This article was originally published by Henry Clifford on