When it comes to the Big Data Analytics space – Guavus is playing a pivotal role in delivering a unique value for communications service providers (CSPs) and Industrial IoT. The company’s products handle very large volumes of data that’s sourced from subscribers and devices on CSP networks and applies AI-based analytics to this data, delivering critical insights and key value-add use cases that CSPs seek. These use cases ultimately aim at maximizing their subscribers’ experiences.
Stephen Spellicy, SVP of Products & Marketing at Guavus (a Thales company), speaks with TelecomDrive.com about AI, analytics and IoT, and how Guavus is adding value for CSPs today.
What’s the state of play in the IoT domain, what kinds of AI and analytics use cases do you see really taking off?
With IoT, the CSP use cases typically break down into 4 major areas, where we see the most traction:
- Monitoring – business-intelligence-style activities where you generate reports for certain key operations performance indicators.
- Profiling of the devices themselves – what is it spending most of its time doing, etc.
- Predictive Maintenance – interpreting the behaviors of the devices to predict the need for unplanned maintenance.
- Device Security – interpreting the behaviors of the devices to determine if they are engaging in suspicious activity
Which of these areas offer the greatest value to companies today in your opinion?
All have business value. However, generally speaking, as you increase the sophistication beyond monitoring, the ROI increases.
Monitoring is clearly the most mature area to-date as it yields immediate value, is comparatively simple to do and everyone understands it. This informs capacity planning, investments, preventative maintenance, etc.
In predictive maintenance we are focused on failure prediction. If it’s going to fail, you want to diagnose what’s going on. With such analysis, it’s hard to do as you can’t touch these devices. After you’ve found the root issue, a key cost saving results from figuring out if it can be resolved remotely rather than incurring the expense and delay of direct intervention.
How do you determine if a device will fail, what kind of indicators do you have and how is your system set up to monitor that?
A simple analogy might be our laptop devices. For example, your fan on your laptop is constantly on. Typically the fan is on with high CPU usage. If we see it is always on even when CPU usage is low – what does this mean? Is this normal behavior or is something wrong such as with the fan itself? So, these KPIs help you tease out abnormal behaviors and also provide you with diagnostic information. The KPIs themselves include the device’s ever-changing physical characteristics, its upstream and downstream traffic frequency, amounts, types, etc.
In addition, devices log their activities. These are interpreted to add to their behavioral profile and typically provide more contextual information about what might be going on in the device.
Are you hardware agnostic? Who do you work with in the IoT ecosystem?
We are hardware agnostic. Let me go back to how companies get data. There are 2 principal ways of getting this data. Observing the device in its operating environment by looking at the device doing its job, collecting the data, and monitoring use cases. IoT devices were never designed for that purpose, to manifest that data so you need to infer that from the network. Guavus has spent a lot of time looking at that, looking at the periphery of the device and what’s happening around it. And by virtue of doing that, we’re hardware agnostic.
Now some manufactures have started instrumenting their devices to directly provide key performance indicators. For example, your cable modem at home – you can log in to see the signal, metadata can indicate if the device has failed, and so on. We’re getting these things in real-time, so we are then able to understand what’s happening with the device.
We’re uniquely positioned in our space and it gives us the opportunity to work within the ecosystem. Those IoT devices are like your cell phone connection, like any other device on your network. Our technologies have been enabling the improvement of the performance of devices in CSP networks for more than a decade. So, technically, it’s just a straightforward extension for us. We have that as part of our DNA.
Are there additional issues you see for CSPs and their customers?
The majority of IoT devices, especially those in smart city and industrial sectors, don’t require the same speed and bandwidth as consumer cellular devices. However, power consumption can be an issue where analytics can be applied, especially to enable the understanding of battery consumption for long lifecycle devices, such as smart meters. Excessive network communications can prematurely drain battery resources.
Analytics can assist utility operators to reduce and refine the data collection frequency to limit impact. In addition, utilities can better protect their grids with an understanding of device behavioural patterns to identify potential intrusion or infiltration.
On the opportunity side of the equation, IoT technology can help make our economy and society more sustainable. IoT can be leveraged to save energy in the long run. Smarter objects mean more efficiency.
Collecting, analyzing, and measuring the behavioral aspects of IoT devices will enable societies to finetune their energy consumption and can be used to reduce the impact on the environment. An example use case would analyzing the data from traffic sensors. By doing so, cities and governments can measure, then predict, vehicle traffic patterns to better optimize commuting at peak times. Beyond that, insights into the data can be used to model the impact of carbon emissions on specific geographies. Regional governments can leverage this data to apply and enforce vehicle traffic policy. Smart cities thus become smarter, more responsible societies. That is a win-win for us all.
Some companies want machine learning and AI but forget that they need data, and how to get that data. Is that a common problem you run into or do your customers already understand what data they need and how machine learning/AI fits into their solution?
People intellectually understand data, but from an operational perspective that’s where people are lost…where do you get the data. We have a consultancy approach to this with our customers. It’s a fundamental paradigm shift that we try to get across to our customers. Normally people think of technology as in/out — performs a specific function and that’s clearly understood.
Yes, analytics are primarily delivered through software but there’s also the data. Machine Learning and Artificial Intelligence are designed to take that data and turn it into value to a human or downstream to optimize itself. The nature, statistics and quality of the data, relative to the information you need, is a new dynamic in addition to the otherwise deterministic functionality of the software.
In the IoT world, data matters now more than ever…
Definitely. But data is an alien concept…what do you need in the context of that data? What is the actual info you’re seeking and what data do you need to support that, how can you find it (executing processes in the org that can pull data)? In some cases, it requires going back to the vendors of the IoT devices to make changes in their systems so they’re providing the right data. Even if machine learning is not in place, you still have these problems.
A lot of our value is in helping customers to think through those challenges – what’s the use case, how does it relate to business value, what info. is needed, what data is needed from which to derive that info, where in their organization can they get it, how can they ensure its of quality, curate it, etc. That’s part of our collaboration with our customers.
Isn’t security an issue as well?
Yes it is. In the world of IoT, it’s important to have robust security measures. In critical environments where there’s no margin for error – in industries such as transportation, defense or cybersecurity — solutions can’t be the same as those developed for the general public and trust is crucial. It’s our responsibility to help customers build a trusted environment by providing for instance strong encryption and secure authentication solutions for data, devices, networks and platforms.
Any data flowing through a network-based system needs strong encryption and authentication technology to ensure integrity. Furthermore, the identity of each user or device needs to be authenticated and verified. Every part of the ecosystem requires trust implicitly. This is why Thales adapts technology – and more explicitly AI through our Thales TrUE AI approach – to the constraints of the environments in which our customers operate, where critical operations require safety, responsibility and compliance with standards, laws and ethical principles.
Trust comes only if users or devices can be identified, their identity authenticated, verified and we can understand and explain what the decision is and why we the decision has been taken. These principles extend beyond network entities such as users or devices to the actual integrity of data that is collected and analyzed. This is why the entire end-to-end operation of network communications requires security at its foundation and core — all the way up to the top of the stack to the data interaction layer.
Machine learning and AI are used as buzzwords a lot and there’s a lot of confusion. How do you define them, what’s the distinction?
In terms of distinction, the key word is intelligence. It’s like if you are walking on your favorite hiking trail and see an army of ants going in one direction and another army going in another. You might at first think they’re intelligent, but on second thought realize they’re following pheromones and matching them with their reflex reaction until they get to a food supply. That’s just pattern matching – that’s machine learning, it’s trained to learn patterns…and coming up with a mathematical model, and every time data comes in, it runs it through this model.
Contrast this with crows, which do problem solving…they’ve been known to drop stones in vials of water to get food to rise to the top. That’s transferring an experience they’ve done into a reasoning process, extrapolating to create hypothesis, test, improve and ultimately achieve an objective – inventing completely new information/patterns.
You’re applying reasoning and predictive analytics so customers can take prescribed actions?
Yes. The applications of analytics to IoT fall into two basic areas: the application of the device itself, like a smart thermostat, so that the manufacturer can leverage data to know when to turn up the thermostat, for example. Then there is the data that runs over the network — so the network provider has to have the smarts in the system to ensure continued connectivity, to ensure the device is performing, not going to a strange IP address (for security reasons)…that’s the network layer of the IoT device and that’s where Guavus specializes.
How critical do you think AI and machine learning really are to IoT? What’s their ROI?
The interesting thing about analytics is you can do it offline and show the potential functionality of the analytics of machine learning and AI and the calculated ROI from it. But, in the end, most customers want to see it trialed in their system and see the ROI at the end of the quarter. These applications aren’t just about processing but translating the results of AI and then customers can use them in their day-to-day jobs.
We have lots of predictive analytics, 24 hours in advance, that a device will fail but then Guavus makes sure it gets into the right hands so action can be taken. The ROI then can be completely recovered because of that. If it sits on someone’s screen, it loses that value.
What does the future hold?
We’re focused on business value problems we can solve, and we can leverage artificial intelligence/machine learning for that purpose. We’re driven by the current problems our customers are facing. The holy grail is to make these devices autonomous, so they can recognize when there are issues using a knowledge base on their own – in other words, self-optimizing systems.
Bear in mind these things don’t happen overnight…we’re slowly getting there, that’s the nature of technology. For certain devices and their operations, it’s a piece of cake – you just need to update the firmware or reboot the device — we’re already at that point today. But then there are the interactions of the network — that will be next, and the way companies deal with the network will change. It will evolve over the next 3-5 years towards doing its own optimization and maintenance.
The interview was first published inside the latest issue of Disruptive Telecoms, a TelecomDrive.com initiative
