Global
EU
USA
PA
PAK
MEAAn IT team of five people supporting four hundred employees spends a good part of the week answering the same question: why is the Wi-Fi slow? Sometimes the cause is an overloaded access point, sometimes a client running an outdated driver, and sometimes a DHCP service that runs out of addresses mid-morning. Finding it means opening consoles, cross-checking logs and testing hypotheses. Marvis, the virtual assistant in HPE Mist, was built to shorten that journey, and the interest it is generating among infrastructure managers justifies a calm look at what it does and what it does not.
How a virtual network assistant diagnoses problems: telemetry, root cause analysis and Shapley values
Marvis was introduced as the first virtual network assistant on the market and is now part of the HPE portfolio. It runs on a cloud platform that receives real-time telemetry from access points, switches and WAN devices. HPE stresses that this telemetry arrives complete and continuous, without sampling, and that it is brought together in a single data model alongside information about users, devices and applications. Without that foundation, any AI layered on top has little to work with.
On top of that data, Marvis performs root cause analysis using Shapley values, a method that comes from game theory. Its core idea is to distribute the credit, or the blame, for an outcome among all the factors that contributed to it. Applied to a poor Zoom or Teams call, the system does not simply report that the call dropped. It calculates how much each element weighed, whether that was signal strength in that area, the load on the access point, the state of the internet link or the user’s own device, and ranks the causes by their contribution. For a technician the difference is significant, because they start from a prioritised list instead of a hunch. It is worth adding that the tool explains its reasoning rather than just delivering a verdict, something every team should demand before allowing it to act.
Marvis Minis and natural language queries: finding the fault before the user calls
Most monitoring systems wait for something to fail. Marvis Minis work differently: they simulate real users who connect, obtain an address, resolve names and reach applications, and they repeat that test periodically at every site. If an authentication service stops responding at six in the morning, the fault shows up on the dashboard before the workforce arrives. They do not replace a person sitting in front of a laptop, because they only see the path through the network, but they bring the warning forward and that reduces calls to the helpdesk.
The third layer is the language model. HPE has integrated LLMs into the platform so that technicians can ask questions in their own words, such as why María has a poor connection on the third floor, and receive an answer based on the data from that network. The language model acts as an interface: it interprets the question and presents the result, while the diagnosis comes from the telemetry analysis described above. It also helps with documentation queries and configuration questions without leaving the console. This design makes sense because it limits the risk of invented answers, although no generative system is free from error and human review remains necessary.
What changes in a hospital and in a school when the network monitors its own behaviour
The two scenarios below are illustrative. They have been built from the capabilities described and do not correspond to specific customers.
In a mid-sized hospital, a patient monitor on an inpatient ward starts showing higher latency than usual. Nobody has noticed yet and the device is still sending data, but its behaviour departs from the pattern the platform has learned. Marvis identifies the device, correlates the anomaly with the rest of the network and, following the defined policy, moves it to a restricted segment so the problem does not reach the clinical systems it shares infrastructure with. The IT team receives an alert with the probable cause and coordinates with biomedical engineering to inspect the device. There is an important nuance here: in a clinical environment, isolating a device is not a neutral decision, so the policy must be agreed in advance with clinical staff, and HPE advocates an approach in which the team can review changes before they are applied.
In a school, teachers in one classroom have been complaining for weeks that video lessons keep breaking up. Marvis cross-references the data from the video conferencing application with network data and detects that video traffic is competing with downloads from the pupils’ tablets on the same access point. The system proposes a policy adjustment that gives priority to video conferencing in that classroom during those hours. If the school has enabled automatic application of changes, it is carried out without anyone needing to open the configuration. If not, the person in charge approves it with one click. In both cases nobody has had to spend an afternoon hunting for the cause by hand.
What 90% fewer tickets means for an IT department of 3 to 10 people
HPE publishes striking figures: up to 90% fewer support tickets, up to 85% fewer on-site visits and deployments up to nine times faster. These are vendor figures and in every case they are ceilings, not promises. The actual result depends on the starting point. A well-maintained network has less to gain than one that has carried years of recurring incidents.
For a team of three to ten people, the value of these numbers lies not in the percentage but in the time recovered. In a company of this size, the same person is often systems administrator, network manager and first-line support. If out of every hundred weekly incidents thirty relate to connectivity, halving that share frees up several hours that are currently lost to interruptions and could go to pending security or business continuity projects. The savings on travel matter especially for companies with branches spread across several cities or countries and no dedicated technician at each one, because every avoided visit means hours of travel and a day not spent on other tasks. And faster deployment means that opening a new office comes down to shipping the equipment, plugging it in and letting it configure itself from the cloud.
As with any HPE Aruba Mist solution, performance depends on what is connected to the platform. The more complete the wireless and wired estate managed from Mist, the better the visibility. A mixed network with equipment from other vendors limits what the assistant can see. Before deciding, it is worth measuring for a few weeks how many network tickets are opened and how much time they consume, so there is a baseline to compare against later.
How to assess this technology with an advisor before committing to an investment
Every company starts from a different network, with its own sites, legacy devices and budget constraints. The teams at Beyond Technology work with this platform and can review with you which part of your infrastructure would benefit most from an assistant of this kind, which figures are realistic for your case and how to plan a gradual adoption. If you would like to talk to an advisor, get in touch through beyondtechnology.net and we will arrange an initial conversation with no commitment.
