Sandvine and the Price of Seeing Everything

The internet can be working perfectly and still be unbearable. A connection is alive. Packets arrive. Somewhere, a video call freezes on the least flattering frame of your face. For the person paying the bill, the distinction between a functioning network and a functioning application is academic. For Sandvine, it became a business.

The revealing question is what happens after a company makes something visible. An operator can use the knowledge to improve a call, delay a costly upgrade or design a family phone plan. A government can use network controls to restrict access to information. Sandvine’s history runs through both possibilities. It is an unusually concrete story about the distance between an engineering feature and a business decision.

Five men, one van, a traffic problem

There is a photograph from August 31, 2001, in which Sandvine’s founding team poses around a Volkswagen van bearing the company’s name. Don Bowman, Tom Donnelly, Dave Caputo, Marc Morin and Brad Siim look considerably less formal than the networking industry’s usual diagrams. All five were alumni of PixStream. The company they launched in Waterloo would spend its life trying to understand the applications sharing a broadband connection.

The economic logic is easy to grasp. Capacity costs money. Customers judge service through applications. Knowing that a link is busy tells an operator less than knowing which application is struggling, where, and under what conditions. Sandvine sold a way to make those distinctions, along with tools to act on them.

By 2017, the business had attracted a takeover agreement from a Francisco Partners affiliate at C$4.40 a share, implying approximately C$562 million in fully diluted equity value. The acquisition closed that September, and Sandvine combined with Procera Networks under the Sandvine name. This was a purchase of an established network-intelligence business, not a venture funding round.

The app inside the app

The useful detail in Sandvine’s technology is smaller than the application icon. Facebook video and Facebook messaging are different activities, with different demands on a connection. Treating them as one undifferentiated bucket misses information that an operator could use. The AppLogic classification engine is designed to identify those content types as well as the application itself.

AppLogic combines machine learning with application signatures. Its current product page advertises coverage of at least 95%. That is a company claim, not a guarantee that every operator will see the same result. Applications change, encryption evolves, and the classification library needs updates. The expertise lives partly in continuing to recognize tomorrow’s traffic, rather than merely identifying yesterday’s familiar names.

The product names describe successive jobs. AppLogic classifies. ActiveLogic supplies inline visibility and traffic enforcement. AppLenX enriches flow records with application context. Maestro Policy Engine turns rules into network behavior. Operations Insights and Enterprise Insights organize the data for people who must investigate complaints and incidents. Professional services help customers put these capabilities to work.

This puts the business between application classification, traffic management and network assurance. Allot, Enea Qosmos, ipoque and NETSCOUT address overlapping parts of those markets. The comparison depends on what the buyer needs. AppLogic Networks’ pitch connects classification to experience scores and inline action; that is a useful distinction to evaluate, rather than proof that its software wins every comparison.

When visibility became a liability

In February 2024, the United States added Sandvine to its Entity List. The State Department said its technology enabled human-rights abuses. The Commerce Department later described mass web-monitoring, censorship and targeting of activists and dissidents, including misuse of commercial spyware. These were findings stated by U.S. authorities, with consequences for the company’s access to U.S. goods and technology.

The failure was in the conditions under which powerful technology was sold and used. Explaining an application does not require reading every encrypted message. Even so, classification and control can affect what people are allowed to reach. That distinction matters: a product can raise serious rights concerns without possessing every capability people associate with the word surveillance.

“The right answer at Sandvine is always to do the right thing.”

Dave Caputo, in court testimony reported by Communitech

The founding maxim became a difficult measure against which to read the later history. By September 2024, Sandvine announced acquisition by a group of U.S. investment firms that had been its lenders, a significant debt reduction and a new capital commitment. In October, BIS removed it from the Entity List after reforms. BIS recorded 32 country exits and another 24 in progress, alongside ethics oversight and human-rights expertise.

Delisting did not finish the financial repair. Court-supervised CCAA restructuring followed. The operating-business transfer closed on March 3, 2025, and new entities emerged under AppLogic Networks. Mark Driedger became CEO. The successor described new owners, leadership, funding and a software-only strategy focused on democratic jurisdictions. Calling this a name change alone would leave out much of what actually happened.

A DNS server offers a better demonstration

A smaller story from the successor’s work is more useful than another sweeping promise about AI. In its account of a live trial at JANOG57 in Osaka, published in March 2026, AppLogic Networks described a DNS server that stopped responding. Operators could see incoming queries without outgoing replies. That helped confirm the problem. The unanswered request supplied the clue.

The trial also streamed data into Grafana and used Juniper Mist access-point information to add location context. The attraction is practical: application intelligence can contribute to tools a team already uses. The company’s report offers a specific operational example, though it does not establish the savings another deployment would achieve.

AI AppAnalyst adds natural-language questions to this investigation process. Its published deployment requirements include a cloud-based language-model connection and compatible software versions. Customers can choose whether to send query results to the model for narrative analysis. In the alternative mode, the model helps generate queries without receiving those results. Buyers have a concrete architecture choice to inspect.

The bill, and the boundaries

The customers are businesses running networks: fixed and mobile providers, cable and satellite operators, and increasingly enterprises, managed services and cloud providers. C Spire’s 2025 parental-control announcement shows another use for the technology: packaging filtering and family controls into a consumer phone service. The announced integration is evidence of a commercial relationship, not a measured outcome for every family.

Revenue comes through software licensing, support, update subscriptions and services. A buyer must weigh those charges against integration work, infrastructure and the cost of maintaining accurate classification. An operator should test whether the proposed intervention changes an actual experience problem. A busy link, an unhappy customer and a compelling chart do not automatically establish the same cause.

There are useful lessons to copy. Investigate experience at the application level before assuming capacity is the answer. Add location and user context where it clarifies an incident. Keep human review of AI-generated investigations. The approach offers less value when the observed traffic cannot explain the complaint, when classifications go stale, or when a proposed policy cannot affect the underlying bottleneck. More precise observation still needs a sound decision.

The successor also has to demonstrate its boundaries. In January 2026, it reported human-rights due diligence, mandatory training and Global Network Initiative observer status. In June, it said it had contributed 1% of its 2025 annual profits to Freedom House. Those are specific commitments and actions. They deserve continuing examination alongside customer selection and product use.

The company began by making the network less mysterious. Its harder assignment is now to make the exercise of that knowledge less mysterious too. Anyone buying its software should want an answer to both questions: what can you see, and what happens next?