'What if we have a magic algorithm?'

Graduation Day 2019

When magic starts, craziness follows. This is so true for EIT Digital Doctoral School graduate Abdelhadi Azzouni. After discovering a ‘magic algorithm’, founding the startup Packet AI in 2019, and receiving €2.1 million in funding, things got crazy. Now, he is working overtime to launch his product, managing more than ten staff members, and planning to expand to other European markets and the US.

Packet AI is the French deep tech brainchild of the EIT Digital Doctoral School graduate Abdelhadi Azzouni and his co-founder Hardik Thakkar. Packet AI provides an Information Technology Operations (ITOps) - more precisely AIOps - solution for companies to predict IT incidents. “We help companies to have more uptime and less downtime.”

Abdelhadi explains that most businesses nowadays suffer from IT incidents and inconsistencies happening all the time. “The more complex the infrastructure is, the longer it takes to find the root cause and fix it. On average, this is between 30 minutes and three hours.”  According to Abdelhadi, this can immobilise a business - which costs money, both direct and indirect.” Think of the hidden cost as the time engineers are working on fixing IT issues, instead of building value for the company. We want to give back that time to the engineers.”

The question is how. And to that question, Azzouni found an answer based on his PhD thesis titled Smart and Secure Network Virtualisation – using machine learning to optimise networks at large scale, at the EIT Digital Doctoral School at the Sorbonne University and the University of Waterloo in Canada. “I worked on the perfection of the routing algorithm to predict the behaviour of internet users in terms of traffic and data flow.”

After finishing his thesis in 2018, he went to do a six-month business development experience (BDx) through the EIT Digital Doctoral School. First, he went to the Paris incubator Agoranov to see if he could build a product based on his research. “That did not work. The market was not ready.” He let go of his initial thought and continued his BDx at the British incubator Entrepreneur First (EF) in Paris, where he met his co-founder Thakkar.

Searching for his Eureka moment, he discovered the EDGE framework that helped him define a goal. “When people start a business, they tend to want to find a solution to a problem. They raise money, work hard but don’t sell. It shouldn’t be the case. We learned to apply the EDGE framework. It is about finding the edge of the founders, doing something that most people do not know. We started with what we can do. Because we were not sure of which problem we had to solve yet, we listed a set of possible problems, then defined which is the most pressing one and asked  around in the market for possible interest.”

Solution

This problem was the inevitability of IT incidents in complex infrastructures. ”From discussions with prospective clients, we learned that they have a huge need today to help with the complexity of IT infrastructures.  We thought, what if we have a magic algorithm that proactively in real-time predicts the smallest deviations, we would have a business. The key to the solution was finding algorithms to predict incidents. “It is not very different from my PhD research.” 

Market-driven approach

They then followed a market-driven approach. In so doing, they positioned themselves as the company they wanted to be.  “We were talking eight hours a day on the phone with people from banks, e-commerce websites, IT-managers, CTOs, engineers. People are nice. They will not criticise something that doesn’t exist yet and will encourage you to go for it. It can be fatal to build something the market doesn’t want to buy.  Thus, we talked about the product as if it was already there and asked, ‘Will you be interested in trying it?’ If then they invite you to come over, you know you are on the right track. People will not want to spend an hour on nothing.”

This kind of approach: a ‘shortcut to get an answer from the market’, also led to some pressure when an invitation came in. What to show if you do not have a product yet? “Each time we had a good response, we built prototypes to show results, like details of the root cause of IT incidents. Clients like that.”

Proof of Concepts

In December 2019 and January 2020, the founders made two proofs of concepts with two large French companies, whose names cannot be mention yet.  “With this, we had enough signals that people were interested in our solution.” Azzouni and Thakkar officially launched Packet AI in January 2019. EIT Digital provided them with a workplace at the co-location centre in Paris.

Packet AI

From January to March they were focused on building a product that would be compatible with platforms like Amazon web services, Azure, Google Cloud, OpenStack, Vmware, Kubernetes, and Docker

Their product models the normal behaviour of a client’s IT infrastructure, so the algorithms can spot anomalies. The ‘magic algorithm’ basically consists of a packet of algorithms that first do real-time anomaly detection, including Natural Language Processing, to spot the smallest deviation. The next step is the root cause analysis. Because complex systems might have a 1000 anomalies per hour, there are additional algorithms in place to keep it manageable. There is, for example, an algorithm that estimates the impact of each anomaly on the future of the system, and one that ranks all anomalies on the probability of occurrence and priority. “Based on this, we can tell for example that there is a 90 per cent chance that within 25 minutes you have a major latency on database X on server Y because of the misconfiguration of A and B in library Z. IT managers can thus prevent a failure from happening, before it impacts millions of users and business operations.”

Timing

The solution they built sounds very logical. People have been attempting to come up with IT-incident prevention for a long time. Azzouni says that now the time is there to get a solution like this in place. “Two factors allow us to build this today. One, computing power is cheap now. You can buy huge servers for hardly anything. The second point is the open sourcing of major Machine Learning projects, like Google’s TensorFlow. It democratised a branch of algorithms in machine learning that require computer data.” Also, Azzouni sees that within the IT world, all big organisations are moving their IT infrastructure to the cloud. “In the moving process they face a lot of pain, incidents. It is a perfect time for us!”

€2.1 million fundraising

From April to August, the co-founders put their efforts into fundraising - with success. In September they raised 2.1 million euro from different parties, such as venture capitalist Aster Capital, BNP Paribas Developpement, Entrepreneur First and SGPA. “There is also some money from the government.” The money raised was for finalising the product for first public release. This was planned for Q2 of this year, but due to the Corona crisis, it has inevitably been delayed a little. Azzouni now expects the launch to be in August this year.

The second goal was to hire people, like machine learning engineers, machine learning researchers and web developers. “We now have 13 engineers, and will have four more in the next months. We grew very fast and we also need to go fast. Therefore, we were focused on hiring senior people, people with experience in startups. €2 million seems a large amount of money, but it is not. It allows us to try something for the coming two years. If we do not succeed, we should raise more money, or the company will die.”

Teambuilding

Building a team is difficult, Azzouni found out. “Managing smart engineers is challenging. We hired people who know something that we do not know and now we need to lead them. In the beginning, it is easy. But step by step, each one is working in their own way, losing sight of the big picture of things. Transporting our dreams to them is hard. It has changed me. I became more managerial. If you have twelve different opinions confronting you, you have to convince them all the time to work in line with a unified vision of the company.”

Customers

In the meantime, a pipeline of customers is waiting for the product to be launched. The software packets of algorithms will be sold as a software as a service product (SAAS) on a subscription basis. When the first customers are happy, Packet AI will directly scaleup, Azzouni says. “If we know the first version can be handled, the true business will start. We will expand across Europe. First to the UK and Germany.” In parallel to a further out roll in Europe, Azzouni wants to cross the ocean and make it big in the US.

EIT Digital Doctoral School

His aspiration to start his own business was dormant in his mind, but got unlocked by the EIT Digital Doctoral School, he explains. “I went to an EIT Digital Summer School in Stockholm and followed seminars. And I met a lot of interesting people. I know people in Finland, Scotland, Italy, and all across all EIT Digital hubs. If I will expand to Italy or Finland, I would just have to call the offices there. That is a huge addition of the EIT Digital Doctoral School programme. The network is amazing!”

© 2010-2020 EIT Digital IVZW. All rights reserved. Legal notice