How to start an AI service business

AI in healthcare: KenSci's launch

The quest to bring AI's broad potential to healthcare is KenSci's mission. The Seattle-based company uses machine learning algorithms to learn patterns based on millions of data points that patients generate over decades of care. AI analysis can help doctors make better treatment decisions, empower patients to improve their health and provide health insurers with insights that enable them to improve customer service.

To get started, KenSci needed support from research institutions, IT partners, healthcare organizations and highly skilled IT professionals.

Before the company was established, its co-founders and childhood friends, Samir Manjure and Ankur Teredesai, already had the right kind of work experience. Manjure held senior engineering positions at Microsoft across the company's Azure, Bing and Dynamics teams. Teredesai is a professor of computer science and systems at the Institute of Technology at the University of Washington Tacoma. A cancer care research project at that university led to the launch of KenSci.

"The University of Washington project was successful in deepening our understanding of cancer care, but it had a limited focus," said Sunny Neogi, KenSci's chief growth officer. "What Manjure and Teredesai realized is there was a healthcare market that needed a platform to scale and address multiple challenges across the care continuum. That's when the idea of building a platform that could address many illnesses began, and out of this, KenSci was born."

The result is a risk prediction platform that's built on Microsoft Azure Machine Learning, SQL and Power BI.

One key aspect of KenSci's initial development was its partnership with Microsoft, which helped the company tap into an ecosystem of tools and talent, along with cloud technology that provided scalability for the large quantities of data necessary to support the project. Another important part of KenSci's early development was the company's ability to start its operations with a customer in hand.

"We were very lucky that our research papers on the cancer care project caught the attention of medical executives at The U.S. Army Medical Corps," Neogi said. "When they came to us with the proposition that they wanted to embark on a project using our machine learning algorithms for their patient data, we decided to start the company. That was in 2015."

Noting that KenSci's success growing its customer footprint is not typical, Neogi said the attraction stems, in part, from its vertical market focus and the ability to apply machine learning algorithms that address many aspects of the healthcare problem. This healthcare emphasis has affected the company's sales and marketing strategy.

"One of the hardest things for a startup is getting the first few customers," Neogi said. "We were very lucky in the sense that our first few customers came to us. People came knocking on our doors because we have that research background and we understood the depth of the healthcare problem versus just having a shiny AI story.

"Everybody has an AI company these days -- everything that you can do you can slap AI on top of it. But we came at it from a healthcare perspective saying we will use data science to improve patient outcomes in healthcare."

Vectra AI's story

Vectra AI Inc., a San Jose, Calif., company that uses AI to detect and respond to cyberattacks, has its own story.

Like KenSci, the company's founding executive team had previously worked together. Prior to launching Vectra, Hitesh Sheth, Vectra's president and CEO, was COO at Aruba Networks. Before that, he was the executive vice president at Juniper Networks, where he was also general manager of the company's switching business. At Juniper, Sheth met Oliver Tavakoli, Vectra's CTO, who spent seven years as CTO of Juniper's security business.

As Mike Banic, vice president of marketing at Vectra, tells it, Tavakoli had enough experience to know that when a team has too many data scientists, you should look at the problem mathematically. On the other hand, in Banic's view, hiring too many security researchers could result in looking for threats that come from bad actors with specific domains and IP addresses. The bottom line: Many of these professionals often miss cyberattacks on their systems because they aren't looking in the right places.

Mike Banic

Tavakoli's strategy, Banic said, was to pair a data scientist and a security researcher -- as both have specific skill sets that, combined, can create highly efficient machine learning algorithms.

"If a team of people decide to enter the AI space to start a business, understanding how to build a business team is vital," Banic said.

Vectra's technology was developed out of a need to find new ways to use tools, like machine learning, to understand patterns in network traffic and detect unwanted or malicious behavior in the network. The company's Cognito platform uses supervised and unsupervised machine learning algorithms that often utilize deep learning techniques, such as recurrent neural networks and long short-term memory.

Vectra's technology has gained recognition in Gartner's Magic Quadrant for intrusion detection and prevention systems. But Banic said there was a lot of skepticism as to whether the company would gain a foothold in the industry. Part of the problem was getting customers to hand over their data so that Vectra could prove the value of their software.

"In those early years, we only had data from published reports or public sources," Banic said. "To get off the ground, we decided to go to people that we had strong relationships with, and we asked them to allow us to put our software in their network environment to collect anonymized metadata in order to build these AI models."

Many of the organizations that allowed Vectra to use their networks to access the data for free are paying customers today, Banic added.

Data analytics skills are critical for companies looking to launch an AI service, said Carolyn April, senior director of industry analysis at CompTIA.

"If you are able to harness that data, analyze it and then bring it back to the customer with some actionable recommendations, you've got yourself a great business consulting practice," April said. "The problem, I think, for IT service providers has been that they have been very slow on the uptake when it comes to becoming skilled with data analytics even before AI. We continue to see that learning curve be a bit of a challenge."

SADA's approach

While some AI companies have launched their AI capabilities by focusing on a single vertical or a specific IT problem, others have gradually grown their AI service offerings in response to the business needs of existing customers.

That's what happened in the case of SADA Systems Inc., a Los Angeles business consulting and technology services firm specializing in Google Cloud. The company provides a range of services to Google Cloud users, including assessment, planning, implementation, customization, development and change management.

We moved cautiously at first.

SADA, which launched in 2000, transitioned to an AI offering in three stages, said Dana Berg, the company's COO. That shift began six years ago when customers using its business intelligence and analytics solutions started asking questions about how their businesses could benefit from AI. That initiated stage 1 of SADA's journey toward an AI practice.

"We moved cautiously at first," Berg said. "We were aware of AI's relevance and we knew the technology would be at the forefront of data analytics, but as an organization, we were still building out a formalized practice with a go-to-market strategy."

Stage 1 involved educating customers on AI and how the company's business intelligence and analytics solutions could benefit from AI capabilities further down the road.

Dana Berg

"We needed to make sure that customers knew our solutions would ultimately compliment AI models. We sent our staff to train with all the various cloud providers, as well as our AI partners, to help them understand what it would mean to really be an AI company," Berg said.

In 2015, two years after beginning the process, SADA took advantage of advancements in technology to build momentum for its AI vision. With the power of cloud computing and sufficient storage capabilities, AI became easier to adopt, making it a technology that could address a plethora of use cases across the enterprise.

At that point, SADA moved to stage 2, a proactive AI strategy, in which it hired data scientists and software engineers and created the company's first AI projects developed specifically for customers' IT enterprises.

"For SADA, a proactive strategy meant we were at a maturity level where we had a strong level of education among our staff that allowed us to identify use cases and AI projects that addressed the problem, showed vision and revealed the art of the possible," Berg said.

Berg added that, over time, service providers can use AI to establish themselves within certain specialty areas, such as merging AI and IoT applications.

"That's when I think stage 3 happens -- when you have a very specialized strategy that is unique for every single customer. At this stage, you are building a specialized AI go-to-market strategy, introducing new sales pitches, marrying your other competencies and making artificial intelligence very unique both to you as a service provider, but also to your customer base," Berg said. "That's when you are going out and actually leading with artificial intelligence as a single distinct type of capability that you can go to market with."

Editor's note: This is the first of two articles on developing an AI service business. This article focuses on how to launch an AI practice, while the second part examines how to expand an AI business once it has been established.