Part of:The ways and reasons to outsource analytics
6 reasons you may need data science as a service
There are plenty of reasons to outsource all or part of a data science project to a service. Find out how enterprises are using DSaaS in their analytics pipeline.
Data science as a service, or DSaaS, has been old news for a while, but it has become a hot topic once again. In March, AWS announced a partnership with Change Healthcare to offer DSaaS for health analytics.
That same month, mobility solutions vendor Comviva announced its own DSaaS offering to help telecom providers with marketing. And in February, data services provider Calligo acquired data analytics vendor Decisive Data in order to expand its DSaaS offerings.
With data science as a service garnering more attention, it's important to enterprises to know whether they should invest in the service and what are the reasons for considering it.
What is data science as a service?
According to Anand Rao, partner and global AI leader at PwC, data science as a service is the outsourcing of data science activities to an external provider.
"The client provides the data and the DSaaS provides the insights from the data to the client," he said.
It's particularly useful for temporary work, for sudden or peak workloads, or for standardized tasks like running analysis on monthly or quarterly reports, he explained.
One reason you might not hear much about data science as a service these days is that the term can mean a lot of different things. It covers everything from analytics tools embedded into popular SaaS platforms such as Salesforce to specialized vendors offering prebuilt models for specific business apps that they can customize and manage for customers to standard consulting deployments.
But whatever the flavor it comes in, DSaaS has a lot of value to offer to enterprises, whether or not they have an in-house data science team.
The more different systems you have to tie together, the less likely you are to consume data science as a service.
Kjell CarlssonPrincipal analyst, Forrester Research
A company's own data science team may be able to build something more customized than these applications, he said.
"But it's really difficult to get those models into the hands of the end user," he said. "The end user is already using those SaaS business applications, and the model is already right there, embedded in their workflow."
In many cases, the benefits of having an embedded model that's already there and easy for users to get started with outweighs the potential benefits of building a bespoke model, according to Carlsson.
However, if the data resides in different systems, the scale of the integration challenge might require an in-house effort.
"The more different systems you have to tie together, the less likely you are to consume data science as a service," Carlsson said.
DSaaS as business intelligence platforms
General-purpose business intelligence suites delivered as a service increasingly have augmented BI and augmented analytics. Tableau Online, for example, offers self-service analytics in the cloud. Microsoft offers Power BI as a service and IBM also offers its analytics tools as a service.
Dave Costenaro, chief data officer at AI-powered help desk company Capacity, uses DSaaS options such as Tableau for analytics and AWS for data storage.
"Infrastructure for data collection and storage can be easily outsourced to a number of cloud database vendors," he said.
Companies can outsource individual steps of the analytics pipeline to external providers, he said, even if they have in-house teams for other parts.
DSaaS as AI development platforms
Vendors such as C3.ai offer components and prebuilt AI modules that enterprises can stick together to build their own predictive applications, Carlsson said.
Using an external data science platform can make sense, even when a company has an internal data science team, because it allows flexibility to scale models up and down as needed and to spin up test environments quickly. It can also reduce capital expenses or licensing costs, and the provider is responsible for keeping the infrastructure maintained and updated.
Using a DSaaS platform can also provide access to the vendor's proprietary data science algorithms, said Hugh Burgin, U.S. data and AI leader of the Microsoft Services Group at EY.
There are some disadvantages, however.
"In a hosted model provided by a vendor, this can often feel like a black box, where the company has less impact and visibility into how the data operations and data science works," he said. "For some companies, executive buy-in on using external resources could be a challenge as well."
DSaaS as AI services or machine learning services
Major cloud vendors and AI-focused startups offer pretrained models for vision, search, recommendations, speech recognition, natural language processing and other common data science tasks.
It can be much faster and easier to use those offerings instead of building a system in-house. However, there could be compliance, privacy or security issues that can limit the use of outside providers, Carlsson said.
"The infosec team might give you a hard time if the data has to leave your walls," he said.
DSaaS as AI accelerators
Some DSaaS companies have prebuilt models they will tweak based on a particular company's data, then charge on a monthly basis to keep the model current with the latest data, Carlsson said.
"Certain use cases are so common for a lot of different companies that it's not going to be a differentiator for you," he said.
There's a trend of businesses moving to using these productized, digital offerings to support the data science lifecycle, said Doug Henschen, vice president and principal analyst at Constellation Research.
"On the development side, we're seeing data services, for example, that support model training with organic and synthetic data," he said. "On the operational side, we're seeing model monitoring services designed to optimize model selection and maintenance."
Common use cases involve fraud detection in banking, staffing management in healthcare, curt management in the telecommunications industry, and pricing optimization and customer targeting in the ecommerce sector.
He's seen companies using internal data science teams to experiment and identify opportunities, then use a DSaaS provider to take those projects to scale.
"I've seen plenty of long-term arrangements where organizations continue to rely on service providers, rather than attempt to develop data science expertise internally," Henschen said. "If you're going to invest in building and maintaining data science expertise, it should be something that clearly delivers differentiating value to the company."
DSaaS as consulting
Professional services companies are also happy to build AI models from scratch. Even companies that have internal data science teams may want to bring in a consulting firm to do specialized projects.
"Even if you have a data scientist on staff, a data scientist who knows how to build vision models is rare," Carlsson said. "Someone who knows how to do a speech recognition model is extremely rare."
Plus, building the model also requires labeled training data, which an enterprise might not have or would only be able to get with a great deal of effort.
A common situation is a company needing intelligent text extraction in order to deal with workflows that involve a lot of scanned documents. Ingesting those documents, pulling the data into a format the company can use and doing text analytics is not a set of capabilities most companies are likely to have in-house.
Enterprises are also likely to use consultants for one-off projects.
"Data science as a service can provide a quick hit to solve a quick business problem," said Chandana Gopal, research director of future of intelligence at IDC. "You're able to deliver business results in the fastest manner when you don't have the skills in-house."
Outsourcing can also make sense if a company doesn't know whether the project will be a success or not. There's no point in hiring a team of data scientists only to find out the data doesn't give you the result you're looking for.
"Then you might hire an in-house team once you know that there's a proven benefit to using data scientists," Gopal said.
"And the shortage of skills is so profound and big that organizations might wind up using service providers long term because they can't hire people," Gopal added.
The size of the company also makes a big difference, said Dan Simion, vice president of AI and analytics at Capgemini.
"If you are a small company, data science as a service may be a better fit than it would be for a large-scale enterprise that already has its internal data science team up and running," he said.
One such company is DayaMed, a Nevada-based startup building a mobile health app. In March, the company announced a pilot project with U.S. Veterans Affairs focusing on medication management and adherence, said company CEO Justin Daya.
To get that data science and AI -- as well as the integration into a scalable, mobile application -- DayaMed turned to SenecaGlobal for data science as a service.
"We now have plans to expand nationally to multiple health systems, pharmacies, [accountable care organizations], health plans and other clients," Daya said.
