Most data managers are turning to partners for help expanding their machine learning efforts beyond the experimental stage.
That's one takeaway from a Capital One machine learning study of 150 data management decision-makers. Two-thirds of the survey's respondents said they're partnering with third parties to evolve their machine learning (ML) strategies and fill staffing gaps. Forrester conducted the poll on the financial services firm's behalf.
While the survey doesn't specifically mention IT services partners, consultants and MSPs routinely target AI/ML and have launched practices for it. Service provider executives cited AI/ML and data analytics as among their top technology trends for 2022.
Partners with hands-on ML experience can help guide customer adoption of the technology, according to industry executives.
Dave KangSenior vice president and head of data insights at Capital One
"Partnership takes many forms," said Dave Kang, senior vice president and head of data insights at Capital One. "The between-the-lines message I take away from the study is that whatever the route -- be it investment, consulting, co-innovation -- the risk that ML becomes a mere shiny object is there."
Decision-makers must remain focused on "demonstrating ROI and real business impact, in no small part by leveraging partners with firsthand experience," Kang said.
Partner experience in ML critical to success
Forrester's survey suggests partners potentially have a large and expanding opportunity in machine learning, although the bulk of businesses are just getting started.
Just over half of the respondents said their organizations have been developing and releasing ML applications for one to two years. The report noted that most of those organizations are experimenting with the technology, adding that a mature strategy doesn't emerge until about three years or more. In addition, 21% of respondents said their companies were six to 11 months into their ML journeys, while 5% were fewer than six months along.
The report recommended organizations seek out experienced ML partners to help move them out of the experimentation phase. That can prove a difficult task, however.
"Anyone can implement and run an algorithm," said Ryan Ries, practice lead of data science and engineering at Mission Cloud Services, a cloud managed services provider with headquarters in Los Angeles. "It's all about the interpretation of what the results mean, "as well as thinking through and properly constructing the data set, he noted. "You have to have a lot of training and understanding on how to look at the data, whether your data makes sense."
When a contractor lacks that understanding, projects fail. As a result, organizations are "starting to look at partners that have more proven track record results," Ries said.
Mission launched a data, analytics and machine learning consulting practice in February 2021. Ries joined Mission in January of that year, having previously rolled out an AI/ML practice at Onica, a cloud consulting firm.
Leidos, a technology, engineering, and science solutions and services provider based in Reston, Va., has also dedicated consulting resources to the AI/ML field.
"That's definitely an area that we do a lot of exploratory pilots, sandboxes and build-outs for our customers," said David Chou, director of cloud capabilities at Leidos. "We can use our different models from our AI/ML accelerator that are applicable specifically for our federal space. We're able to turn those models into practical deployments."
Two big machine learning challenges
Finding the right partner to tackle an ML project isn't the only obstacle customers face.
Data silos. Fifty-seven percent of the Forrester survey respondents said silos between data scientists and practitioners inhibit ML deployments, Kang said. Capital One has been addressing this issue.
"We have prioritized getting teams on the same stack and focused on collaboration, bringing down silos, and prioritizing reusable components and frameworks across all ML efforts," he noted.
Ries said the silo problem stems from data fiefdoms in organizations that have resisted the cloud's role as a general data store. "Companies have just had knockout brawls amongst themselves over who owns the data products," he said. "The hope of the cloud was, 'Hey, let me get everything into a central repository, then you can have advanced analytics and data science.' But no one inside the company wanted to allow that to happen."
One response at companies with groups that insist on data control has been enterprise discussions around data fabrics and data meshes, Ries noted. The fabric-and-mesh approach lets corporate groups, such as finance, own their own data but create data products that data scientists can subscribe to and use.
ROI is elusive. Lack of clear results also slows the growth of ML, according to the study. Nearly half of the data management decision-makers polled said executives struggle to see improvements from AI/ML adoption.
"People don't realize how hard it is to actually do some of the machine learning bits," Ries said. "We talk about this to some customers on the ROI side. It's not unheard of to take six to eight months to build a really good model. And is that model going to improve your business [and] get enough ROI back to pay for that effort that you put into it?"
Partners will need to help their clients produce results to keep ML moving. Benefits such as efficiency, productivity and improved customer experience are among the goals, Kang noted.
When decision-makers prove observable outcomes and gain executive leadership buy-in, their organizations can "further scale and operationalize ML applications," he said.