Connections

For organizations that are looking to make better business decisions using machine learning (ML) without writing code or requiring ML experience, Amazon SageMaker Canvas provides the solution they’re looking for, according to Danny Smith, senior manager for product management, AI Platforms, at Amazon Web Services (AWS).

Amazon SageMaker Canvas, after all, enables business analysts to achieve effective business outcomes with a visual point-and-click interface allowing them to understand their data better and generate accurate ML predictions without writing even a single line of code.

“If we look at the machine learning industry kind of advancements over the last 10-plus years or so … what you saw was we were focused a lot on building machine learning technology,” he said Aug. 16 during the “AWS Online Tech Talk” webinar “Make better business decisions with ML using Amazon SageMaker Canvas, without code.”

Since then, what we’ve seen over the past five or so years, is people have asked “how do we ease the adoption of that technology,” he told viewers.

“Really now, the expectation from our customers is ‘how do we take machine learning mainstream,” he added.

During the session, he explained how business analysts, supporting ML professionals and others can use Amazon SageMaker Canvas to quickly and easily build ML models, analyze models and generate accurate predictions with only a few clicks.

He also explained how they can use SageMaker Canvas to address business problems including sales forecasting and how they can share models and datasets with data scientists so they can validate and further refine ML models.

“In early days, it was data scientists and really research scientists building out this ML technology – things like deep learning and stuff,” Smith said. But “when we got into easy ML adoption, new characters came into play,” he noted, adding: “So we had data engineers who were focused on getting the data ready for machine learning. We had data scientists doing the model, building and stuff. And then we had ML engineers preparing the pipeline.”

But he told viewers: “If we want to get to mainstream ML machine learning adoption, we need to help business analysts.”

Why? The reasons can be chalked up to several things that AWS customers were saying, he said, including: “We want to unlock the value in our data.” “ML is a competitive advantage.” “How do we get more ML builders?” “We want to democratize ML.”

“So, if we go and look at where the machine learning value creation process is today, as you saw, we had a bunch of focus over the last 10 or 12 years from personas like a research scientist or a data engineer or data scientist or ML ops,” he pointed out.

“We tended to spend a lot of time talking to those people who are concerned with data preparation and feature engineering, model development and training, model deployment,” he said.

“But if we think about where the journey starts, buried somewhere in the line of business there is an analyst who is not a machine learning professional, who is not a coder, doesn’t know Python [and] all these things; they have a business problem and so they’re thinking about this business problem like, ‘hey, I’m a marketing analyst and I want to predict when my customers leave,’ he said.

As they think about this problem, “they have data that might be appropriate to help them get some new insights,” he pointed out. “But then they kind of get stuck” and that is because there is a “barrier to entry,” he told viewers, adding customers have said: “We get stuck with weeks and months of delay trying to get to the value and we only cover a really small part of the value” or a “small set of use cases.” Customers, therefore, asked AWS to help them, he said, noting the key was to scale ML value creation.

ML team productivity can be enhanced by leveraging the most complete, end-to-end ML service and increase productivity by 10x, he said. Another option is to expand the ML development team in proportion to the need, he said. And then you can also democratize ML innovation to enable more groups of people, including business analysts to build ML models, he added.

The main challenges analysts face in building ML now include: the lack of deep ML expertise and the learning curve is steep; business needs explainability and validation from experts; and available no-code ML tools tend to lack transparency and also have upfront fees, he said.

To view the entire webinar, click here.