With machine learning, more business processes will be automated

February 18, 2022

Source: AI Trends Team | AI Trends

Machine learning has the potential to automate many more business processes than are currently automated in enterprise software, based on all previous generations of software development methods.

That’s a suggestion put forward by Claus Jepsen, chief technology officer at Unit4, an ERP software provider based in Denmark.

Claus Jepsen, CTO, Unit4 Denmark

“In my experience, typically less than 20% of business processes are automated in enterprise software. I think in as little as two to three years, we could see up to 80% of routine business processes automated by ML,” Jepsen said in a recent review in Forbes.

Much of machine learning, which he describes as the ability to create automation through AI algorithms, is statistical analysis from calculations, identifying patterns and predicting future outcomes based on past results. All of this can be done with standard logic programming.

The extent to which ML can improve business results is “currently marginal”, he suggests, with the accuracy of financial forecasts, for example, sensitive to many factors more important than the algorithm’s ability to fine-tune itself. over time. If you don’t have harmonized, accurate, and complete data to start with, just applying ML to it won’t by itself lead to better business decisions,” Jepsen said.

Defining the business problem is the same challenge that software developers have always faced. “In terms of Gartner hype cycle, ML is currently at the top of inflated expectations,” he said. “You can’t just throw ML on a bucket of big data and expect it to magically come up with a perfect business plan.”

Points in a business process where judgment or prediction is needed, and where a small improvement in accuracy would have a strong benefit to the business, are candidates for ML automation. The humans surrounding the effort to make the AI ​​work are essential. They need to decide on the use case and ensure that the data is of sufficient quality to be useful, before giving the algorithm a task and then training it.

“The human mind is by far the best pattern-matching machine in the universe,” Jepsen said. “The average two-year-old can probably correctly identify a cat after seeing two or three, whereas an ML algorithm might need to see 2,000 to be sure. But, once trained, ML excels at dealing with huge volumes of data and processing it very quickly, never getting bored doing repetitive and tedious tasks day after day.

Machine learning is gaining momentum in Africa

This idea of ​​machine learning extending automation beyond what software development has achieved so far extends to Africa, where machine learning is making headway. IDC analysts have forecast AI spending in the Middle East and Africa to maintain its strong growth trajectory as companies continue to invest in projects using AI software and platforms, according to a report in Smart CIO Africa.

An IDC survey of IT leaders found that ML improved customer and employee experiences and led to faster rates of innovation in the organization.

Fady Richmany, Senior Director and General Manager, UAE Dell Technologies

The same challenges apply: pick a good candidate business problem to automate with ML and ensure the data is available to make it work. As part of this, “It is critical to identify and understand whether the problems they are trying to solve could be more efficiently and accurately addressed by machine learning rather than conventional software,” said Fady Richmany, Principal and General Manager, UAE Dell Technologies.

Speaking of candidate applications for ML, Richmany said, “Companies can use machine learning for customer retention because ML systems can study customer behavior and identify potential customer retention steps. they can use ML to aid in market research and customer segmentation, enabling them to deliver the right products and services at the right time, while gaining valuable insights into the buying habits of specific groups of customers so they can to better target their needs.

Considerations for Buying or Building an ML Platform

Companies that commit to pursuing machine learning for AI software development must decide whether to purchase or build the necessary ML platform.

“Building a solution takes years and people,” says Charna Parkey, data science manager at Kaskada of Seattle, in a recent review in integrated. Kaskada is building a machine learning platform aimed at enabling collaboration on feature engineering and repeatable success in production.

Airbnb, for example, took three months to decide what to integrate with its ML platform and four years to create it; they call it Bighead. Its developers used a range of open-source technologies, striving to “fix the gaps on the way to production” with their own services and user interface. This meant they had to support multiple frameworks, feature management, and model and data transformation. In a similar experience, Uber has been working on its platform, dubbed Michaelangelo, for five years. And Netflix started over four years ago on its platform, which continues to be developed, according to Kaskada.

Finding the necessary talent is always a challenge. The basic decision is whether to hire a classically trained data scientist or hire a domain expert and upgrade their skills. “I chose to perfect myself,” Kaskada said, and she’s not alone. Some 46% of organizations surveyed by PwC in 2020 said they were rolling out AI skills upgrading to manage the shift to more AI, and 38% were implementing accreditation programs.

Purchasing a pre-built ML platform saves initial build costs, integration costs of “custom and fragile workflows”, and it comes with dedicated external support, a she declared. It also reduces the time it takes to onboard new employees to proprietary software. The costs of moving to a pre-built platform, including the need to adopt new workflows instead of building the ones the company has in place, and perhaps telling developers that their tools favorites are out of fashion.

“Not all platforms will support all of your ML operations or the unique needs of your business,” Kaskada suggested. “Evaluate carefully.”

New book: Real-World AI: A Practical Guide to Responsible Machine Learning

In the real world of applied ML applications, the challenges are only just beginning to be understood, suggest the authors of a new book, Real-World AI: A Practical Guide to Responsible Machine Learning, by Alyssa Simpson Rochwerger and Wilson Pang, two experienced practitioners of applied machine learning. Rochwerger is a former product manager at IBM Watson, and Pang is the CTO of Appen, a company focused on improving data quality for ML applications, based in Chatswood, Australia.

“Only 20% of pilot-stage AI in large enterprises makes it to production, and many fail to serve their customers as well as they could,” Rochwerger and Pang write in real world AI, according to a book account recently published in TechTalks. “In some cases, it’s because they’re trying to solve the wrong problem. In others, it’s because they don’t account for all the variables – or latent biases – that are critical to a model’s success or failure.

The real world collides with the academic roots of AI in data.

“When building AI in the real world, the data used to train the model is far more important than the model itself,” Rochwerger and Pang write in real world AI. “This is a reversal of the typical paradigm represented by academia, where data science PhDs devote most of their focus and effort to creating new models. But the data used to train Models in academia are only intended to prove the functionality of the model, not to solve real problems.In the real world, accurate, high-quality data that can be used to train a working model is incredibly difficult to collect . »

Read articles and source information in Forbes, in Intelligent DSI Africa, in integrated and in TechTalks.

Comments are closed.