International Institute for Analytics

Having just come out of stealth mode in July, University of Michigan Professor, former Ford control algorithm innovator and now startup co-founder and CEO of Refraction, Matthew Johnson-Roberson has jumped into the autonomous vehicle race. But he has a particular angle that has yet to be addressed – delivery robots that can operate in rough weather, such as Michigan winters. Matthew will talk about Rev-1, his new delivery robot, and the decisions that went into defining a market opportunity that leveraged the analytics potential of what could be developed.

Artificial intelligence has become the hottest commodity in recent years, and business, academia, and government have embraced it to propel complex use cases. As AI becomes more woven into the fabric of organizations (and its criticality increases), enterprise infrastructure becomes essential. AI is only as strong as its weakest link. The ability to build out use cases, deploy into production, scale, and secure all relies on the supporting solutions and infrastructure. There are many different decisions to make when choosing the right solutions and infrastructure: On-premises or off? GPUs or CPUs? Which storage system and framework to use? The list goes on. Drawing on real-world considerations, use cases, and solutions, Nick Curcuru discusses different decisions—and the associated considerations and best practices—you need to exercise to build and deploy a successful AI.

The vast majority of people building analytics and data science processes have every intention of being good and ethical. As a result, most potentially unethical and evil processes arise in situations where that wasn’t the intention. The problem is typically that proper focus and governance is not in place to keep analytics and data science processes on the side of good. On top of that, what is good and what is evil isn’t nearly as clear cut as we’d wish it to be.

Information Economies in Organizations

The data warehouse revolution began in 1991 when Bill Inmon published Building the Data Warehouse. Inmon observed, early in that book, that every organization has a naturally occurring information economy, and that most naturally occurring information economies were inefficient, duplicative and prone to produce suboptimal decisions.

This observation of Inmon’s has not gotten anywhere near the credit, or attention, it deserves. A decade’s worth of collective practice in advanced analytics should tell us that everything we know about real-world economies applies to our information economies. There is demand for information by people and functions in an organization, and there is a supply of (some of) that information. There is (some amount) of technical and procedural infrastructure – some kind of market — to bring demand and supply together in an organized way. That “market” infrastructure is often partial, fragile and in some cases ineffective. There are competitive alternatives (like cloud service providers and SaaS vendors), over- and under-regulation (various data governance models), excessive demand-side taxation (cost allocation strategies), failure to invest in infrastructure, and all other elements of economies.

When organizations are planning strategy-driven large-scale advanced analytics programs, they should begin their planning by characterizing their as-is information economy.

We’ve had technical people focused on the ingestion and management of data for decades. But, only recently has data engineering become a critical, widespread role. Why is that? This post will outline a somewhat contrarian view as to why data engineering has become a critical function and how we might expect the role to evolve over time.

Why do some analytics projects succeed while so many fail? According to Gartner analyst Nick Heudecker, as many as 85% of big data projects fail. However, the ROI from the other 15% that succeed is incredibly promising. With such a clearly high barrier to competency in executing big data strategies, there remains significant opportunity for first-mover advantage for enterprises that can crack the code to improving their outcomes.

So, what can organizations do to increase their chances of big data success? Part of the answer lies in creating a data engineering culture. This is the necessary foundation underpinning a big data analytics proficiency and enables companies to outperform the competition.

Multi-model databases (MMDBMS) have been expanding the definition of database for several years. A multi-model database combines several data stores in one database. Those data storage services support distinct data models. Data models include relational, graph, documents, key-value, time-series, and object stores. But simply storing different kinds of data is insufficient to call it multi-model. Specialized programming services must exist for each data model. In the best MMDBMS, a single query can combine data from all data models.

GE aspires to be an algorithmic business, but recognizes this transition will not occur overnight. It will occur in stages as the company develops new capabilities and implements multiple emerging technologies. This transition requires building solid foundational systems and encouraging broad experimentation and innovation using new analytics technologies.

Beyond getting experience with next-generation technologies, transitioning to an algorithmic business requires cultivating an enterprise-wide data culture and changing how people work throughout the company, particularly on the front line.

It’s No Game to Find and Keep Your Data Scientists - EA Battles The Market Forces for Talent

In 2013/14, EA’s voluntary turnover among data scientists was 21-22%. It is now 8%, with consistent improvements. These improvements occurred without major changes in compensation and without disproportionate change in investment in the analytics platform, which are common data scientist complaints.

Employment and Training in The Era of AI

As AI replaces some jobs and changes others, it raises questions of, “What is the role for humans in the AI world?” It is most useful to see humans and AI working together, taking advantage of the strengths of each.

The training and learning tracks will vary by role. Foundational learning will be required in all technical roles including basic software engineering, data wrangling, predictive analytics, and data visualization. Data scientists will require additional training in advanced machine learning; data engineers will require more immersion in distributed computing.

The demand for data scientists and analysts is estimated at 140,000 to 190,000. But the demand for data-savvy managers is even greater at 1.5 million. It is unlikely universities will be able to meet this demand. Universities tend to be more theoretical and less focused on practical application. Private training will be needed to fill the gaps.