Three Popular Ways to Fail with Big Data - How to Overcome Common Pitfalls on the Way to Creating Business Value with Data

Driven by the success of companies like Google and AirBnB, many enterprises have high aspirations for creating business value with data.

However, few achieve those goals. A Gartner study from Q4 2016 states that only 15% of enterprises have big data systems in production, while 30% are piloting and experimenting. The remainder are gathering knowledge or developing their data strategy. This suggests that fewer than 15% are truly successful. In this post, we examine the root causes behind this relative lack of success.

All too often, data is seen as a technology-driven effort without a clear line of sight to business outcomes. The same Gartner study shows that data efforts are initiated and led by the IT function about twice as frequently as they are led by the business. Only in 1 of 10 companies who participated in the survey are these efforts led by the chief data officer (CDO).

Silicon Valley Data Science was recently approached by a CIO who had spent more than $10m on hardware, software and services related to a data platform, without as yet having a single use case in production that would show value back to the business. We are sure those use cases existed, due to our work with other clients in the same industry.

Wherein lies the problem?

A technologist might point to gaps or friction in the data value chain that need to be mastered: discovering, ingesting, processing, persisting, integrating, analyzing data and exposing the resulting insights to the business.

A more holistic view recognizes that a range of business and technology capabilities need to be mastered to achieve full success with data. Relevant business disciplines include:

• business and data strategy
• organization and skills
• service portfolio and program management
• analytic application development for the business

Technology disciplines include:

• architecture and governance
• infrastructure and operations
• data management and security
• privacy and compliance
• data science and machine learning applied to business problems

All these disciplines, and a wealth of discrete capabilities in each, need to be orchestrated in pursuit of the “global optimum” state for the enterprise. That is precisely why enterprises are creating the CDO role: to institutionalize new thinking and data best practices. 

The CDO role can be compared to that of the central nervous system in the human body. It is a signal processing function, staffed by a small team of specialists (perhaps 30-50) that enable better decisions across the whole body of the organization (perhaps reaching 30,000-50,000 colleagues). Only a small number of enterprises have a CDO function that is fully staffed up, as the acute talent shortage in the data space continues. 

Three failure patterns

We have observed three main failure patterns on the way to creating business value with data.

1. Failure to motivate and mobilize the organization 

Becoming truly data-driven is a multi-year process even for well run organizations with excellent change leadership skills. All too often, boards have inflated expectations, driven by data success stories in the media. The hard work of building foundational capabilities, experimenting and learning from failure, and establishing a data platform as a shared service for the enterprise, never makes any headlines. 

Successful data leaders (whether they hold the CDO title or not) are those who can channel excitement by the board into a concrete roadmap that balances tactical quick wins and lighthouse showcases with sustained investment into technology platforms and talent development. This is a balancing act of expectations management, between delivering the art of what is possible today (incremental capability for the business every quarter), and recognizing the magnitude of the required changes to the way business is done (akin to an industrial revolution).

Organizations whose culture punishes failure will not succeed, nor will leaders who lack persistence.  Some organizations hire energetic young data leaders from the Googles and AirBnBs. All too frequently, these up-and-comers don’t know how to get things done in a large enterprise, get frustrated and leave the organization within a year. 

2. Last mile problem

Ultimately, the largest value tends to come from a closed loop of insight-to-action. While data leaders often control the middle of the value chain (e.g., oversight of data engineering, data science, analytics), and ultimately succeed on-boarding most commercially relevant data sources (not without pain), they almost never control the end of the value chain (at least in a traditional, largely off-line business). The large investments in data platforms and analytics may result in better management decision making. However, that only translates to better front-line decision making on a day-to-day basis if the ongoing stream of recommendations and insights is processed all the way through to the point of sale, the call center, the manufacturing shop floor etc. That requires application integration, application development, training, worker involvement etc. 

You need strong top-down sponsorship to close the last mile problem. One of the few exceptions is for CDOs who also are in charge of (some) sales channels and customer care and loyalty etc.—they have some end-to-end purview, but most analytics leaders do not.

3. Choking change with excessive bureaucracy 

Making change happen also requires money that is well spent on the right mix of talent and technology. Historically, enterprises have instituted a very defensive set of policies and procedures to govern money, talent, and technology, notably for mature parts of the business. 

Money: Budgeting processes (and associated stakeholder expectations) are based on deterministic outcomes with respect to the value of data. A return on investment needs to be quantified, even though data analysis starts with discovery and exploration. ROI hurdles are particularly ill suited for foundational platform investments. Some of the information requests from finance functions (in conjunction with preparing an investment board case for data initiatives) are as nonsensical as trying to quantify the ROI of introducing accounting software in the 1980s and 1990s. Both capital and data are key production factors that need to be run professionally. Furthermore, detailed waterfall plans are requested, even though the organization is in learning and experimentation mode. 

Talent: Most roles in an enterprise are scoped so that they can be done by the cheapest, least qualified resource that can still get the job done. They are also compensated such that more qualified or ambitious resources look for better paid opportunities elsewhere. This is completely counter-productive in the data domain, which requires premium talent at its current stage of evolution. Consider the role of a CDO, who needs to combine a deep understanding of both business and technology, coupled with stellar communication and change leadership skills. Consider the role of a data scientist, who needs to be highly mathematical yet creative. For any individual enterprise, a necessary step in overcoming the data talent shortage in the market is to escape the HR determined boundaries of what talent can earn at a given grade.

Many enterprises have tended to rely mostly on large SIs and traditionally never built out much capability in-house. Others find that the talent market is so hot that top talent prefers to remain in contractor mode. So 80% of their data engineers and data scientists have no skin in the game since they are not employees. If you do hire an external firm who brings experienced talent, pick one who is committed to make you self sufficient, helps you build out competencies, works in joint teams, assists in defining roles & responsibilities, provides quality assurance in the interview and new hire on-boarding process etc.

Technology: There are still many organizations even today who struggle to bring innovation into a world of IT that is neatly compartmentalized into various IT domains and often relies overwhelmingly on third party IT service providers. This might work in well understood, commoditized domains, but - at present - not in the data domain, which calls for fostering a culture of experimentation. 

As a result, the data change agents spend too much time planning and managing the organizational 'overhead' that gets in the way of delivering value incrementally, early and often. 

The offense play book

Big data requires enterprises to go on the offense, which is played differently than a defensive game.  Core to the offense play book is the ability to iterate quickly towards ever greater business value by embracing Silicon Valley best practices, notably with respect to building minimum viable products / platforms through agile development. This includes principles such as:

• Building a data science capability, in order to conduct experiments and learn how to respond to the changing environment
• Cloud-first, to make technology building blocks (notably foundational infrastructure) readily accessible
• Open-source-first, to experiment with and prove the value of software without a prior procurement and licensing discussion
• Dev-Ops, to move successful experiments into production rapidly and iteratively
• Data platforms and APIs, to make it easy for the organization to consume data and insights 

For a fuller discussion on how to overcome these three common failure patterns, join our panel discussion this Thursday at Dataworks Summit / Hadoop Summit.

Disclaimer:

This post does not cover some of the common technical challenges. In our experience, these are less intractable for most organizations. Common technical challenges include data quality, data context (data dictionaries), legacy / shadow systems, network infrastructure, operating models / service transition, approvals for security / privacy / compliance.