The excitement about data needs to be balanced with an uncomfortable truth. Most data projects fail. Perhaps as many as 85%. This doesn’t mean your data modeling project or big data analysis tool is doomed. But it does mean you need to know how to swing the odds back in your favour. In this article we look at the reasons many data projects don’t deliver value. We also suggest practical steps you can follow to make your project a success.

Why are data analysis projects so hard to get right?

Spending on data analytics software is expected to hit $150Billion in 2017, a 12.4% increase over 2016. But that doesn’t mean that returns have kept pace. If Gartner’s estimates are accurate, and 60%-85% of data projects do indeed fail, billions of dollars each year are being wasted. And that’s without adding in other investments such as consultancy services, training and hardware.

What’s going wrong? It’s a complex issue. However a review of recent research, along with our ongoing work delivering data projects with public and private sector leaders, has highlighted a few common challenges.

Bad commissioning

Data projects stand little chance of success if they don’t have a clear question at their heart, a clear idea of how data will answer the question, and a clear idea about the ways the answer will be useful.

From poorly articulated questions, to projects that don’t have a clear connection to organisational goals, it’s the role of budget holders and leaders to get things started the right way, to provide clarity around the problems that need solving, and an understanding of how useful action might be taken.

The data skills gap

Studies suggest a growing imbalance between the demand for data skills and the availability of those skills in our working population. Without specialist data science skills such as analytics, engineering and programming, the design and delivery of many data projects becomes very hard.

But it’s not just about the well publicised lack of data scientists. Leaders who aren’t data literate will find it tricky to provide strategic guidance to projects and oversee their delivery. And it’s this lack of data literacy that’s an important factor behind the commissioning challenges common to many projects.

When data analysis techniques and technology don’t line up with the problem.

Even exploratory projects need to make sure that they can access and use both the data and the data analysis technique and tools they need to conduct the analysis. It may sound simple but it’s essential for everyone involved in a project to understand how the data and the techniques to be used will combine to get useful results.

Not keeping pace with changing circumstances

Data project leaders need to remember the hard-won lessons from the software development community - to deliver something of value you need the flexibility to adapt and learn along the way, particularly as circumstances and data changes. Which is why data projects that don’t have the ability to iterate and evolve alongside an organisation’s evolving context, goals and needs are more prone to failure.

Ignoring existing patterns

Whilst data might seem like a new phenomenon, in reality many of the tools and techniques have been around for a while, particularly in the academic and research sectors. It’s tempting to believe that your situation is unique and that your solution will be unique too. But, that’s rarely the case, and whilst you will most likely need to tailor what you do, following established methods, using proven tools, is more likely to yield valid results.

Believing in magic bullets

Yes, existing patterns might point to a potential route to success, but projects that promise quick results, with little effort or input, using off-the-shelf solutions, tend to result in disappointment. Similarly the belief that hunting in data using magic tools will result in outstanding results is one that regularly results in failure.

What can you do to avoid data project failure?

Whether you’re starting your first project, or have run a few before, there are some best-practices that together build the foundations for success.

A few of these practices will seem like common sense. Many are unique to data. Everything is based on the current state of research, alongside our experience conducting projects as part of in-house data teams and as data science consultants to governments and companies across the world.

A summary of the key things to consider follows. If you want a more detailed, downloadable checklist for your records please click here.

Initial checkpoints

Before committing to planning a data project in detail you need to be sure that what you’re trying to do is suited to a data-informed approach, and that it can have an impact, in the right way. Specifically check that:

• The problem or question is important and connected to organisational goals
• That data will play a central role in answering this question
• That you and your organisation are ready to take action based on what the project will accomplish
• That you can access the data you need from internal or external sources
• That your team or your organisation have the capability, capacity and resources required to address the problem.
• That the project approach or data itself is free from ethical concerns

Getting your question right

The key to successful data projects will often rely on the clarity of the question at its heart. This means finding a question that’s focused on your goal and being clear how data can answer it. You need to consider:

• Is your question simple and clear? Does it contain a precise hypothesis?
• How will data help answer this question?
• Who is interested in this question and how will they expect to see it answered?
• How will results be delivered?
• How will answering this question allow action to be taken?
• How will you re-evaluate the question based on early data exploration?
• The impact of timings on your question, particularly in fast-changing datasets

Building a business case

To get the support and resources you need to deliver a project you’ll need a compelling case that connects the question you are addressing to an outcome people care about. In practice this means:

1. Being clear how your objectives align with organisational goals
2. Describing how action will be taken based on your findings or outputs
3. Getting buy-in by defining how your project supports objectives of key stakeholders
4. Weigh up alternative approaches, including the impact of doing nothing
5. Anticipating concerns and objections

Agile ways of working

As with many technology projects, adopting an Agile approach will enable you to get results quicker, allowing you to adapt and learn as you go. Think about:

• Defining a Minimum Viable Data Product (MVDP) - the approach that will get you closer to a result whilst helping you iterate both the question and the analysis approach
• Failing fast by using a MVDP as a way of identifying issues in your data and approach
• Establish how you will communicate with your delivery teams and sponsors
• Adopting Agile as your project management approach
• Using a diagram or visualisation to sketch out how the project will work
• What project tools you’ll use to plan and manage the work

Data Analysis

A key step in the project process is deciding how your analysis will be conducted. Consider the following:

1. Conduct a search for similar projects or papers that have addressed similar questions.
2. Do your data sources suggest this will be quantitative analysis, qualitative analysis or a blend?
3. Describe how this analysis is the correct way of addressing the question
4. How will the kind of analysis you need to undertake shape the resourcing and structure of the project?
5. What initial exploration of the data can be quickly done in order to refine both the question and the analysis approach?
6. How will you verify that the analysis has been conducted correctly?

Putting Data To Work

As data is the core resource for your project you need to plan how to use it. Answers these questions to understand your data needs, select tools and plan your resources.

• Do you have access to the data you need?
• What size and format will data come in?
• How often will core data come in and/or be updated?
• Are the data sources internal, external or open data?
• How will you access data?
• Is the data licensed openly or do you have the appropriate usage rights to use it?
• What are your data accuracy and cleanliness requirements?
• Have you budgeted for 70-80% of your time to be spent on finding, cleaning and preparing data, ready for work? This is the average time needed on data projects.

Scope Your Data and Skills Infrastructure

To deliver an effective project you will need the appropriate infrastructure to support it - from access to tools, to having the right people with the right skills available. Check for the following:

• That your organisation’s data governance and data management policies permit you to access, use and share data in the way you need it
• Internal systems and tools that can support the project
• Do your and/or your team have the skills and capacity required to undertake the project?
• Have you identified support you will need from internal or external sources?
• Does the project team contains the variety of skills you need? For example: data science skills, communication skills and domain or subject matter expertise.

Summary - data driven projects need both focus and flexibility

As we’ve seen in this article data projects can be tricky to get right. They need to be aimed squarely at a specific question or challenge, preferably one that can provide a useful basis for action. At the same time they need a degree of flexibility in order to adapt when things change. All whilst making best use of appropriate techniques, deployed by skilled people.

This may seem like a tricky balance. However our experience suggests that the more of the areas and questions above you can tackle the greater your chance of success.

If you’d like a detailed, downloadable checklist of the seven areas to cover, including an extra section covering “traps to avoid”, click here.