Creating a strategy to put wind in your data sails
Data is powering our economy. Businesses are using data to make better decisions, refine operations and create new revenue streams. Or are they?
Companies are certainly amassing vast amounts of data every day. Whilst the number (and variety) of data professionals on their payrolls is growing each year. Yet, data technology isn’t for free — as skyrocketing revenues from the large tech companies attest.
“Hope is not a strategy.” — Benjamin Ola Akande
While many companies are investing fortunes into data, data professionals and data technology, Gartner and others are consistently reporting devastating failure rates of data science projects.
Apparently, throwing money and talent at data — potentially a very valuable business asset — is not enough. The fatal 80% and upwards failure rate can eliminate even the smartest investments. And it is making (not only) Bill Schmarzo mad as hell!
So, what is the problem?
Simply, a lack of data strategy.
Yogi Berra paraphrases this data disconnect perfectly with: “If you don’t know where you are going, you’ll end up some place else.”
What is a data strategy?
Data strategy sounds like a boring corporate document. But in principle, it follows a very simple approach to maximising the benefits of data. I like to think there are three parts to a data strategy:
1. What business problems do you want to solve?
2. What do you need to do that?
3. How are you going to get it done?
At a high level, it is difficult to argue with such a generic problem solving approach, but let me expand it a little.
1. What business problems do you want to solve?
Every data strategy needs to be targeted. There needs to be a clear vision of how is data going to support the business, plus clear goals to be achieved in (say) the next two to three years. Data strategy needs to be aligned with business strategy. It must regard the business and operational needs, as well as ethical principles of the company.
The business goals might be minimising downside risks, supporting business objectives or a combination of both. Your company might want to increase revenue, profitability or customer satisfaction, or decrease costs. Alternatively, it might want to detect and limit fraud or better govern financial or supplier data.
The more specific the goal is, the better. For example, increase revenue is a good goal. Increase average ticket size is better. Increase average basket size even more so. The reason is that it helps narrow the focus of data teams and reduce room for misunderstanding.
2. What do you need to do that?
It is no secret that to deliver the value of data you need:
· technology,
· people,
· analytics and, naturally,
· data.
Consider all these dimensions when thinking about what is needed to achieve the solution to the problems outlined in point 1. I find it helpful to quickly think ‘blue sky’ before diving into today’s realities.
Think about what data you need to have available to solve the problems you have defined. Where can you get them, what’s their format, what’s the quality? How quickly can you acquire them, for how much? Time and time again it has been proven that having relevant data in a reasonable quality is superior to large volumes of some data.
Next, think about the technology required to build and deliver the solution into production. The end game — having the solution actually deployed and solving the business problem — is absolutely critical. The latest developments in the field of MLOps are showing that building a model solving a business problem is just the beginning. Deploying it into production, monitoring it, ensuring data, model and code quality is at least equally important.
Design the right architecture of data infrastructure enabling you to collect, process, analyse and deliver the data. Nobody knows what the future might look like so try not to make any architectural or vendor lock-ins.
Consider what analytics you might need to utilise as part of your solution. This, too, often dictates requirements on technology, data and people.
Next, think about what roles and personalities you need to have in the data teams, where should the data teams sit in the organisation and how should they operate.
Lastly, don’t forget about the rest of the organisation and possibly other people, like customers or suppliers, who might be involved with or be affected by your data strategy. Consider what cooperation you need from them, how you want them to behave, what are the responsibilities, roles and governance rules?
3. How are you going to get it done?
Once you know what you are solving, where you are and where you want to be, it is time to break it down into actionable steps. This is, by no means, easy.
One needs to create a roadmap with a sequence of projects, quickly delivering value, opening future opportunities and ultimately solving the original problem (as much as possible).
I tend to run multiple streams in parallel, with individual streams focusing on:
· business and analytical problems;
· data infrastructure; and
· people — both in the data teams and within the broader organisation.
An example of the first bullet might be version one of a recommendation engine or demand prediction model. The second could be represented by implementing a technology enabling the deployment of said recommendation engine or demand prediction model. The third requires hiring, training and developing the data teams, as well as building bridges with the organisation — understanding the business problems as well as explaining what data and analytics can and cannot do.
It is also advisable to, wherever possible, throw the net a bit wider. Collect data you need for the problems at hand and a bit more, put together a technology that solves the current problem and provide flexibility for future. That way, while solving the problem, you are also preparing the business for (inevitable) future complications. Analytics also gives firms an edge in learning and adapting.
How to create a data strategy
Creation of a data strategy is a collaborative process. It cannot be done in a vacuum. It requires a close collaboration between chief data officer, C-level executives and other relevant people.
It takes a couple of weeks to understand the current situation (or ‘as is’), discuss the business goals and outline the solution and how to get to it. Once the data strategy is formalised, it needs to be approved together with timelines, milestones and budgets.
What not to forget — execution of the data strategy
Now it gets interesting. Having the data strategy is great, but the proof of the pudding is in the eating — executing the strategy successfully is what matters.
Let me share three points related to the execution of the data strategy that are important to be covered in the data strategy itself.
· Data strategy principles
· Specifics of project management
· Metrics and measures of success
Including meaningful principles in the data strategy has two benefits. Firstly, it provides guidance to the data team when facing operational decisions without the need to align on every detail again and again. And secondly, the process of creating the principles provides the initial alignment between business and data teams.
Managing data science projects is notoriously demanding, largely because of the science component (requiring creativity, and often with uncertain timelines). Acknowledging this in the data strategy and outlining the ways to minimise delays and miscommunications is a great first step.
Metrics and measures of success need to be agreed, designed and tracked from the beginning. With over 80% failure rate there is simply no other way. The chief data officer as well as CEO and other executives need to know if the execution of the data strategy is on track, what has been achieved and what is not working. And remember, data strategy is not set in stone — the progress and latest developments should be regularly discussed with the top management, as well as with the data teams.
Please refer to the article below for further tips for execution of the data strategy.
Conclusion
Having a data strategy fundamentally makes a difference in delivering on the expectations of data. Pragmatically thinking about what business problems you want to solve; what do you need to do that; and how are you going to get it done is what it takes. Make sure your data strategy covers all data-related levers: business, data infrastructure, data and analytics, as well as people.
Plus, the specifics of data project management should be acknowledged and addressed in the data strategy too — make it a living document and regularly track progress with both data teams and senior executives to genuinely deliver on the promise of data in your organisation. And, given that we are at the start of a new calendar year, what better time to kick of the strategy drafting process or pull out last year’s strategy for a refresh. Best of luck!