Five Steps for Building a Great Data Product

If the necessary platforms don’t exist, create them.


PlanAllen Kelly and the #beyondprojects movement gave us a new mindset for our work: think continuous.

Instead of framing everything as a project with a fixed value and a fixed end date, we organize ourselves around continuous, product-like work streams.

They emerge, grow, shrink or die, according to the value they deliver, much like a “survival of the most value-generating”.

To work efficiently, we prioritize by value and effort.

In the beginning of data science initiatives, low effort is more important than high value, as quick wins will generate valuable trust.

However, good planning is about choosing a suitable approach, not about following a fixed route: The market dynamics of today’s digital economy make value “a moving target”, and the experimental nature of most data science efforts impedes a precise estimation of time frames.

This hinders long-term planning and necessitates an iterative approach.

To reduce risk, we move in small steps to create a minimum viable product (MVP) for which we obtain customer feedback to learn and improve.

In the Agile Data Science Manifesto, Russell Jurney introduces the idea of the data-value pyramid: layers of data usage with increasing value building up on each other, from data records to charts, reports, predictions and finally action.

Keep this in mind when planning your work: Data science is only effective when it leads to action, be it at strategy level or for micro-decisions.

Thus, think beyond models and analyses: Plan to integrate data science products into organizational processes and make sure your results “make it through” by leveraging your social skills.








However: Experimenting isn’t about trying random stuff, but finding out what works or doesn’t, i.


discovering the path to success by experience.

In the digital economy, a great idea is much more valuable than those few percent additional efficiency you get from narrow-mindedly pursuing your current task.

Take the time and broaden your horizon.

Combine different ideas.

Keep in touch with the latest developments and innovations.

Oftentimes you’ll get stuck, but something that doesn’t work now might come in handy at another time.

Keep an open mind, and many valuable insights will come to you along the way.

Discuss and share your approaches, ideas, experiences and results with your team and the wider data science community, both offline and online.


DeployNow, at the latest, the true data scientists shine: They’re capable of not only building great models that answer real business needs, but also creating products and services from them.

Either closely collaborate with software engineers or learn their craft, and you will witness your experimental workflow being stripped to its bare essentials and being reborn as a production-grade application.

Don’t keep interfaces or functionalities because you needed them in the past or might need it again in the distant future — it’s wiser to add that when the time comes than to cripple your product by complexity right from the start.

Don’t throw away your exploration notebooks, but archive them.

Learn how to use deployment technologies to bring that piece of code into action.

To move swiftly and release an MVP early on, work closely with other roles, such as product owners or sysadmins.

Devise autonomous and heterogeneous teams, encompassing all necessary roles, which are not obstructed by organizational boundaries.

A sense of joint responsibility is gold: If everybody cares not just for his part, but for the product as a whole, people will work goal-aligned and pragmatically close responsibility gaps.


GrowScale your product.

Look forward: Which improvements, features or changes will further increase its success?.Look backward: What lessons can you derive from the journey so far?.How will this help you with your next endeavour?Communicate your results and successes to claim “data science credibility” for your team and yourself.

Can you put a number on the savings or revenue increase you’re responsible for?Good work will create trust, which will reward you with further opportunities.

In the long term, you’ll go from changing processes to changing mindsets, growing beyond the scope of data science and unfolding a truly transformational effect.

Do you miss anything?.Disagree entirely?.Share your opinion in the comments!.. More details

Leave a Reply