Three meta-steps for data project effectiveness

January 2023 Issue

Your data expertise, your time, and your mental energy are valuable. Your work can help your stakeholders accomplish great things, but that work needs to be focused and efficient. Devoting more time, people, and resources to data work is not the answer if that work is misguided or irrelevant to your stakeholder's goals. All of you deserve better.

In that spirit, I'm offering you three "meta-steps" – or steps of steps – for data project effectiveness. The meta-steps are simple:

1. Build a clear consensus on workable goals with your stakeholders
2. Develop a laser focus on actionable insights
3. Be mindful of ROI (return on investment) at every step

Let's go over each of these meta-steps briefly and see how they can make your data work more efficient and more effective.

Build a clear consensus on workable goals with your stakeholders

The first step in any data project, not surprisingly, is to decide what the goal is. That is obvious. But, as anybody who has worked on projects (of any kind) knows, just because something is obvious doesn't mean that it always happens. On a data project, it's important to spend enough time with your stakeholders – your client, your supervisor, the people who would be affected by the results of the project – to get a clear consensus on workable goals for the project. Those terms are important, so let's take a closer look at each one:

• Clear. Your plan needs to be explicit. Describe the finished product in sufficient detail that everyone knows what to expect. You might consider writing a short memo with a list of project deliverables (that is, what you will actually give to the stakeholders at the end of the project). A project memo can be helpful even for internal clients or, really, even for yourself. It's a specific guide.
• Consensus. Everybody needs to agree on what the project is designed to accomplish and what the end product will include. Getting on the same page is particularly important when the project includes terms that mean different things to different people, such as "optimal strategy," "best clients," or even "profits." For example, what is included (or excluded) when calculating profits? Does it include projects that cross over accounting timelines? Does it include potential returns? Does it include fluctuating currency exchange rates? Even a single term can have many meanings, so take the time to make sure you all agree on what the project should do.
• Workable. You should perform a reality check on your project. Is it possible to give the stakeholders what they ask for in the time allowed and with the resources allocated to the project? Is the necessary data available, and is it in a format that will fit your needs? Do you have the necessary expertise in-house or will you have to reach out for parts of the project? Are you juggling other projects that may have to wait while you work on this one? And are the anticipated outcomes in line with your stakeholders' business model? For example, a store may only conduct business online (or in person) and proposing a change in modality could be a huge adaptation. Or, for a company that provides professional services, licensing requirements may restrict involvement to certain countries or states.
• Goals. This gets back to the idea of knowing what you're trying to accomplish with the data project. While it's possible that your project is purely "exploratory," more often than not, that just means that people haven't taken the time to figure out what's important to them yet. Your work will have greater impact when your goals focus on "actionable insights," so that's what we'll look at next.

Develop a laser focus on actionable insights

This newsletter is called DataIsForDoing, but that's a curious phrase. It's adapted from a line by the American philosopher and psychologist William James. In his 1890 book The Principles of Psychology, James said "My thinking is first and last and always for the sake of my doing." The idea is that the brain doesn't just think random things for no particular reason. Rather, it's trying to solve strategic goals and that our mental processes can best be understood by looking at their functions. For data projects, the idea that connects to "actionable insights," or data-derived observations that lead to stakeholders taking specific actions to bring about the outcomes that are important to them. That is, data should lead to recommendations for behavior.

Unfortunately, many data analysts and data scientists see their responsibility as finding interesting patterns in data, but leave the actual recommendations to others. While there may be situations where this kind of division of labor works well, it contributes to a passive approach that can be more like turning a car on but not putting it into gear and going somewhere. That is, it misses the point of the work.

Fortunately, most "interesting patterns" can be turned into "actionable insights" by just taking one or two more steps, as in the following hypothetical examples:

Example: Website Engagement

• Interesting pattern. People who visit your website on mobile devices are more likely to engage than are people who visit on desktop computers.
• Actionable insights. Your stakeholder could (a) make engagement easier on desktop; (b) make calls-to-action on the desktop sites that encourage people to engage on mobile devices; or (c) double-down on offers targeted to the higher-engagement/mobile-user group.

Example: Museum Attendance

• Interesting pattern. Most people who visit a local museum on Sundays come for their outdoor performances, and not to see the art collection
• Actionable insights. The museum could (a) close the collection on Sundays, thus saving in operational costs without significantly impacting collection visitors; (b) offer incentives such as special pricing to view the collections on Sunday; or (c) bringing art displays – physical objects or virtual displays – out to the performance area.

Example: E-bike Usage

• Interesting patterns. E-bike sales boomed during the pandemic when people needed a way to travel locally that was healthy and affordable (compared to automobiles, which became scarce during the pandemic), but bicycling infrastructure is a challenge in many cities.
• Actionable insights. Cities could support e-bikes by (a) creating tax incentives for e-bike purchases that are similar to the EV incentives; (b) providing safer bike lanes and bicycle traffic signals to support all forms of bicycle travel; or (c) promote e-bike sharing programs at key places in their cities.

What should be clear in each of these examples is that it doesn't take much work to go from an "interesting pattern" to an "actionable insight," and that the actionable insights are the things that the stakeholders are generally hoping to get from the analysis in the first place. By meeting your stakeholders' needs directly, you can make sure your analytical work has greater impact.

Be mindful of ROI at every step

Your data work should focus on actionable insights, but not all actions are equally efficient at turning effort (or time or money or resources) into desirable outcomes. This gets to the idea of "return on investment" or ROI, which is central to the business world but often unfamiliar to data analysts and data scientists. And while it's customary to think of ROI in terms of literal financial investments – for example, a dollar spent retaining a customer has more expected earnings than a dollar spent finding a new customer – the same concept applies to every stage of a data project.

• Gathering data. All else being equal, more data is better than less data, but more data takes more time and costs more money, especially when it comes to proper cleaning and preparation of the data. At what point does gathering more data lead to diminishing returns for your analytical conclusions? What is the ROI for investing in additional data?
• Choosing analyses. Some analytical methods, like linear regression, are flexible, generally straight-forward to set up, and easy to interpret and apply. Other methods, like random forests or neural networks, may be better adapted to the intricacies of the data (especially in the case of non-linear relationships), but they can be much more sensitive to set up, much slower to execute, and much more difficult to interpret and apply. In this case, ROI refers to the value of the additional insights given the additional costs of the methods.
• Presenting results. Data analysts and data scientists are often justifiably proud of the detective work they do in order to get their data into shape, design their analyses, and find interesting patterns. Stakeholders, on the other hand, often just want to know what they should do. Your presentations should only add details about methods to the extent that they help the stakeholders take informed action. Here, the ROI calculation compares the level of detail you provide with the certainty of the actions that they will take.

In these non-financial examples, ROI is also similar to the so-called "80/20 rule" or Pareto principle, which is often summarized to say that 80% of returns, for example, come from 20% of actions. The economically wise choice in that case is to focus on the 20% of actions that produce the greatest returns. The same is true in data work: at each stage of your project, focus on the steps that will get you the furthest towards your final goals of actionable insights.

So what are your next steps?

This short collection of "meta-steps," is not a final answer or a finished guide to your analytical work. Instead, take it as a starting point for thinking about how you can make your own data work more effective, more impactful, and more efficient. Taking the time to reach clear consensus on workable goals, focusing on actionable insights, and choosing high-ROI solutions at each step will get you started on maximizing the impact that your valuable work can – and should – have for your stakeholders.

Thanks for joining me here. And remember, sharing is caring! Follow me on LinkedIn and share this newsletter with a friend who you think would benefit from it.