If you don’t plan where you are going from the start, you can’t expect to be satisfied with your final destination. In the business world, as in many others, having a well-defined strategy, and working to that strategy, ensures you arrive where you hoped.
In the context of data science, strategy can be applied at a number of different levels. There are distinct contrasts between applying a strategy for the whole organization, for a whole team, and for a particular project.
This chapter takes a high-level view, working through the process of developing a data science team strategy that is aligned to the goals of the organization as a whole, creating a document that records and communicates the strategy, and then considers how to use the strategy to the best advantage of the organization by referring to it for decision making.
A lot of the emphasis in this chapter will be on people who are in data science teams in larger organizations. In this role, although you are not a CEO or divisional head who needs to put together the grand plan of the business, as a leader in your organization (and all data scientists should consider themselves thought leaders, even if you are not anyone’s manager), you still need to understand the top-level strategic plan, and have some sympathy for how it is produced.
However, if you are a data scientist in a consultancy, although you may skim some parts, you will still get a lot out of the overall theme of strategic alignment, so as to demonstrate the value of your work to your clients more effectively.
The Idea of Strategy
One of the key ways military strategy has been taught in the United States is according to the formula Strategy = ends + ways + means, with future military leaders using this summary to analyze past commanders’ campaigns.1
Using this formulation provides an ability to analyze how strategies are varied according to the objectives and the means available to achieve those objectives, although I will often blur the distinction between ways and means.
Arthur Lykke offers some examples of military objectives in his article discussing ways and means. They include defend a homeland, restore territory, and deter aggression. It is intuitive that, for example, deter aggression could be achieved by projecting an appearance of powerful forces without firing a live round, but restore territory will require incursion into an area currently held by the enemy.
In his original article, the metaphor that Lykke paints is one of a stool with three legs, where each leg needs to be same length and angle for the stool to remain balanced (and therefore resist being knocked over). While we will more often use a two-sided metaphor, the point remains that the elements need to be balanced to ensure success.
Data analysis is a field where pursuing details is especially likely to be attractive, from which it follows that not seeing the wood for the trees is especially risky for a data scientist. Taking care to explicitly consider the big picture is one possible antidote to this problem and is the first step to knowing how far to chase down the detailed aspects of a data set.
Although data science has been promoted heavily as an answer to many of the world’s problems, there are signs that at least some people have begun to feel that the promise hasn’t been lived up to. A very plausible reason for this gap is that the overall objectives for data scientists, both as teams and within projects, may not be clear enough. This can cause a situation where data scientists are delivering projects that are excellent on the metrics the data scientists themselves set, but which fail to deliver the organization’s goals.
This brings to mind another military adage, “War is too important to be left to generals,” which is loosely meant as a warning that generals can win wars and battles without gaining anything of value for their country if they don’t communicate well enough with their political leaders.
Winning Battles Without Winning the War
Military strategies have elements in common with data science strategies. It has been recognized that military means are often used in support of political objectives. This was starkly shown in an exchange during a meeting around the end of the Vietnam War between US Colonel Summers and North Vietnamese Colonel Tu when Colonel Summers observed “You know, you never defeated us on the battlefield,” and Colonel Tu retorted “That may be so. But it is also irrelevant.”2
The North Vietnamese strategy , which recognized that on the one hand the United States had unbeatable military strength while at the same time US troops could only remain engaged in Vietnam as long as there was political support in the United States, was effectively to avoid confrontation and wait the United States out. This has since been recognized as crucial to the success of the North Vietnamese.
This may appear to have little to do with data science. However, matching resources and capability to goals is critical to achieving those goals in any sphere. In the context of data science this may dictate what software you need, what training you need, and hiring decisions. For example, there’s no point in doing a course in text analytics if you’re not going to be able to use the new knowledge.
Alternatively, if, as was the case for the North Vietnamese, your capabilities are to some degree fixed, it means considering what projects you can take on carefully—at some level, even if you are an in-house data science department, success in the current project is needed to keep securing future projects, so accepting projects with a low probability of success has a strong chance of backfiring.
Tools: An Embarrassment of Riches
Data scientists are blessed with an abundance of different tools and a lot of time mastering their methods, but there is a lot less consideration put into the objectives. Intuitively, this risks disappointing customers and users, and there are now signs that this has been happening.
The flipside of such a large array of possibilities is that there isn’t much agreement over what the standard toolset is. There isn’t all that much agreement on what a data scientist is. Therefore, where there is always a tendency among experts to disagree on the best method to deal with a specific problem, this difficulty is especially acute for data scientists.
It’s inevitably the case when data scientists can come from a statistics background, a computer science background, or some other background that doesn’t instantly seem related. A strong sense of strategy offers a solution to this problem, as it allows you to define the concept of a data scientist that is most applicable to your team in your context—including both the skills available and the problems you are trying to solve.
This situation can be exacerbated by the need for statistical modelers to aim for very specific goals when they are modeling, evaluating the results by formal measures that do not always have a straightforward link between what the organization or client needs. It is incumbent on the data scientist to keep a firm grip on the customer’s idea of value. This will ensure not only that the model or data product succeeds in providing that value but that the customer sees the value.
Frank Harrell, in Regression Modeling Strategies,3 offers a number of strategies that effectively imply two orthogonal axes—one measuring predictive ability from none to superior, and the other measuring ability to explain the role of inputs from none to superior. He also suggests complementary strategies, where effectively separate models are created for prediction and inference.
This analysis provides low-level objectives (build a model that predicts well even if it can’t be understood or build a model that can be understood, even if other models could be better predictors), but the higher-order objectives aren’t in view. Note that Harrell’s text offers fully developed strategies, in that after presenting objectives, he then outlines a method to achieve those objectives.
The different strategies from Frank Harrell are all modeling strategies. However, there are also other kinds of low-level strategies to consider. Models have different real-world uses. Sometimes they are used to make decisions. Other times they may be used to aid communication or to persuade an audience.
However, number of times the word visualization is used in articles for data scientists (referring to tools for producing graphs as a means of communication) compared to the word communication, or, even more so, words like convince or persuade shows the focus on ways and means among data scientists over the ends they serve. Even more telling are articles where data science tools—the means to execute a data science strategy—are presented without a discussion of the objective they could be used to achieve.
The preceding strategy formulation also implies that there is a strong connection between available tools and objectives—strategy texts regardless of context often include examples of strategies where the objective is carefully chosen in recognition of minimal availability of resources—and often the ingenuity of a strategist is noted when they overcome a lack of certain vital resources in still achieving their objective.
Another area that is influenced by your overall strategy is which project management tools you will decide upon. While it is possible to choose a different methodology for each project, in reality, most teams stick with one methodology for general use and fit projects inside it. We will examine some of the alternatives in a later chapter, but the key takeaway is that an understanding of the overall team strategy is essential to picking the methodology that fits best.
The type of projects you select if you are an in-house data science team might also be influenced by the urgency of providing a pay-off. The notion of the quick win is very common and is often necessary to build credibility with upper management, but it will often come at the expense of bigger achievements, so if you know that you are being given more time, you can work toward something that will be more impressive at the end.
Understanding your goals and the resources you have to achieve them ensures that the goals are within your capability, a prerequisite for meeting them, and ensures that the goals you achieve actually assist your organization in achieving its goals.
A reasonable criticism of strategic planning as an activity is that too often the strategy is left on the shelf to gather dust rather than inform business decisions. However, the reality is that those decisions will still be made—individual data scientists and their direct managers choose not just which projects to pursue, but how much effort to invest in them.
Given the decision is unavoidable, you may as well base it on how useful the project is to company goals than more whimsical criteria, such as whether the project lets you use a cool new tidy verse package, or simply your emotional state on the day.
Getting this right means you can unlock a lot of additional value within your team and your organization, because as you are successful, and are seen to be successful, you get additional license and credibility that you can use to pursue more ambitious goals. Keep having successes, and you maintain that license and credibility within your organization.
Moreover, a clear strategy should provide clear links between goals and the actions taken to achieve them. This clear linkage is fundamental for anyone who wants to receive credit for their work, as it provides the means to explain how the low-level activities performed throughout any period contributed to the top-level goals.
Without those clear links, you may need a lot of goodwill and understanding from your audience to convince them that your team contributed to the extent that you believe.
The Value of Data Science in Your Organization
The idea that a data science function is beneficial to companies often seems to be assumed without proof. At least, the benefits are not well stated—it is assumed that whatever the data scientists do will instantly be of use to the company. In reality, there are a small number of types of ways at a high level that data science can be useful.
Stating what they are enables you to ensure that you align what your data science function is doing closely with what your company is doing. Just as it is easy to have a data science project that achieves all its objectives without helping anyone, it is also easy to have a data science function that produces fabulous work without improving any of the company’s goals.
A hazard to developing a successful data science strategy is that a data science manager will be selected from among successful data scientists. As such, they stand a good chance of having an astute strategic mind, but may not have taken any formal study of corporate strategy allowing communication on an even plane with the MBA grads who are most often retained as consultants or in-house strategy gurus.
It was once said that war was too important to be left to generals—from the data scientist’s point of view, it could be said that business management is too important to be left to the managers. Therefore, data scientists should educate themselves not just on how businesses work but on the vocabulary used to manage them, so they can communicate in the right language.
The most obvious company-wide use of a data science strategy is to research the company’s customers, especially in a consumer setting where large numbers of customers allows for statistical analysis.
An inverted version of this is risk analytics, where rather than identify opportunities to market differently among statistically large groups of customers, similar statistically large groups of customers are analyzed to understand the possibility of their causing the business to lose money. The personalities of people likely to be successful in these two areas intuitively could be quite different.
A third kind of data science is operational data science, essentially reducing costs. This is companies attempting to use data science to improve the way their business operates. This can be an approach to data science that lends itself to very creative approaches. It may include people looking deeply into logistics or people using deep learning identify the images in photos taken by drones, for example, photos of pole top transformers or overhead cables to automate the maintenance of these assets with reduced need for a human being to conduct the inspections.
There are also the goals of making decisions and choosing strategies. Here the data scientists are researchers advising senior managers on possible courses of action. The main tools in this arena will be optimization and decision theory, sometimes coupled with Bayesian statistics.
There isn’t a mystery about these categories; however, there is always value in stating out loud what you might have already figured out subconsciously. As always, writing down what you think you know helps check that you really know it (amazing how what you think you know is right in your head can look or sound wrong when written down or spoken), and ensures that everyone else that you think knows the same thing really does know the same thing.
The process of getting the team together to create the strategy is as much about discovering what everyone was thinking all along as it is a process of having brand new ideas.
Strategic Alignment
Earlier I touched on the idea that a team’s strategy needs to assist the wider organization in achieving its goals. In business school jargon this idea is called “strategic alignment” (knowing the jargon is useful when doing a Google search for more information).
Strategic alignment is recognized as a difficult area to master, although it is usually discussed from the point of view of senior managers trying to align their followers, rather than from the perspective of more junior people trying to align their own work upwards.
Another key concept in the way that organizational strategies are communicated throughout organizations is of “cascading strategies.” This is where senior management creates a strategy that is reapplied at lower and lower levels of the organization, each time becoming more detailed.
In my experience, the underlying analogy is flawed in that a “cascade” still functions when the lower levels are passive. In my experience, for this process to function correctly, the lower management levels need to be ready to receive—like a pitcher and catcher.
Although there is usually some lip service paid to the idea that the senior manager at each level will discuss the strategy with their subordinate, too much emphasis is placed on the idea that the more senior manager will direct the conversation. In the case of a data science team strategy, where it will often be the case that the more senior manager is far from expert on data science, this is a dangerous path to take.
Overall, it is critically important that the data science function both serves the overall mission of the organization and is seen to serve the mission of the organization. To achieve this, a prerequisite is that the data science team’s strategy is well aligned to the organization’s strategy.
At the same time, you need to ensure that it is easy to show that the alignment exists, and an intuitive way to do this is to create a document showing this strategic alignment.
Why Alignment Matters
It’s crucial for data science teams to be able to demonstrate their value to the organization by linking their activities to the organization’s mission because there will be doubters who believe strongly that data science is a waste of time.
In fact, while it may sound paradoxical, it’s still important when many people, including senior managers, are convinced that data science can make all the difference, as the kind of senior manager who has adopted data science because it’s the latest fad will be especially fickle. Hence, for both the unconvinced and the completely convinced real results that are meaningful in their context are vital.
While the preceding points relate to the way your work is viewed within your organization—which can still often have real-world consequences—strategic alignment has been shown to be crucial to the actual success of an overall organizational strategy.
A well-known article in Harvard Business Review4 by authors Bower and Gilbert showed that in many ways line management and functional management going about their everyday tasks did more to determine strategy than the activities of the most senior management. It may also be observed that many corporate strategies are vague enough to allow substantial room for an individual to put their own stamp on them.
It is therefore easy to implement the corporate strategy in your area in either a way that does uphold its intentions, or in a way that does not. Making a genuine effort to implement in a way that upholds the intended purpose is a clear way to gain credibility and influence with upper management, and make your functional area more important to the overall business.
Pro Tip
The examples used in the Bower and Gilbert article are also a powerful illustration of how personal influence can trump formal power. Their opening example talks about a factory that was built by line managers without senior management approval by purchasing components in increments under spending authority limits. In this scenario, someone had the idea and influence to get the others involved to go along with it. Maintaining and using influence as data science experts is crucial to ongoing data science success.
There is a very real sense that regardless of any data science strategy above you, the strategy you create for your team will be the data science strategy that is actually implemented in your organization.
Having this real power over your own direction gives you a great deal of autonomy, but it can be yanked away quickly if a perception that the company’s goals are not being achieved ever develops. The way to prevent that happening is to ensure both that the work you do supports the company’s goals properly and that senior people in the organization know that your work is supporting those goals.
Both of those things are far easier to achieve if the data science team has a clear, documented strategy that fully explains how the team’s goals align with the organization’s goals. The first step in developing a team strategy that achieves alignment with the organization is a careful study of the organization’s strategy document.
Working With the Organization’s Strategy Document
Most organizations renew their strategy at intervals and publish at least the top view version (sometimes the detailed roadmap may not be available to everyone) within the organization. It may be in a folder in a shared drive or on an internal company web page, for example. Get yourself a copy of this document, and analyze it.
Every organization has slightly different overall goals and mission. The strategy document describes what success for the whole organization looks like. That may mean chasing technical perfection, it may also mean doing social good, such as improving safety, it might be maintaining a high level of market share, or it may mean having the lowest costs. The challenge for the data science team is to decide how the data science team supports those goals.
Clearly, this will vary widely according to what the goals are, and what resources are available. In an organization driven by cost, the strategy might be centered on a search for automation opportunities. Alternatively, where the data science team is focused on risk analysis, the team’s mission may be chasing incremental improvements on existing practices.
You may think that the connection between some or all of your team’s usual activities and your organization’s top-level strategy is both obvious, and well known to all members of your team. This is great—if it’s true. It’s often the case that incorrectly assuming people know something when they don’t is more damaging than knowingly keeping them in the dark.
At the same time, if you stay with a business long enough, they will change their strategy, and if you don’t you’ll move to a new business—with a different strategy. In both scenarios, your ideas, if not the rest of the team’s ideas, on what the mission is will need to change—just because the mission was understood on day one doesn’t mean you can assume it remains well understood on any later day.
Having your team brainstorm ways to connect or answer the organization’s strategy with you is still a useful thing to do, even if you feel that the connections between the top level and the team level are obvious. You may believe that each team member understands these connections, but only when you hear them say it do you know it.
You may also think that everyone agrees on what those connections are—but data scientists are experts, and any room of five experts has at least six opinions. Again, hearing people express what they “know” out loud means you know what they think rather than just guess.
Documenting Your Strategy
Strategies are complicated and neither easy to remember nor understand. It makes sense to create a record of what’s in them to refer back to and to communicate their content to the intended audience.
Depending on the level of detail and complexity, the structure and content of these documents can be varied. A strategy document intended to be used for a team rather than an organization could be expected to take a relatively light approach. However, there are some parts of strategies created for larger groups of people that can still be useful at a team level.
Firstly, it can be useful to create a document that defines the mission and strategy for your team. These are common at the company level, but are sometimes forgotten at the departmental level—but if a department or team isn’t very directly working on either selling the company’s products and services or delivering them, a mission statement, possibly accompanied by team values can be useful to define how the team fits in the bigger picture.
The other crucial element of a strategy document is the plan itself, with details of the goals, how they are going to be achieved and who is responsible for getting there. Given a theme of this chapter has been the importance of aligning the team’s goals with the organization as a whole, an intuitive addition is a statement on how each of the team’s goals relates to the company’s goals.
However, note that the same temptation to get lost in the details during analysis applies to a strategy for data analysis. A short document that is referred to often, with some items omitted, is more useful than a comprehensive tome that no one looks at because they can’t ever find the section they’re looking for. We will see that the simplest model for the job is usually the right one—it is the same for strategy and other similar documents.
The overall aim is to provide a short document that can be referred to often, and that leaves little room for differing interpretation. It should also be applicable to a range of possible data science projects, and help people understand both when and whether to do projects.
A Mission Statement for Your Team
The idea of writing a team mission statement isn’t new—it’s certainly encountered in Agile practices. Mission statements come in a variety of shapes and sizes, and although they most frequently encountered at the organizational level, sometimes they are useful at the team level.
This may be especially true for data science teams in large organizations that aren’t intrinsically data science or similar consultancies. In this case, the team’s mission statement allows you to align the team’s objectives with the organization’s objectives.
A good mission statement can be read aloud any time you decide whether or not to do a project—if it doesn’t fit the mission, don’t do it (if you have that power). If you don’t have the authority to say no, write a business case for not taking on the work, even if sending it isn’t feasible.
At the same time, the purpose of a mission statement is to unite people behind a common purpose. Just as important as a common purpose is a common way of doing things and approaching problems that allows your team to put a distinct stamp on your work.
The document will need to include details of how your team’s strategy aligns to the top-level strategy. This may not necessarily mean that it fits to every point, but it should mean that it fits to the overall mission.
A mission statement can have four major components5—the mission, the vision, values, and major goals. You may feel now, and eventually decide, that your small team doesn’t need such a large document. However, I’d say not to be too hasty about a couple of these.
When aligning the two strategies, consider what you are currently doing, and how well it fits to the organization’s mission. Do you need to change? If you change, do you have the resources to make that change—will you need new data sources, or will you need to have new training?
A mission statement for your team, especially when created by the whole team together, can be an excellent platform for defining the way the team expects to aid the organization as a whole. Aim for something simply expressed, and make it as accessible as possible so that it gets used.
Presenting Your Team’s New Strategy
A strategy document is no use if you keep it private. By involving your team, you’ve increased the chances that they sign on to the team’s strategy, but it’s still not guaranteed. You need to go through it with them and ensure that they really understand it, and know their part in achieving it.
Having a meeting with your team members is the minimum of what can and should be done to ensure that the strategy is understood. Be prepared to discuss it at the beginning of each project, and during regular team catchups. Make sure it’s easily available to your team members, and encourage them to refer to it often—the best way being by doing that yourself.
To prepare for the meeting consider why you are doing this—answering both the what and the why is crucial to the explanation. Give background on the company’s current situation. This is a good time to give a very brief update on where the company and the department are at (but not a good time for a detailed view that distracts from your message). That way, everyone will be hearing the new strategy from a position of equal awareness of the external situation. It also means that anyone who might have felt left out of things is brought on board at the same time.
While this recap on your organization’s landscape needs to be brief, and therefore you need to be selective and tailor the talk to what matters at this particular point in time, there are some general issues you might like to consider.
You should expect questions, and you should be able to anticipate some of them. That is, it should be part of your preparation to brainstorm some of the questions you would ask, and to attempt to put yourself into the brains of a couple of your employees, and try to predict the questions they may ask in order to prepare some answers. Some of the things you may wish to put into the main presentation, others you may wish to leave, and only ask if they do come up as a question.
Pro Tip
Assign different parts of your team to different team members who can be the champion for that section. For example, someone in your team could be the methodology champion or the data sources champion. This will get the buy-in of those particular team members, and mean that you don’t have to go it alone in rolling out the strategy. However, you may not think it’s a good idea to spring these roles on people at a positive meeting when you’re trying to attract buy-in. Instead, have a one-on-one with those specific people, and use it as an opportunity to road test that element of the plan and the presentation.
Keep a close watch on your tone and overall presentation when you present this strategy for the first time. The point is to get your employees excited, if not as pumped as if they were at a rock concert. Rehearse your talk, and listen carefully to the tone of your voice. It needs to communicate excitement.
Consider the time of day where you have your most energy, and the time of day when you have the least. Don’t schedule the talk on a day when you are running from meeting to meeting or when there is a distracting deadline to meet, or at least do as much as you humanly can to avoid those situations so that you have the time to do the presentation properly, and the time to take questions unhurriedly afterward.
Just as your team has a strategy, and, as we will see in Chapter 2, so does each project you attempt, so does your talk. Be clear in your mind what you are trying to achieve with this presentation. In this instance, merely informing is not the whole story—you are trying to get your team to sing on with you for the way you will be doing things for the next 12 months (or your preferred time frame).
Overall, how well your presentation is received at its first public airing will rely on how well you prepare, and how well you can empathize with your own staff to ensure you have the right answers at your fingertips.
Putting a Strategy into Action
A strategy has very little—if any—value until it has been used to make a decision. The size of the decision isn’t important, but at some point it should be expected that the strategy is used to guide a decision.
Typical decisions in a data science life cycle could include choice of algorithm; choice of data; how much time spent on data cleaning; how much time to spend on feature engineering. While questions like which algorithm to use are part strategic, part technical, questions that begin with “how much time/effort to spend on X?” can be seen as almost totally strategic.
Another important time to go back to the strategy is during the hiring process. I’ve met recruitment agents whose workaround to the problem that neither companies nor candidates truly know what a data scientist is has been simply to ask “How do you define a data scientist?” to anyone they speak to. A clear strategy that supports a strong understanding of the skills you need on your team side steps that problem.
It’s important to get this right, because on the one hand senior management is unlikely to include people with a deep understanding of data science, at least not for some years from now. Hence, it will be up to the managers and team leaders to translate the overall company strategy into a strategy for the data science team.
On the other hand, a frequent reaction from senior managers where there appear to be problems is to manage more. That is, if results aren’t achieved rapidly, or at least not rapidly enough, a plausible outcome is that senior managers will try to involve themselves more and give more instructions, regardless of whether those instructions make sense in a data science context.
Pro Tip
The comment that interference can lead to worsened results isn’t intended as a dig against senior management, but rather as a comment on human nature. Most people, if they see something going wrong want to fix it. That they are at risk of worsening the problem comes from the frequent warning6 found in Statistical Quality Control manuals that unnecessary adjustments frequently lead to exactly the poor process outcomes they are intended to avoid.
If the instructions are inappropriate, this can lead to a vicious circle where the team performs progressively less well, while the senior managers try to manage more, making the problem worse.
All of this can be avoided by setting goals for your team that relate to the organization’s goals, but use words that make sense to your data scientists. Partly because humans have a tendency to forget and to break promises, and partly because data science is usually a group activity, it’s important to write these words down.
When making this strategy, be aware of the need to build support. We will see in future chapters that achieving the best results is often dependent on getting good advice and information from the people in the areas you are trying to assist, as well as from other subject matter experts. Building a pipeline of projects that provide early success can be instrumental in ensuring you continue to enjoy goodwill from this group of people, who are vital to your overall success.
Showing You Mean It—Behaving As If Your Strategy Matters
Nothing kills a leader’s message as quickly as the leader’s followers seeing the leader acting against their own message. Conversely, the most powerful way to communicate a strategy is to act on it in clear view of your followers.
When you are assigning tasks to yourself, therefore, make a point of telling the team what you are working on as much as possible and also make a point of explaining how your work fits into the overall strategy.
Also make sure that you show that you are tying your decisions back to the strategy across your usual activities. Mention that it’s the reason behind picking a particular project or avoiding another, and make it a part of performance reviews.
Giving your team the idea that you don’t care about something is a shortcut to them not caring about it either. It’s not a secret that actions are louder and easier to hear than words. It’s not different in relation to a team strategy than it is anywhere else—if you want people to do something, the best way is for you to start doing it yourself, somewhere you can easily be seen.
Strategy and Culture
Throughout this chapter, we have concentrated heavily on written strategy. However, nonwritten strategy, usually conveyed as a shared culture, can be a critical way to maintain and communicate strategy. For a strategy to succeed as a decision-making tool, for example, reaching a point where the culture within the team makes decisions aligned with the strategy nearly automaticaly creates the best results.
However, the details of how this is done are difficult to capture in a document. Possibly the most famous attempt to document culture in a working document was performed by Jews under the Roman occupation. Recognizing that their written record of religious and cultural practices, the Torah, covered only a fraction of those practices, they attempted to write down their oral law, and created the Mishnah, ostensibly the “Oral Torah,” which was far longer than the “Written Torah” (the first five books of the Jewish Bible or Christian Old Testament).7
While clearly documenting the culture of an entire society is a far greater task than for a functional team within an organization, the point is that creating a document to define culture may be too large a task.
Instead, consider the use of routines and rituals that create the right culture for your team. These will need to suit your team’s makeup and context, and are better invented in conjunction with your team to make them “sticky.”
Also consider how to incorporate some of the other teams that you work most closely with into the data science team’s culture. As one of the groups in an organization with some of the most specialized training, it is especially easy for data science teams to become isolated from other areas, leading to a “silo” effect.
The most obvious remedy is to interact more with the teams around you, and it may sometimes be easier to do that within organized or semi-organized social occasions than during your day-to-day work. Talk to the people around as well as the people inside your team to get the maximum effect.
The benefits of incorporating the other teams around you are especially important, considering the need to understand the organization’s goals as deeply as possible. It’s inevitable that people in those other areas will have a different idea of what those strategic goals mean—indeed, this is intuitively a reason why executing on a top-level strategy is so often very difficult.
Maintaining links with the other teams is crucial to ensuring that you are able to keep tabs on what the top-level strategy means to other people. It, therefore, ensures you avoid aligning your team’s strategy to an interpretation of the top-level strategy that others in the organization won’t recognize.
Prepare for Friction—Overcoming Obstacles to Your Success
A great deal can go wrong when delivering a strategy. Some of the obvious things that have been covered include poorly communicating the strategy, and poorly aligning the strategy with the organization’s goals. Obviously, these are just a couple of obstacles you may encounter.
One of the assumptions of this book is that there are organizations out there right now which have incorporated data science into their overall strategy but will not get the results they want due to a poor understanding of what data science can and cannot deliver.
One of the most difficult and most universal pieces to strategy implementation is change management. A strategy that requires no change is unlikely to achieve anything, and unfortunately, as all employees including data scientists are humans, there will be people who do not want to change.
- 1.
Ensure you explain the reasons for the new strategy.
- 2.
Address any concerns that arise as you explain.
- 3.
Maintain a positive attitude (you are senior management’s ambassador).
- 4.
Have some of the answers to hand on things that will be needed to achieve the strategy, such as training, new tools, or new data.
Although ultimately people need to sign on to a new strategy because it’s their job, there are still many things that can be done to make the process easier. Like many things in life, a lot of these come down to knowing what to expect and preparing accordingly.
Change management as a process is one of the most studied aspects of management and leadership, and as a result, there are many resources available. As a data scientist, mastering this particular skill will be crucial for your ability to lead not just your own team but often to how well your models are received and used in the wider world.
We will return to this point multiple times throughout this book—how much your models are used and valued by your customers will very frequently depend less on how effective the model itself is, but on how well you manage the adoption process. Hence, although this skill has been introduced in the context of how to properly introduce your team’s strategy, in reality, your skill in this arena has much wider implications for your overall success.
On a final note, research8 has shown that one of the most significant reasons that strategies fail is that they are developed by senior management who are then unable to directly verify the implementation—instead they are reliant on their middle managers to communicate the results.
In addition to the simple actions listed previously, there is no doubt that one of the most effective ways to assist with the process is to be seen to embrace the change yourself. Compared to a senior manager without the advantage of working alongside many of the workers in an organization, you have an advantage that you can be seen actively implementing the strategy on a daily basis.
If you are a data science team leader or a data science functional leader, you are in a more fortunate position of being able to see directly how the strategy is implemented on the ground. Although you should take the challenge of change management seriously, there is a lot of reason to feel optimistic about implementation.
Close the Loop—Check Your Results at the End
A common mistake is to produce a strategy, and then fail to evaluate its success—or otherwise. In contrast, by returning to the goals you set out to achieve at the beginning and evaluating your performance, you ensure that the lessons of what went wrong and what went right aren’t forgotten. At the same time, you can adjust your view of what you are capable of, and what your goals should be.
Somewhere along the way, you made some mistakes, so you should reap their value. Somewhere along the way you realized that there was a better target than what you were actually aiming for—again you should adjust. It was expensive in both cases, so don’t let yourself be short changed by not getting the full value of those mistakes.
Meanwhile, circumstances have changed since you wrote the original strategy. Different tools are available, and probably different data are available. If not a complete new source, then at least you now have a greater sample of the kind of data you were using originally, as data accumulated while time marched on.
Your team, too, has changed—they are more experienced, and hopefully more skilled and a little wiser. It should be possible to demand more of them than it was at the beginning of the process.
On the other hand, maybe some team members moved on, and the overall team composition changed. Take the opportunity to reassess your strategy in light of the overall team’s new skill set.
Keep a record of these sessions, and they will become valuable resources, not just as an input to the next round of strategy creation, but as a generalized “lessons learned” library. You wouldn’t throw out your code snippets, or data sets, you’d archive them. Do the same with these lessons learned.
A Strategic Thinking, Planning, and Doing Life Cycle
A strategy, like a model, has a useful life beyond which it is used up or wears out. The final step in the process, the review step, helps you decide when this has happened and sets you up to get the maximum benefit from what you learned applying the old strategy when you sit down to begin to write the new strategy.
Aside from the lessons learned that relate to how well the strategy was executed, there is also the question of how fit for purpose the strategy itself continues to be, partly determined by any changes that the organization’s senior management have made to the top-level strategy.
There may be an automatic process in your organization that reviews that top-level strategy and, therefore, prompts you to look at the team strategy. There may not.
Where there is a process, you can get the most out of it by using it as a prompt to realign your team strategy to the top level. If there isn’t a process, make your own process, and put aside time on a regular basis to check whether your strategy still supports the top strategy.
That time is also an excellent opportunity to share some of your wins. Although your organization most likely has a formal performance review process where you are expected to account for yourself over the year, they are a poor forum to let senior management know your wins. Moreover, as they are one-on-one, you lose the morale benefits that come with praising your team members in front of more senior managers.
Take the opportunity when you do your own review to compile highlights of what you’ve achieved, and present it to senior management. This allows you to achieve the final important benefit to having a well-defined team strategy in alignment with the high-level strategy—the chance to link your work with the organization’s success in the minds of the organization’s top managers.
The chance to promote your team’s work and highlight its value is what this has been leading up to. The process of creating a well-aligned strategy, allocating time to the projects that support it the most, and reviewing your progress against your initial goals ensures that when you do so, the result can only be that your team is respected and credited for the value they bring to the organization as a whole.
The review process, which completes the cycle by checking the outcomes against the goals and feeding the results back into the planning process for the future iterations of strategy planning, is the essential last step that ensures that the value of your team’s efforts is fully realized.
Summary
According to Herb Kelleher, founder of Southwest Airlines, “Strategy is overrated. We all have a strategic plan, it’s called doing things.”9 That may be true, but it’s unlikely that by fastening pieces of wood together with no guiding principle that a house will naturally emerge. As noted earlier and reaffirmed by the preceding quotation, the decisions that make up a strategy are made on a weekly, daily, or sometimes hourly basis. Without a strategy the reasoning will be local, and will often be arbitrary and isolated from the other decisions. With a strategy, those decisions can be validated against a higher purpose.
A data scientist embedded in an organization that is not primarily a data science consultancy needs to align themselves with their company’s overall strategy—there is a task of translating the goals of the company as a whole into goals within the data science function which support those goals. That process is crucial to the success of the strategy.
Even data scientists who work in consultancies can benefit from a greater understanding of their clients’ larger goals. Every project has a context—even if the client understands you did an excellent job on a project, it can still feel tainted on their side if it was a poor fit for their overall goal.
The resultant data science functional strategy is both a decision-making tool that informs people on both what work to take on and how much importance to place on it. At the same time, it is a communication tool that ensures you can explain how each project and the department as a whole are supporting the overall business.
Any strategy’s success depends not just on its quality as a strategy but how well it is communicated. Involve your team with its implementation and keep the way that it is expressed as simple as possible.
As senior management are seldom data scientists themselves, they aren’t able to stay on top of what’s happening in your area at a detailed level. It is therefore not to be expected that they are able to provide a strategy that is perfectly aligned to your function.
For best results, you need to have a certain amount of self-reliance, but self-reliance that is sympathy with the organization’s goals, rather than in pursuit of your own goals. In this way you have a strong foundation to fill the gaps in data science strategies that more senior generalist managers cannot.
Of course, more than that, for a strategy to be effective it has to be used. Make it a habit to think through decisions in terms of your strategy. When deciding which of two projects should go ahead, or which should get the most time or deserves the most experienced person on your team in case her dance card is full, consider which project advances your strategy most effectively.
We will see later in this book that models need to be maintained to ensure they perform as expected throughout their life. The same is true for strategies. Not only that, reviewing a strategy provides a great opportunity to learn from your experience, and to make sure that what you learned is captured.
There are people who argue, similar to Herb Kelleher, that it isn’t necessary to have a formally documented strategy. While the necessity may be arguable, having a simple team strategy that aligns well to company goals is a great way to ensure that your work not only supports those goals but also can be seen to support those goals.
It may be that there are occupations where it’s instantly clear how they support the goals of the organization as a whole. It’s intuitive that airline mechanics and pilots may know easily how they fit within the structure of an airline and support the organization’s goals (although this doesn’t mean it’s guaranteed).
Data science isn’t an occupation where it’s instantly obvious how the contribution is made—let’s face it; there are nearly as many definitions of data science as there are data scientists. Expressing your particular definition out loud and writing it down will pay dividends in added clarity.
A strategy has a number of parts. They can include a mission statement, a strategic plan covering details of intended goals and how they are to be met, or a set of cultural practices that have the effect of leading people to answering certain problems in a certain way. Understanding and untangling these elements gives you a platform that will allow you to not just achieve more but also to receive your fair share of the credit for achievements, and therefore continue to be given opportunities to do innovative work for your company.
No one has enough time to do everything that appears worthwhile, both as individuals and from a team perspective. Understanding the company’s goals yourself and ensuring everyone in your team understands them, combined with a knowledge of how your data science activities support those goals, ensure that every ounce of your effort improves your company, and that you can explain to anyone outside your team how you contribute.
In Chapter 2, we will move closer to the metal by discussing strategy for individual projects. Although some of the philosophy will carry over, we will still see a great number of new ideas.
Team Strategy Checklists
In order to cover all the important aspects when developing a strategy, it can pay to have a short checklist to consult. The following is a nonexhaustive list of useful questions to ask yourself about your team’s strategy to ensure everything important is covered.
Team Context
How much goodwill currently exists? The level of goodwill often determines the timeframe within which you are supposed to deliver
What skills currently exist in the team? Are they a good match for the data that is usually available?
Where does data mostly come from? Do users or clients bring it to you, does it currently exist within the organization, are you expected to effectively obtain data for the organization, for example, via web scraping?
Following on from the above, will you usually be looking at the same data set or looking at different data sets depending on the occasion?
Is there an accepted industry standard way of looking at problems, for example, is a generalized linear model considered the “gold standard”? Is there an advantage of a new approach?
How accepting of new approaches is your industry generally? What sort of blowback will there be if you introduce something new?
Are there resources available in your organization to help educate users, or will that be your responsibility? Who are users likely to ask for help if they encounter a problem?
Alignment
Can I use the team’s strategy to explain how data science helps the organization achieve its goals?
If I follow the team’s strategy, will I automatically achieve the organization’s goals?
Will the organization’s strategy be stable over the intended period of the team’s strategy?
Strategy Documentation
Is your strategy document easily available to your team?
Does the formatting of the document make it easy to read?
Have you made a point of communicating the new strategy with your team members and discussing how it relates to their individual work?
Presenting Your Strategy
Choose a time when both you and the team will have as few distractions as possible
Anticipate the most likely questions and prepare responses
Speak one-on-one with champions for particular aspects of the strategy prior to presenting it
Culture
What are your team’s rituals? Do they help or hinder in achieving the team’s goals? Do they make the team open to change, or do they reinforce a team filter bubble?
Do your rituals include the teams and others around you?
Acting on Your Strategy
Is it standard practice to consider whether a new project under consideration supports the team’s strategy?
Do you adjust the descriptions in job ads to attract candidates whose skills support the overall team goals?