When you think of the perfect data science team, do you imagine a copy of the 10 of the same computer science and statistics professor? I hope not!

Google's Geoff Hinton is my hero and a great researcher for deep learning, but I hope you don't plan to let his application data science team work with him, not everyone else!
Google's Geoff Hinton is my hero and a great researcher for deep learning, but I hope you don't plan to let his application data science team work with him, not everyone else!

Applied Data Science is a highly interdisciplinary team movement. The diversity of perspective is important! In fact, opinions and attitudes are at least as important as education and experience.

If you want to make data useful by making intelligent engineering methods, then this is my view of the order in which the team evolves.

#0Data Engineer

Of course, we count from scratch, because before you talk about data analysis, you need to be able toObtaindata. If you're working on a small dataset, the data engineering actually enters some numbers into the spreadsheet. When you operate on a more impressive scale, data engineering itself becomes a complex discipline. Someone on your team needs to be responsible for dealing with the tricky engineering aspects of providing data that other employees can use.

#1decision maker

Before hiring a Ph.D. trained data scientist, make sure you have an art and science decision maker who understands data-driven decision making.

Decision-making skills must be in place before teams can derive value from the data.

This person is responsible for determining the decisions that are worth making with the data, building them (from designing the metrics to making decisions based on statistical assumptions), and determining the rigor of analysis required based on the potential impact on the business. Looking for a thoughtful person, he won’t always say,"Oh, hey, I didn't even think about it when I thought about this decision."They have already thought of it. then. thats right.

#2 Analyst

Then the next employee is... everyone has already worked with you. Everyone is qualified to view the data and get inspiration, the only thing that may be missing is the familiarity with the software that is familiar with the job. If you have ever seen digital photos, then you have completed data visualization and analysis.

Learning to use tools such as R and Python is just an upgrade of MS Paint for data visualization; they are just multi-functional tools for viewing more types of data sets, not just red, green and blue pixel matrices.

If you've ever seen digital photos, you've completed data visualization and analysis. This is the same thing.

Hey, if you have a stomach, look at the first five rows of data in a spreadsheet, then this is still better than nothing. If the entire workforce has the right to do so, then you can better grasp the pulse of the business, rather than no one is viewing any data.

Nessie 1934: This is data. Draw wisely.
Nessie 1934: This is data. Draw wisely.

The important thing to remember is that you should not draw conclusions other than data. This requires professional training. Just like the photo above, here's all you can say:" This is what is in my data set."Please don't use it to conclude that the Loch Ness Monster is real.

#3Expert Analyst

enterLightning version! This person can view more data faster. The game here is speed, exploration, discovery... fun! (Another term for analysis is data mining.) This is not a role that focuses on rigorous and careful conclusions. Rather, this is the person who helps your team focus on your data as much as possible so that your decision makers can be more cautious about what is worth pursuing.

The work here is speed and encounter potential insights as quickly as possible.

This may be counterintuitive, but don't work with the most reliable engineers who write gorgeous, powerful code. The job here is speed, experiencing potential insights as quickly as possible, and unfortunately those who are too good at code quality may find it difficult to scale data quickly to work in this role.

Those who are fascinated by the quality of the code may find it difficult to play a role in this role.

I have seen analysts on engineering-oriented teams being bullied because their peers did not realize what "good code" means for descriptive analysis. The great thing is "fast and humble" here. If the fast but sloppy programmers don’t get much love, they will leave your company and you will wonder why you are not keeping the pulse of your business.


Now that we have made all these people happy to explore the data, we better let the surroundingPeopleImpeding the feeding frenzy.It's safe to look at Nessie's "photos", as long as you have the discipline to keep yourself away from things that actually exist... But what about you?Although people are very good at thinking about photos reasonably, other data types seem to convey common sense outside the window.It may be a good idea to have people around you prevent the team from making unfounded conclusions.

Inspiration is cheap, but rigor is expensive.

Lifehack: Don't make a conclusion, don't worry about it.I am only half joking. Inspiration is cheap, but rigor is expensive. Pay or satisfy yourself with pure inspiration.

Statisticians help decision makers draw conclusions safely outside of the data.

For example, if your machine learning system works in a dataset, you can safely conclude that it isThe data set works. Does it work when running in production? Should you launch? You need some extra skills to deal with these issues. Statistical skills.

If we want to make serious decisions without perfect facts, let us slow down and take a cautious attitude. Statisticians help decision makers safely draw conclusions beyond the data analyzed.

#5Application Machine Learning Engineer

The best attribute of the applied AI/machine learning engineer is not an understanding of how the algorithm works.Their job is to use them, not to build them. (This is what the researchers do.) Professional debate code allows existing algorithms to accept and drain through your data set, which is exactly what you are looking for.

In addition to quickly coding your fingers, you also need to find a personality that can handle failure. You almost never know what you are doing, even if you think you did. You can run the data as quickly as possible through a bunch of algorithms to see if it works properly... It is reasonable to expect that you will fail a lot before you succeed. A large part of this work is to get involved blindly, and it takes a certain personality to enjoy this.

Perfectionists tend to struggle as ML engineers.

Because your business problems are not in textbooks, you can't know in advance what works, so you can't expect to get the perfect results in the first place. It doesn't matter, try many methods and iterate the solution as quickly as possible.

Speaking of "operating data through algorithms"...what data? Of course, analysts think these inputs may be interesting. This is why the analyst makes sense as an early employee.

Although there are many tinkerings, machine learning engineers must deeply respect the rigorous process part of the process: assessment. Does the solution really work for new data? Fortunately, you made a wise choice in your previous hiring, so all you have to do is pass the baton to the statistician.

The most powerful application ML engineers know very well how long it takes to apply various methods.

Impressing options when a potential ML employee can try them on various data sets is impressive.

When a potential M L employee can try them on various data sets Wait for the selection of the item, leaving a deep impression.

#6Data Scientist

The way I use this word, the data scientist is a fully expert in the first three roles. Not everyone uses my definition: you will see the work application there, people call themselves "data scientists", when they really master one of the three, it is worth checking.

Data scientists are all experts in the previous three positions.

This character is in the 6 position because hiring a real three-in-one is an expensive option. If you can hire one within the budget, this is a good idea, but if your budget is tight, consider upgrading and cultivating existing single role experts.

#7Analysis Manager/Data Science Leader

The analysis manager is the goose that produces golden eggs: they are a mixture of data scientists and decision makers. Their presence on the team acts as a power multiplier, ensuring that your data science team doesn't break away from weeds, rather than adding value to your business.

The decision maker + data scientist hybrid is a force multiplier. Unfortunately, they are rare and difficult to hire.

This person stays awake at night, the problem is " How do we design the right questions? How do we make a decision? How do we best distribute our experts? What is worth doing? Are the skills and data meeting the requirements? How do we ensure good input data?"

If you are lucky enough to hire one of them, please stick to it and never let them leave. Learn more about this role here.

#8Qualitative Expert /Social Scientist

Sometimes your decision maker is an excellent leader, manager, motivator, influencer or navigator of organizational politics... but not skilled in the art and science of decision-making. Decision-making is not just a kind of talent. If your decision makers do not hone their craft, they may cause more damage.

You can use qualitative experts to enhance them instead of firing an unskilled decision maker.

Don't fire an unskilled decision maker and strengthen them. You can hire them to upgrade in the form of a helper. Qualitative experts add their skills here.

This person usually has a background in social sciences and data-behavioral economists, neuroeconomists, and JDM psychologists receive the most professional training, but self-taught people can also excel.This work is to help decision makers clarify ideas, check all angles, and translate ambiguous intuitions into language.

Thoughtful instructions so that other members of the team can easily execute. We are not aware of how valuable social scientists are. They are often more capable than data scientists to translate the intuition and intent of decision makers into specific indicators.

Qualitative experts will not call any of the lenses. Instead, they ensure that the decision maker has a complete grasp of the shots available for the call. They are also trusted advisors, brainstorming companions, and decision makers' sound boards. Engaging them is a great way to ensure that the project begins to move in the right direction.


Many hiring managers think their first team members need to be former professors, but you don't actually need those PhDs unless you already know that the industry won't provide the algorithms you need. Most teams don't know in advance, so it makes more sense to do things in the right order: before building your own space pen, check if a pencil can get the job done. Start by, if you find that the existing solution doesn't give you too much love, then you should consider hiring a researcher.

If the researcher is your first employee, you may not have the right environment to make the most of them.

Don't bring them directly to the bat. It's best to wait until your team develops enough to figure out that they need researchers. Wait until you have exhausted all available tools and then hire someone to build expensive new tools.

Before you invent a pen that works in space, check to see if your existing solution meets your needs.
Before you invent a pen that works in space, check to see if your existing solution meets your needs.

#10 +Other people

In addition to the characters we see, here are some of my favorite people who are welcome to participate in decision intelligence projects:

  • Domain expert
  • Ethicist
  • software engineer
  • Reliability engineer
  • UX designer
  • Interactive visualization/graphics designer
  • Data collection expert
  • Data product manager
  • Project/project manager

Many projects cannot do without them-the only reason they are not included in my top 10 is that decision intelligence is not their main business. On the contrary, they are geniuses in their own disciplines, and they have enough knowledge of data and decision-making to be very useful for your project. Think of them as having their own profession or profession, but they are passionate about decision-making intelligence and they choose to fine-tune it.

A large team or a small team?

After reading everything, you may feel overwhelmed. So many characters! Take a deep breath. Depending on your needs, you can get enough value from the first few roles.

Revisiting my analogy of applying machine learning as a kitchen innovation. If you personally want to open an industrial-scale pizzeria that makes innovative pizzas, you need a large team or need to cooperate with a supplier/consultant.If you want to make one or two unique pizzas this weekend-caramel anchovies surprise, anyone? – Then you still need to consider all the components we mentioned.You have to decide what to do (Role 1), which ingredients are used (Roles 2 and 3), where to get the ingredients (Role 0), how to customize the recipe (Role 5), and how to give it a taste test (Role 4) Before you can serve someone who wants to impress, but for a less important random version, you can do all of this yourself. If your goal is to make a standard traditional pizza, you don't even need it: grab the recipes that others have tested and tested (you don't need to reinvent your own formula) and ingredients to start cooking!