5 useful lessons from modest salmon
Three waves of artificial intelligence development according to Evo Pricing, based on DHL research.
Data is an unreliable friend, and one can hardly call it scientific. What to do with data science?
Over the past 5 years, I have interviewed over 1000 candidates for data analysts who want to work for Evo Pricing. In the process, I learned that the way the media portray this profession is fundamentally wrong; We do not just substitute data into ready-made algorithms.
There is a need to radically rethink our understanding of data science.
70 years of history in two paragraphs and one picture
At its core, data science is a beautiful name for a wide range of complex mathematical operations, which, for the most part, were invented in the distant past, but gained a second wind thanks to the use of significantly improved technical devices: more data, more computing power, more reasonable results at a lower price.
As the cost of storing and processing data has decreased, the amount of data has grown: a simple law of supply and demand works here, or, one can say, demand elasticity at the price of data. Price decreases – volume grows. Accordingly, someone must do something with all this information. And so the science of data came about.
Common misconceptions about data science
What Data science?
According to the University of California Berkeley, this science is one of the most promising career paths for qualified professionals.
In my opinion, the name “Data Science” implies a special approach to solving certain problems. There is certain data; and what shall we do with them, whatever?
In fact, investing a huge amount of money in some kind of data, from which, perhaps, one day you get something useful, does not sound very optimal both in terms of career growth and in terms of business strategy. Unfortunately, the industrial revolution of the 19th century left us a legacy of schools and universities for the training of a large number of workers, capable of only giving uniform answers to standard questions; and since then, little has changed.
And what will happen if we teach people to ask the right questions, and machines to find the answers?
Data Science could be a career dead end
Despite the fact that data science in its various forms is gaining popularity, such as artificial intelligence and everything related to this topic, the profession itself is good only for beginners.
Salary promises of more than 80 thousand dollars a year may seem tempting, but it’s not as simple as it seems. To truly succeed in working with data, you need to succeed in solving specific, significant and well-defined problems, rather than becoming a universal expert in the field of data or, even worse, in science, which, as the image at the beginning of the article shows, is quite outdated from an academic point of view.
Data and algorithms are powerful tools. But, like any tool, their effectiveness depends on how good the one who uses them is.
Want to succeed – grow Business Science
How to become successful while working with data? Focus on the problem you need to solve, and not on the data as such.
For those who want to use the data for commercial purposes, business science offers excellent strategies:
- a business problem must be identified, studied and resolved;
- research should be done scientifically;
- impact on business: a measurable, objective result;
For non-commercial options for working with data, the logic, however, is similar: start with a question / hypothesis, adhere to a strict methodology, then return to the aspect / question being studied and determine whether the effect of the data on the business has been proven or not. And so on a new one.
The question arises: how to do all this? We can explain this using a rather funny analogy.
Salmon Lesson # 1: Start Over
Modest salmon is not only delicious, but also for its 5-10 years of life manages to do a lot of right, logical things: his life begins at the end (at the mouth of the river), and then returns to the beginning (source), where spawning occurs, after which he leaves the river, giving way to a new generation of salmon.
Salmon offspring is born at the source, then floats downstream, learning the wonderful world of the ocean, and then returns back to the river, where it can claim the right to produce new offspring.
The average data analyst has a lot to learn from modest salmon. No matter how comfortable (and intellectually stagnant) to go with the flow to a new volume of data, such a simple, childish strategy will not lead to success in the long run.
Mature salmon will begin to move downstream, looking closely at the goal (river) that it wants to achieve and which it wants to influence, and then decides to slowly and painfully move upstream, gradually narrowing the amount of data (water) through which it needs to break through .
Salmon Lesson # 2: Don’t Swim Close to the Waterfall
I worked as a management consultant in Mckinsey & company 10 years. During my tenure, I strictly followed the traditional cascading model of work – it’s the “Waterfall” model: I always started by investing a huge amount of time, effort and client budget in the project. Explored everything as detailed as possible. Roughly speaking, the ocean boiled – and, in the process, killed poor salmon!
In fact, my team formulated the initial hypothesis, and then searched for relevant data to prove or disprove it. A kind of hypothesized thinking. In the best case, this can be called an effective quasi-scientific approach to business, in the worst – an expensive example of confirmation bias, when the data is selected to justify the result to which it was decided to arrive in advance.
This strategy may be suitable for highly strategic, long-term plans, but it does not give any guarantees to the client that tomorrow, and then in a year they will still follow the plan, especially in a world that is developing faster, more complex and more chaotic. My boss, Pwrap diamondlikes to say this: business is a film, not a static picture.
The risk of this approach is that it is possible to answer the wrong question, and also it does not provide the feedback necessary for the development, which affects success, despite the constant disorganization of the market. Today, data is the right model!
In the end, it was for this that appeared flexible development methodology. To allow for additional adjustments.
Salmon Lesson # 3: 80/20 Principle Helps Avoid Bears
At the top of the waterfall, even the most dexterous salmon can run into its sworn enemy – a large furry beast.
Watch out for the bear. Photo: “Leap of Death”, Peter Stahl.
When swimming upstream, each salmon can encounter unexpected, sometimes insurmountable obstacles, with terrible predators. Once the quiet waters suddenly become a seething waterfall, it becomes very difficult to swim.
In a desperate attempt to overcome the obstacle, the salmon jumps at its best and falls into the trap of a large hairy bear waiting for dinner.
Perfectionism – the main enemy of Business Scientist’s
Perfectionism is the trait that turns life into a solid report card. This can cause a person’s unhappy existence.
Water (data) can quickly become a shelter for a bear waiting for its prey. Ideal for swimming and life, it can suddenly become death. To avoid this, we need a different, more pragmatic approach.
Our salvation is 80/20 principle – focuses on what is really important, and allows you to get around obstacles, and not go ahead. Swim around and look for ways that can help avoid bears. Extreme, borderline cases practically do not affect the business! Then why strain?
Lesson from salmon No. 4: the less (data), the better (information)
The preparation of data-based results should take longer than their research. And by “more time” I do not mean “to put everything together in one pile at the last moment”.
Salmon is born in a small reservoir (data volume) – where it sets itself a narrow task, formulates a question; after that, he goes to a huge ocean of exploration, where he swims in a large volume of water and with big data; then he returns back to his little pond. Indeed, in order to explain the result, you need to thoroughly study the data obtained and form what will affect the business.
Work based on data should be carried out from the bottom up, but effective communication from the top down.
At some point, the scientist needs to finish cooking in the ocean of data and move on to formulating a message – how to get the idea from scratch to the recipient? To do this, go to top to bottom.
No need to create a fancy visualization that is so dynamic and confusing that it is clear only to techies. On the contrary, the wording should be simplified as much as possible. Spend LESS time on data and MORE on planning discussions.
Effective communication starts from the end, and then goes on to a million reasons why such a conclusion was made, and a million evidence supporting this conclusion.
I highly recommend reading Barbara Minto’s “Pyramid Principle,” which details how to best convey information through facts.
Salmon Lesson # 5: The Evidence Is Available
Starting from the end, talking about the tangible impacts that the study can have, you will gain the approval and confidence of your clients in your work, which could otherwise be called some kind of obscure algorithmic tricks. To understand if a black box works, you must first open it.
Just by reading about satellite navigation, you will not learn how to use it.
I really like working with pricing and supply chain applications, and in both cases, most often, you can only hit the big jackpot by moving upstream: planning, designing. But, there is always a “but.” You will never get anything unless you first prove the effectiveness of your work.
Therefore, I recommend starting with EOL cases such as reordering (for value chains) and markdowns (in case of pricing). And I understand perfectly that the best markdown is one that you DO NOT offer at all, because everything was correctly planned from the very beginning. But how problematic would it be to disrupt planning without first winning the hearts and minds of customers with tangible facts?
Be a Business Scientist, and the money will follow
What is more likely to pay off: to go, like everyone else, to study the methods and techniques of data science, or to invest in the study of business applications where the impact of data exceeds intuition?
First find your niche and forget about machine learning until there is an exciting problem that you want to solve. After I left McKinsey, I learned R in a day to start implementing my idea in the field of business science. But did I understand well what I wanted to do?
It took me about 10 years to reach the level at which I can safely work with the problem that I wanted to address.
Before you go to get a doctorate, you must first formulate the task of research – in contrast to the master’s work, the topic of which someone else gives you.
Before becoming a businessman, you need to come up with an original business idea – unlike a regular profession, where someone else gives you all the tasks.
This is the philosophy of life.
Dare to swim upstream, and instead of becoming another data analyst, try to find yourself in the field of business science.
Have a good swim!
Learn the details of how to get a sought-after profession from scratch or Level Up in skills and salary by completing SkillFactory paid online courses:
- Machine Learning Course (12 weeks)
- Learning Data Science from scratch (12 months)
- Analyst profession with any starting level (9 months)
- Python for Web Development Course (9 months)
- The coolest Data Scientist does not waste time on statistics
- How to Become a Data Scientist Without Online Courses
- Sorting cheat sheet for Data Science
- Data Science for the Humanities: What is Data
- Steroid Data Scenario: Introducing Decision Intelligence