ONES CTO Feng Bin｜How to make high-quality decisions at low cost

On January 6th, at the TOP100 Global Software Case Study Summit, ONES co-founder and CTO Feng Bin gave a speech entitled "Mechanization Principles for Efficient Management of Large Teams".

The following is the main content shared by Feng Bin that day.

Technical management issues for large teams

Let's first look at the problems faced by large teams in technical management.

First, is there a way we can make good enough decisions cheaply? Because in R&D scenarios, especially technical managers, decisions need to be made every day. This decision may be related to recruitment, including deciding who to hire, deciding which colleague to promote; it may also be related to a function or a requirement, including whether to select a customer, enter a market, and so on.

Secondly, in the R&D team of hundreds of people, how can the quality of everyone's work continue as expected? In large and medium-sized teams, technical managers are subordinated to managers, and managers subordinated to front-line engineers, and there is a certain hierarchical relationship with each other. As the top technical manager, there is no way to directly know how each person is working every day.

Secondly, if we have defined some standards, but after the training, people have not fully understood or are not used to implementing them, how can these standards be implemented quickly?

Finally, when we have standards, how can we make everyone follow the standards as much as possible while working without being rigid and flexible?

Here, I want to draw a conclusion first: we need to do "mechanized" execution based on structured standards and processes. "Mechanization" here is not a pejorative, but a neutral word.

structured standard

How to further understand "mechanization", we have to start with the difference between expert systems and structured standards.

What is an expert system? Taking recruitment as an example, when deciding whether to admit a R&D candidate, the company's engineers, technical leaders, HR, and even management will all express their opinions and express their views from different perspectives. Come to a conclusion - this decision-making method is called an expert system.

However, we found that there are many problems with expert systems: many practices in behavioral economics have proved that the success rate of expert systems in making judgments and decisions is not particularly high. At the same time, expert systems are often ambiguous. This ambiguity will make the whole The team felt that it wasn't clear enough, and maybe even unfair to some colleagues.

Relatively speaking, structured criteria have a better effect on decision-making.

The structured quantification method means that before we make a decision, we should first decompose the relevant things in some dimensions, then score it, and quantify the objects that need to be decided by various means. We can choose first.

This quantitative standard is easier to implement, because once we can quantify these things, we can even use machines to calculate points. A consequential and clear standard can produce a more equitable effect, and can also bring a sense of security to the team, and a sense of security can generate the driving force of multiple teams.

If we have a way to use structured standards to judge all aspects of a thing, then it will be easier to be implemented by our digital tools. Once it can be implemented, we can easily do statistics, analysis and prediction.

Structured Criteria for Decision-Making

Since the key to "mechanized" execution is structured standards, what principles should we follow?

From the perspective of technical management, structural standards can be divided into two categories, one is about decision-making, and the other is about execution.

So, what are the structured criteria for decision-making? When we make structural standards, we will list the various dimensions of judging a thing, such as drawing a radar chart. These dimensions are some principles that should be satisfied.

The first is the MECE rule. It emphasizes that each dimension should be independent of each other, not repetitive, and cover as much as possible the questions we want to judge.

The MECE rule is a thinking tool proposed by McKinsey consultant Barbara Minto in "The Pyramid Principle". Visually, it is like a jigsaw puzzle. Pieces are used to make a complete picture. If the pieces are spelled correctly, there must be not many pieces and not many pieces in the end.

The MECE rule is the basic skill of "structured thinking". When using the MECE rule, pay attention to three points: keep in mind the purpose of decomposition, avoid hierarchy confusion, and learn from mature models.

The second is the focus principle. How do we achieve a good standard? And why can this standard be implemented at low cost? This is where "focusing" should be taken into account.

"Focusing" means that when we judge a thing, we don't need to have too many dimensions. The most critical four to six dimensions are generally enough to allow us to make a relatively accurate judgment on a thing. Once we seek the greatest or the perfection, it is very easy to lead to problems such as high cost, poor execution, and poor training.

In many cases, we set some structured standards, and in the end they didn't work. Looking back, it is because there may be twenty or thirty dimensions in this standard that need to be judged. For example, in the recruitment process, we may judge the technical ability, communication ability, values, etc. of the candidate. Divided into many sub-categories, from computer basics, to his specialized technology stack - these complex dimensions make execution difficult.

The third is the principle of quantification. When we have dimensions, it is not enough. We also need to quantify each dimension, and quantification will involve the scoring of each dimension. Its scoring standards can be different according to different situations, but there are some directions that need to be grasped Live: The distinction should be high, clear, and easy to measure.

To give a counter example, when we describe someone's ability, sometimes we will evaluate it as good, excellent, excellent, etc. In fact, these words are all "degree words", whether it is a person who determines the standard or a person who implements the standard. People, it is difficult to make judgments here, because these words are very vague, and everyone may have different understandings.

At the same time, the points should not be too many. If there are too many points, the distinction will not be high enough, and everyone needs to think a lot. We generally use a five-point system or a ten-point system.

Implemented Structural Standards

After talking about the principles of structured standards for decision-making, let's look at how the structured standards at the execution level are implemented.

After the practice of ONES, we found that if the relevant structural standards are digitized on the tool, the efficiency in practical use will be very high.

First, if we are to actually drive behavioral change, we must internalize our entire habit. For example, in a software team project management scenario, if we want to standardize our work methods and processes, the best way to develop a habit is to let everyone work in a tool, and let the tool tell everyone that when the requirements are decomposed, What is the next step, solidify a series of processes in the whole system.

Secondly, if everyone can work in a digital tool, whether it is making a decision or executing a process, you can use the tool to collect relevant data and form some new conclusions and actions.

In the end, it will be a more open and transparent scene, allowing everyone to work better because it is more secure.

Let me give you a classic example so you can better understand how to enforce structured standards.

"Apgar Score" is a model designed by anesthesiologist Apgar in 1953 to judge whether a newborn is healthy. He considered five indicators in total: skin color, heart rate, facial expression, muscle tone, and breathing.

The second step is scoring. Set an integer fractional interval for each metric. For example, each indicator in the Apgar score can be scored as 0, 1 or 2. Like skin color, if the whole body is pink, it is 2 points; if the limbs are blue, it is 1 point; if the whole body is blue, it is 0 points.

The third step is to calculate the total score. There is no need for weighted average, just add them together. The Apgar score is out of 10. Then this judgment system stipulates that a total score of more than 7 points is healthy; a score of 4 to 6 is not healthy; a score of 0 to 3 requires immediate emergency measures.

This is a "simple and easy to implement" structured standard. Many diagnoses in the medical world today, such as some cancer screenings, use a similar scoring system. This method decomposes complex decisions into simple judgments in several dimensions. It is easy to operate, less affected by the doctor's experience and level, and has high accuracy because noise is greatly reduced.

Digitizer Embedded Standards

Regarding embedding standards with digital tools, I will introduce the specific practice of ONES.

In ONES Project, we can set up the workflow engine. It is what kind of process we follow when we are going to deal with a bug or requirement. Here we see that there are seven steps, which are too expensive for someone to memorize completely. Therefore, we can embed the entire process through a data tool, and use the tool to remind everyone what needs to be done at each step when the process is executed. In this way, our training costs and even documentation can be greatly reduced.

For example, when creating a new task in ONES, there are many custom attributes, which are required, and we can manage some tasks in a structured way.

Regarding the functionality of the ONES Wiki, here is a template for failure analysis, which is more straightforward than documentation and training. Digital tools can directly provide us with a preset standard, and everyone works within this standard to quickly complete some necessary things.

When the standard can really run, we must review it regularly, because simple standards are easy to implement. In essence, it is a trade-off between simplicity and effectiveness. With the passage of time and changes in the external environment, it is There will be limitations.

In addition, I also share the model of ONES for technical recruitment. Our model will be divided into four aspects. The first aspect is professional ability, that is, technical ability. The second aspect is learning ability, the third is thinking expression ability. We believe that a person’s expression ability reflects whether his thinking ability is structured, clear enough, and whether he can focus on the goal. The fourth is thinking level, Mainly determine whether the source of a candidate's driving force is external or internal.

In terms of learning, we will divide it into six scores from 0 to 5 to judge whether a candidate has active learning behavior and whether his active learning behavior is fragmented. Compared with the active and firm we just said For the words , excellent and good, it can distinguish different scores by one characteristic and one behavior. In this case, the scoring will be simpler and more accurate.

Make sure the decision is correct

To sum up, in the technical management of large teams, we often have to make many decisions, and we need to ensure that these decisions are correct enough.

In fact, we don't need to be 100% correct, we just need to be correct enough, and let everyone work according to a standard in their work to ensure a certain quality.

First, mechanized standards and processes are more efficient and operating costs are lower.

Structured standards and processes should be focused, clear, and simple. When we have such standards and processes, we can manage them with digital tools.

When everyone goes back to work in a digital environment, they will naturally follow some of the structural standards we have set before to ensure that they are implemented in place. At the same time, we only need to click a few buttons on the system to aggregate the desired data and complete the quantification efficiently, which is very helpful for our review and improvement.