GTLC Global Technology Leadership Summit | Progressive Way of R&D Management Improvement

On October 17, ONES attended the 2021 Global Technology Leadership Summit-Hangzhou Station (referred to as "GTLC Conference"). At the conference, Chen Liangyu, Director of ONES R&D and head of ONES Core business center, gave a speech titled "Progressive R&D Management Improvement Road", sharing the practical experience of ONES's R&D efficiency improvement with technical experts from all walks of life .

The following is the main content of Chen Liangyu's speech.

R&D Management

R&D management is a huge system, which needs to be considered from the big and small. In practice, business managers often set big goals, but they will encounter various difficulties when implementing them in a "small place". The root cause of these difficulties is that there is a lot of "noise" in the entire process of improving R&D performance.

Daniel Kahneman, winner of the 2002 Nobel Prize in Economics, stated in his new book "Noise" that "noise" refers to unnecessary random variability in the decision-making process. Where there is judgment, there is "noise".

There are many judgments in the process of R&D performance improvement, and a lot of noise will naturally be generated. Most of these noises are invisible, you can't see it but will be affected by it.

The following are three common "noises" in the development process.

The first is the "noise" generated by the execution of predictive decisions. In the process of promoting the implementation of management methods such as agile and DevOps, teams often need to transform their organizational structure. This is a risky decision that requires the trust and support of the company's management, and making this predictive decision will produce random variables.

The second is the "noise" caused by target understanding deviation, in which target understanding deviation is more common. For example, the OKR of corporate managers is "increase R&D efficiency by 20%." After this goal is split into different departments, the testing department believes that the test efficiency should be increased by 20%, while the R&D team believes that the R&D efficiency should be increased by 20%. But in fact, what managers want is to achieve end-to-end R&D efficiency improvement in the entire R&D process. This is the noise of target understanding deviation.

The third type is "noise" caused by subjective emotional changes. The improvement of the R&D process usually changes the work process of the team and the work habits of frontline employees. Requiring front-line employees to leave their familiar working methods is an "anti-human" thing, and employees may have some sense of resistance, which will affect the final implementation of the improvement measures.

It can be seen that the improvement of R&D efficiency is a complex activity that is influenced by multiple factors and links. The "Book of Changes" mentions that "easy is easy to know, simple is easy to follow; easy to know is close to each other, and easy to follow is effective". This idea can be applied to the process of performance improvement, turning complexity into simplicity, and adopting simple and gradual progress. The improvement measures make it easy for the team to understand and implement, thereby reducing the noise impact in the entire R&D efficiency process.

Make it simple

Progressive R&D performance improvement

The R&D process is actually a process of value flow. We imagine the R&D process as a road, and the improvement of R&D efficiency is to remove the obstacles that may occur on this road (such as narrowing of the road, traffic accidents, etc.), avoiding more and more obstacles due to a road obstacle, and eventually leading to traffic jams.

We can use "theory of constraints" to "unblock roads" in the research and development process. The theory of constraints means that the beat of the bottleneck node in the actual business determines the entire business process. It is divided into five steps: identifying constraints, exhausting constraints, coordinating constraints, breaking constraints, and then returning to the first step. Carry out cyclical improvements. According to this theory, we can continuously identify bottlenecks, break through bottlenecks, and finally achieve gradual improvements in efficiency.

Theory of Constraints

Take the performance improvement practice of the ONES team as an example.

ONES has established a delivery team to respond to customer needs and improve customer satisfaction. Over time, the team’s effectiveness has become a bottleneck. We first thought of increasing the size of the team, improving the quality of team members, or improving team effectiveness by introducing automated processes. However, each of the above methods requires a large investment of resources, and may even require the team to suspend the business to make improvements, which is not realistic. Therefore, ONES has adopted a gradual improvement method.

We started by analyzing the core dilemma of the delivery team.

First of all, the ONES product matrix is rich, and 8 products have been released. The delivery team must fully understand all the products and their details, otherwise it will cause the team to optimize the demand and the final product quality will not be high.

Second, the customer has an expected delivery time for the demand, and the team spends a lot of time estimating the working hours and giving a more accurate time. However, due to various complex factors in the development process, delivery delays have been caused, and demand has been accumulating. Finally, even small optimizations require a long time to respond, which ultimately leads to a decrease in business satisfaction.

Furthermore, the number of customer needs and the time when the requirements are raised are extremely uncertain, and the estimation of new needs will interrupt the team's current iterative research and development, resulting in reduced efficiency.

At the same time, when faced with large and small online defects, the ONES delivery team fully responded, and handling the defects would also interrupt the R&D work.

In order to solve the above-mentioned problems, ONES conducted an efficiency analysis on the delivery team:

• 15-20 new demands are added every month, and a large number of demands are waiting for research and development
• The scale is estimated to need to complete 15-20 per month, and the estimated time ranges from half a day to one day
• No matter what kind of defect, it needs immediate response, often interrupting R&D
• R&D completed requirements still need R&D investment to repair test defects after starting the test

Before improvement: Discrete demand cycle time

Before improvement: the demand for research and development is higher than the demand for release

After comprehensive analysis, we finally determined that the constraint of the delivery team’s efficiency improvement is the R&D link, and formulated a solution for this:

• Set up an SLA (Service Level Agreement) response level that optimizes demand and replace precise estimates with rough estimates, which greatly reduces the time the team spends on demand estimates and devotes more resources to activities that directly generate customer value.
• Set WIP (work-in-progress) restrictions in R&D and R&D completion together to reduce the need for "near completion", thereby accelerating the flow of value.
• Set up a defect SLA, and a serious defect can temporarily break the WIP limit. Through Kanban, we carry out priority management on defects, and visually show the processing procedures and conditions of defects, so that upstream and downstream teams can understand the R&D capabilities of the R&D team more clearly, and coordinate with the R&D team to adjust the priority of their own needs.
• Hold monthly demand planning meetings based on Kanban, and negotiate with upstream and downstream the demand priority in the Backlog.

We have continued to observe the improvement measures. After two or three months of implementation, the team’s new demand cycle time was concentrated in 5 days, 10 days, and 20 days, and the predictability of delivery time was enhanced. At the same time, the number of pending R&D needs continued to decline and stabilized at a healthy level, and the number of parallel tasks was maintained at 2-3, and the bottleneck of the R&D process was cleared to a certain extent.

After improvement: the demand cycle is predictable

After improvement: the number of pending R&D requirements

In order to complete the dredging to a greater extent, the next step is to break the constraints. To this end, we also did three things:

• Investigate and analyze customer service problems in defects, and set up an independent department to deal with them, so that the R&D team can be more focused
• Separate roadmap requirements and negotiate response strategies with upstream departments and product departments
• Expand the size of the team and increase the limits of WIP

In the practice of gradual R&D and improvement, team effectiveness and business satisfaction have been significantly improved. From the progressive performance improvement experience of the ONES team, we have summed up two core concepts:

First of all, maximizing the use of non-bottleneck resources will only cause accumulation and waiting. Performance improvement needs to focus on clearing the bottleneck, so that the improvement becomes simple and executable.

Secondly, the gradual improvement of R&D efficiency can give positive feedback to the team in a short period of time, thereby enhancing the team's self-driving ability. At the same time, by improving the predictability of demand, it can effectively improve business satisfaction and establish a transparent and trustworthy team atmosphere.