Brief Summary: This video emphasizes that simply automating repetitive tasks has clear limits for organizational productivity, and introduces process mining--a data-driven analysis technique--as a practical and concrete way to identify core bottlenecks and innovate how work gets done. It clearly lays out a strategy for developers to go beyond mere implementation and proactively think about and practice how an organization should work to grow. This is a valuable guide for anyone interested in advancing work methods and data-driven improvement processes after automation.
1. Topic Introduction and the Speaker's Motivation
The speaker, developer Kim Byeong-muk, first acknowledges that the term process mining may be unfamiliar, but explains that it is actually "not a story about technology, but a story about how we work."
"This is less about technology and more about examining how we work."
He shared that he prepared this talk based on his experience hitting the limits of repetitive automation and operational efficiency. His interest in this topic originated from a part-time job he had before becoming a developer. When customers suddenly flooded in, the unprepared operating system led to frequent mistakes, which generated complaints and damaged the store's reputation--teaching him that "crises can actually arise during the growth process."
"When customers increase, if the way we work doesn't evolve accordingly, crises can actually emerge during the growth process."
As a developer, he continued to be deeply interested in advancing how work gets done.
2. The Need for Data-Driven Work Improvement
He emphasizes that simply automating inconvenient tasks or adopting cutting-edge technologies like AI is not enough for fundamental productivity improvement. The core message is that we must first use data to concretely understand how our team works and where time is being wasted.
"Before implementing anything, it's really important to look closely at how our team currently works, and to examine that not through guesswork but accurately through data."
In practice, even with repetitive engineering tasks like maintenance and error handling, teams often don't know "how many issues we process per month, or how much time each step actually takes." If one step could be shortened with a simple Slack notification, the entire team could save over 100 hours per month.
3. What Is Process Mining?
3.1. Concept and Background
Process mining is a methodology that collects and analyzes actual workflow data left in various systems (e.g., GitHub, Sentry, Slack, etc.) to visualize business processes and identify bottlenecks.
"Process mining is a theory that analyzes workflow based on data. It can be divided into three stages: collecting work data, processing it into an analysis-ready format, and finally deriving a 'process model'--a map of the workflow."
3.2. Scope of Application
This methodology can be applied not only to IT development tasks but to all repetitive organizational work, including operations, marketing, customer service, sales, and order processing.
4. The Three-Step Process Mining Execution Method
4.1. First Step: Data Collection
Assuming that all work happens on some system, you collect the data traces (logs) left by work activities.
- Example: Order-delivery work -> commerce system, admin, shipping company data, etc.
- Workflow data can be pulled from various systems via APIs (GitHub, Sentry, Slack, CRM, call center, marketing tools, etc.)
"Every trace of work can be accumulated as data... External tools can also be collected via APIs."
4.2. Second Step: Refining into Event Log Format
Data is organized into 'event log format'. The key columns are:
- Case ID: A single work item or processing case (e.g., issue number, order number, etc.)
- Activity: The specific steps that make up the work (e.g., development, code review, deployment, assignee allocation, etc.)
- Timestamp: The time each step occurred
"Recording these three pieces of information--Case ID, Activity, and Timestamp--together is what allows you to accurately analyze the time spent at each step, correlations, and bottleneck points."
Additionally, adding a resource (assignee) column makes it possible to identify staffing distribution issues and work concentration patterns.
4.3. Scoping -- Which Steps to Track?
Rather than blindly collecting all data, you need to define the scope of steps to track based on the problem definition.
"You only need to track the minimum number of steps necessary for the problem you're interested in."
For example, in shipping operations, rather than collecting every granular step, you might only record 'shipped' and 'delivery completed.' If bottlenecks or improvement points are found, you can then drill down into just that section for more detail.
- -> Quickly identify bottlenecks with minimal effort, and
- -> Expand and analyze only the relevant area if needed--this flexible approach is emphasized as important.
4.4. Entity Resolution
How do you link data from multiple systems into 'a single workflow'? This is the entity resolution problem.
"You need to properly link data accumulated across multiple systems using 'Case IDs' to accurately see total processing times, bottleneck segments, and more."
In practice, this can be achieved using common identifiers (e.g., order numbers, customer phone numbers), each tool's built-in features, or data integration approaches.
5. Discovering Insights Through Process Mining
5.1. Deriving Process Models and Examples
With well-structured event log data, commercial tools or open-source libraries (like PM4Py) can generate visualization models with just a few lines of code.
- How many flows branch at each step and in what ways
- How long each step and the overall process takes
In one example, after error detection, the flow should have moved to assignee allocation, but 17% of cases were simply closed (unnecessary error alerts).
"Let's reduce the frequency of this flow from 17% to 0%. Every detected error should be one that truly requires handling."
Furthermore, for order processes, breaking steps into finer detail reveals exactly where and why 'order cancellations' frequently occur, enabling targeted root cause identification and selective improvement.
5.2. Identifying Specific Bottlenecks and Optimization Strategies
In practice, the code review step alone accounted for 43% of total processing time--clearly revealing 'segments where wait time can be improved.'
"Steps that take long due to waiting time, like code review, can have their overall speed dramatically improved just by introducing Slack notifications or auto-merge."
This process applies to all repetitive tasks across operations, orders, customer support, and more. For example, in customer service, if the segment where "contact can't be reached and the case sits idle" turns out to significantly affect overall delays, it can be greatly improved by sending scheduled texts or notification messages.
"Coming up with these ideas vaguely is difficult, but the moment you pull out the data and know exactly which step is the problem, you'll find that good ideas naturally flow."
6. Summary and Expanding the Developer's Role
Finally, Kim Byeong-muk reemphasizes that 'advancing how we work is just as important as advancing the product' for organizational growth.
- Process Mining:
- Build event log data
- Analyze workflow based on data
- Discover bottlenecks and generate improvement ideas
- Repeat: verify effectiveness after improvements
"Going beyond developing given requirements as a developer, the process of deeply examining workflows through data and finding unexpected problems was truly fascinating and rewarding."
He also advises that in an era where technologies like AI are accelerating, 'the ability to define good problems' will become increasingly important, rather than raw development skills alone.
"With the advancement of AI and LLMs, if you can define specific requirements well, AI can handle the development itself rapidly. In times like these, it's important for developers to take interest in 'good problem definition' and expand their roles."
The speaker closes with warm wishes that this session might serve as "a practical catalyst for expanding one's role as a developer."
"I hope this session has been at least somewhat helpful in thinking about how to expand your role as a developer... That concludes the session. Thank you."
Conclusion
This video provides concrete guidance for anyone thinking about "what comes after automating repetitive tasks" on how to boldly innovate the way they work based on data. Data-driven analysis and problem definition through process mining will be a key strategy for driving productivity innovation not just for developers, but for entire organizations. It was a presentation filled with practical insights for developers and organizations finding their own better path forward.