Revolutionizing Decision Making: How Hackle AI Pioneered the Leap from Data-Driven to AI-Driven Insights
Automation of the decision-making process: data-based decision making to AI-based decision making
Hi, Yeonju (Wonder) and my OBH, which currently exist only as code in the archive folder
Accelerates customer decision-making by improving the way customers interact with data using OBH
Manual → Semi-automatic → Automatic
$PLTR shows the Levels of Enterprise AI
- Business Intelligence and Search (Chat)
- Decision Guidance (Alerts)
- Tribal Knowledge (Learning)
- Automated Action (AI agents)
Q. Where would the Hackle AI I was trying to implement in 2023 be among the four stages defined by PLTR?
If I put it generously, I think it’s level 2. KPI indicator setting and monitoring, indicator anomaly detection and cause analysis, all automatedIt was an AI that could provide guidance for business decision making. The ultimate goal of our AI was to provide an optimized user experience for individual users who use online platform services.
- People who want to make data-based decisions
- (Short-term) People who want to know the current situation by continuously monitoring (tracking) the KPI indicators of the product for which they are responsible.
- (Long-term) People who want to analyze KPI indicators and find ways to improve them
- When [Situation] I wanted to solve the following problem situations that slow down organizational decision-making and reduce productivity.
– To know the current status of KPI indicators created using a data analyst or data analysis tool, you must periodically visit the dashboard, log in, and check the trend.
– Even if you check, it is difficult to determine whether the indicator situation is good or bad.
– On busy days, there are days when I can’t check.
– The indicator is flat, but if you look at the detailed attributes (categories) & user groups within it, it appears to be flat with a mixture of upward and downward trends, so you cannot tell if a problem has occurred in a specific detailed attribute or user group.
– When the indicator value drops, you have to search for the related indicators one by one to find the cause, or group by with numerous attributes to find the cause, but you do not know what the related indicators are, and when you group by with numerous attributes, there are too many attribute values. Therefore, it is difficult to find which attribute and attribute value is the cause.
– When trying to improve indicators, it takes a lot of time to find out where and how to improve, which will greatly contribute to improving the indicators. I’m at a loss because I don’t even know how to approach it. - I want to [Motivation]
– I want to receive notifications about KPI indicators through work tools I use every day, such as Slack/email.
– When I check the indicator, I want to know immediately whether the current status is good or bad, not just numbers or line charts.
– When the current situation is bad, I want to quickly know the cause and deal with it if I can control it with the product.
– When you want to improve your current indicators, you want to be told where and how to improve to have the greatest effect. - So I can [Expected Outcome]
– You can receive regular notifications via email or Slack of summarized information that can help you understand the status of key indicators.
– When the status of an indicator is good or bad and you want to explore factors affecting it, this can also be checked directly in the notification in a summarized form.
– You can receive recommendations for ideas to improve indicators (where and how to improve) through notifications. - (Optional) What are the symptoms of the Problem?
[1세대 OBH의 문제]Although random indicators are monitored and cause analysis is performed in the form of messages through a dialog window, it was difficult to recognize what indicator was being monitored, what the status of the indicator was, and what the cause was just by looking at the message.
Why can’t I check the KPI dashboard and see whether the indicators are good or bad?
1. Summary form — trend unknown, based on 100%, if it is above 100%, the indicator is rising, and if it is below, the indicator is falling.
2. Charts & Tables – Requires human interpretation
1 +2 combined state (amplitude)
1 + 2 + prediction (amplitude)
Why is it difficult to find the cause when an indicator falls or rises?
- If you don’t know what the related indicators and attributes are and search for them, you won’t be able to find them.
- Searching requires breaking down numerous indicators and the properties of individual indicators and comparing them one by one, which is very cumbersome.
3. If the related attributes and attribute values are diverse, there will be too many combinations, making it difficult to find the cause.
When trying to improve indicator values, why is it difficult to know where and how to improve the indicators that will greatly contribute to improving the indicator?
(Example number of members)
(1) Check the membership conversion rate → (2) Check the membership registration funnel → (3) Recognize the funnel stage from which most people churn → (4–1) Check the user path where they deviate from that stage
(4–2) Analysis of the difference between users who drop out and users who convert at that stage
(4–2–1) User attribute analysis
(4–2–2) User behavior analysis — Compare behavior right up to conversion or abandonment…
(4–2–3) Analysis of time required for conversion
(Optional) Why is it critical to solve now?
- (as-is) self-service data analysis platform — Anyone can derive insights based on data by using the functions provided by the tool (X), extract data (O). Data can be analyzed (△)
- Data analyst: The functions supported by the data analysis platform are limited, and in order to perform the necessary detailed analysis, it must be exported to Excel and analyzed further. There is nothing better than collecting and analyzing data yourself. → Because they can directly process data, they prefer visualization tools with a high degree of freedom such as Tableau and PowerBI.
- Non-data analysts: Learning is required to utilize data analytics platforms. Simple analysis is possible, but what if you want to analyze data from various dimensions at a glance? It also needs to be exported and analyzed further. In that case, all you need to do is extract the data you want. A level of help in the process of finding insight is needed, not just for simple inquiry.
- (to-be) Hackle requires significant resources to develop and add an analysis menu as advanced as its competitors. It is impossible to catch up by approaching it on a menu (function) basis. In this situation, if you want to gain a competitive advantage over them, customers will need to use a data analysis platform. Automate workflowMust do this.