5 characteristics of KPIs supporting artificial intelligence and machine learning
Before speaking about the KPIs, let me explain how we use AI and machine learning so that we know how to define the KPIs with the specified characteristics to support AI as opportunities for data gathering.
In business strategy, we use artificial intelligence to help us in decision making or getting results for the main following areas:
Prediction: AI says that tell me your story of your past data then I tell you the future. Here data scientists make a model to predict the future outcomes. Example: If a retailer has the past data for a good number of years in sales and also the inflation rate and other economic factors then with an acceptable accuracy rate we can predict the sales for the first quarter of next year.
Pattern recognition: In this case, we try to create patterns to understand the outcome better. For example: online shopping websites suggest different items to the users after some time by creating a pattern out of the preferences of specific users. In marketing it is called individualization. The user feels better, more convenient and excited once s/he visits the website. Briefly we can divide the pattern recognition process into 6 steps: sensing, segmentation, feature extraction, classification, post processing and finally decision.
Anomaly detection: Finding anomalies to remove them or to figure out the reasons is an important job of AI. For instance: in a complicated laptop production line (considering all parts are coming from the same sources), by understanding the process measurements, we can create a standard for quality control and quality levels. In this example, we would like to understand the processes so that we can predict the quality and in case we have many defects to stop the line and to find the root cause not to waste time, electricity, manpower, parts, etc.
Clustering for decision making: Data by itself has no value unless we can extract insights from it. To make decisions out of past activities sometimes we need to study different items clustered in certain categories. Data scientists can make this possible. Let me tell you a story: Sales people usually find CFOs as a demotivator! They reach their targets and still the CFO turns them down. Why? Because sales simply sees the revenue and CFO sees revenue and the profit. It is true that a single product in a specific market can sell a lot but the question is how the numbers are globally! Sales people in that specific region are happy but indeed the CFO is unhappy. Clustering data shows the facts more clearly. In the above story, if we have thousands or millions of products sold in 100 different cities then we need to know how many cities are profitable, how many reached breakeven point or they are in loss. How fast or slow were the sales processes, etc. This is all possible by clustering.
Natural Language Processing: Natural language is the way that we speak or we comment and write in social media. Humans understand it very well but machines need to turn it to code to be able to process it. AI and ML provide us the ability to scan all comments about a service or product on social media and then cluster them and get acceptable sentiments. The result may not be 100% accurate but it is reliable surely. We can understand what problems we have in which regions with which product without spending huge amounts of money to conduct surveys. (As I explained, the result may not be 100% accurate but it is reliable surely)
Now that we know the role of the AI, let's get back to the KPIs as the biggest supporter for AI and ML. KPIs used for AI business processes are useful when we define them as below:
Object oriented: Completely based on your organization objectives aligned with your mission and vision. It would be a huge mistake if we copy the industry KPIs without considering creative KPIs to indicate problems faster and more efficiently. Simply: KPIs are like lighthouses showing you the direction.
Online, 24/7: Measuring system and KPIs are not one-shot deals! No period of time for gathering data should be ignored or the results would be all fake. Compromising data gathering in a similar process but different times could create trouble. For example, the fact that production processes in a line are always the same doesn't let us ignore the data from night shift. The process on paper is the same but in reality, temperature, light or human factors differ surely.
Democratized: Every member of a team, organization and process should know where we are standing and where we are heading. This is not about technology but leadership. Like other projects, the success is achieved and gets repeated once everyone is aware of things around. In the story of the CFO and sales people explained above, I should have explained that many sales people don't know what the breakeven point is. That is why they don't understand why the CFO or top management are not happy. This happens when we wouldn't like to share profit or loss with our colleagues. It's sad but it happens unfortunately!
Automated: Employees cannot once do the job and then once again measure the job! This is exhausting. Systems should be so much digitalized and automated that at the same time of doing the job, the measurements happen too. Let's not forget that we pay people to do the job. We don't pay them to do the job and then to measure it for us only because we don't have an automated system! This is one of the biggest mistakes in leadership. We make people do the job and then we force them to spend even more time preparing reports for us! From the other side, AI needs data. The more relevant and accurate data the better. It is not possible to gather the required data without automation. If it is possible then there might be a huge risk of miscalculation and eventually misunderstandings.
Embracing AI: In the first part of the article, I explained what AI can do for us in business. Also, it is obvious that KPIs are the basics for giving direction to the data gathering process. Then we should be careful how to define the KPIs to achieve at least one of the objectives of AI implementation. (Prediction, Pattern Recognition, Anomaly Detection, Clustering or NLP). It is smarter to know what we would like to do with the KPIs. It has been repeated many times that KPIs are not created to control and punish people. They are used to improve processes. Therefore, once we define KPIs we should know WHY we define it and how we use the results.